Cash Transfers, Agriculture and Child Nutritional Status ... · 3/27/2014 · Cash Transfers,...
Transcript of Cash Transfers, Agriculture and Child Nutritional Status ... · 3/27/2014 · Cash Transfers,...
13
DRAFT NOT FOR CITATION
Cash Transfers Agriculture and Child Nutritional Status
Evidence from Malawi
Alessandro Romeo Solomon Asfaw Benjamin Davis Agricultural Development Economics Division
Food and Agriculture Organization Viale delle Terme di Caracalla
Rome Italy 00153
Paul Winters Department of Economics
American University Washington DC 20016 USA
Draft May 6 2013
Abstract
In the last decade cash transfer programmes have become popular components of poverty reduction strategies in Sub Saharan African countries Along with well established benefits such as a reduction in food insecurity and increased school enrolment recent research has found these programmes are also effective in promoting on-farm activities and investments in agricultural assets In addition some cash transfer interventions have proven effective in improving nutritional status of children living in beneficiary households In this paper we use data from the Mchinji Social Cash Transfer pilot programme to assess if the programme had an impact on child nutritional status and to what extent this impact can be linked to increases in agricultural production by beneficiary households At the household level the analysis shows a substantial impact on household food and non-food expenditure as well as a shift in the consumption preferences towards better nutrients At the individual level we find children of age 0-5 residing in beneficiary households being on average taller compared to the control group which translates into a significant reduction in the stunting rate among children Further we find that the programme positively affected food consumption out of own production and that children living in families experiencing a shift toward home production of foods such as meat and fish benefitted more in terms of nutritional outcomes
JEL Classification I15 I38 O12 O22 Q12
Key words cash transfers impact evaluation child nutrition agriculture Malawi
1
13
DRAFT NOT FOR CITATION
Introduction
Mitigating poverty and improving standard of living of future generations is at the core of
social protection policies in Least Developed Countries (LDCs) In fact the entry point of
such type of interventions is human capital relying on the assumption that poverty is
transmitted across generations mostly due to the inability of poor parents to invest in their
childrenrsquo human capital (Maluccio 2010) Cash transfer programmes have been widely used in
a growing number of countries to secure the poorest from adverse consequences of poverty
and market failures while promoting human capital enhancing behaviour
The positive effect of CT interventions over households receiving the program usually the
poorest living in rural areas spans a range of human development indicators including the
increase in household consumption expenditure higher levels of child school enrolment
nutritional status and access to health services (See reviews in Fiszbein and Schady 2009 and
Handa and Davis 2006)
A smaller number of studies looked at CT interventions as an engine of growth to help the
ldquohardcorerdquo poor a segment of the population which is usually difficult to reach with anti-
poverty programmes (Alatas et al 2012 and Banerjee et al 2011) In fact since the majority
of the programmes rolled-out in rural areas cash transfers intervention might have the
potential to promote income gains across smallholder farmers whose livelihood rely upon
agriculture as well as livestock activities From a theoretical perspective since consumption
and production choices of agriculture smallholder families are non-separable due to market
failures as liquidity and credit constraints CT programmes should positively influence
consumption choices while they could (positively or negatively) affect production choices
More recently a new stream of studies provides a mix of evidence on the same topic
Maluccio (2010) found that the lionrsquos share of the Nicaraguan Red De Proteccion Social was
mostly related to improvements in child human capital while long term increases in
2
13
DRAFT NOT FOR CITATION
consumption as a result of other forms of increased investment was limited On the contrary
Gertler et al (2012) found that the Mexican ProgresaOportunidades safety net programme
increase agricultural income of poor Mexican smallholders by 10 per cent after 18 months the
programme was implemented In addition they found that the intervention significantly
increased the likelihood of initiating an agricultural business as well as expanding pre-existing
activities with respect to the implementation programme rolled-out In the African context
Covarrubias et al (2012) Boone et al (forthcoming) found positive and significant impacts
on own production of foods and investments in agricultural activities under the Social Cash
Transfer intervention unconditional cash transfer pilot programme conducted in Malawi
between 2007-2008 Asfaw et al (2012) found the CT-OVC in Kenya had a positive impact
on agricultural assets ownership while in Ethiopia Hodinott et al (2012) reported that
significant increase in the use of fertilizer along with higher level of agriculture investments
only occurred for those households receiving joint transfer from different social protection
intervention
By and large this evidence suggests that CT programmes has the potential to increase
agricultural production and thus have positive effects on the livelihood of the poorest which
were not initially foreseen to occur following the CT implementation Moreover if
investments in agricultural activities as well as food own production raise due to the CT
programmes such type of interventions could improve the livelihood of targeted families not
only by loosening household expenditure constraints ndash households can purchase more food -
but they can also consume more and better food as they shift toward on-farm activities and
therefore moving toward a more sustainable path of development revolving around
agricultural growth
Since part of the CTs has been found effective in enhancing activities related to home
production of food a first question is to investigate on how those enhancements in food
3
13
DRAFT NOT FOR CITATION
production interact with the nutritional status of household members Also given the nature of
the policy interventions ndash CT programmes aiming at mitigating poverty in the short run while
improving the well being of the new generationsndash it is important to conduct the analysis over
the health status (eg height) of young individuals residing in targeted families Given these
premises in this paper we seek to answer the following question do children residing in
families that shift towards on-farm activities due to the transfer intervention also experience
larger benefit in terms of their nutritional status compared to counterfactual children where the
absence of the transfer did not generate a significant shift toward inndashfarm activities If we find
that poor households tend to produce a larger variety of nutrients (particularly protein based
food) out of own production and thus they shift toward on-farm activities we would also
expect household members children age 0-5 in our study to be better nourished and display
some significant improvement in health outcomes indicators as resilience to disease or more
interestingly height of treated individuals compared to the counterfactual Although
households could be net-sellers of food commodities this is unlikely to be the case in our
samples as the families targeted to the programme were selected across the poorest 10 per cent
of the population facing labour constraints and mainly involved with subsistence farming
While mitigating child malnutrition cannot be exclusively achieved through a raise in income
(Alderman et al 2003) the empirical evidence from Latin American programmes provides a
mix of evidence Mostly children of 0-36 months significantly benefited from CCT
interventions and that the impacts were larger for those programmes in which the transfer
value consisted of approximately 17-25 per cent or more of the pre-treatment household
income and older household members mostly mothers attended nutritional counselling
(Fiszbein et al 2009 Manley et al 2013 Leroy et al 2009 Gertler et al 2004 Maluccio et al
2005 among others) For example Maluccio et al (2005) found that children in the
Nicaraguan Red De Proteccion Social were 017 standard deviation (SD) taller compared the
4
13
DRAFT NOT FOR CITATION
counterfactual while Attanasio et al (2005) found that the Colombian Familias en accioacuten
significantly improved the height-for-age scores by 016 SD In Africa the empirical evidence
chiefly revolves around the Old Age Pension (OAP) programme and the Child Support Grants
(CSG) both implemented in South Africa Girls of age 0-5 living in households taking up the
OAP intervention were 223 cm significantly taller after two years the programme was rolled
out (Duflo 2003) However little empirical evidence exists on the link between agricultural
production and health status To the best of our knowledge one of the few studies looking at
the ability of agricultural production to fulfil the nutritional gap over the most vulnerable
families living in remote and rural areas is proposed by Muller (2008) Based on a sample of
Rwandan farmers the Author found that several food groups have a diverse effect on adult
health status However no studies try to directly disentangle such type of nexus under CT
programme framework
In an attempt to explore this relation our study uses data from the impact evaluation of the
Mchinji cash transfer pilot programme in Malawi conducted between 2007-2008 By
combining double difference (DD) method with Inverse Probability Weighting (IPW) to
restore pre-treatment balancing conditions we test the hypothesis that (i) households
participating in the programme significantly increased the home production of food such as
dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally test to test if and to what extent consumption out of home production of foods can
influence child outcomes By carrying out this type of analysis we conclude that households
boosting their home production (and consumption) of protein-rich foods as meat and fish due
to the transfer intervention also experience larger benefit in terms of child nutritional status
compared to those households which did not shift toward on-farm activities
5
13
DRAFT NOT FOR CITATION
The remainder of the paper is structured as follows section 2 gives a picture of malnutrition
in Malawi and describes the Mchinji SCT pilot programme and the research design section 3
describes the theoretical economic framework underlying the analysis section describes 4 the
empirical specification and the econometric techniques used to carry out the analysis section
5 discusses household and child level attrition in the sample section 6 presents the household
and child level characteristics section 7 provides the results and section 8 concludes
2 Malnutrition in Malawi and the Social Cash Transfer (SCT) programme in Malawi 21 Malnutrition in Malawi Chronic malnutrition is usually measured using standard linear growth index called the HAZ
index The HAZ is calculated by comparing the height for age of a child with a reference
population of well-nourished children We only focus on this nutrition outcome since it is
considered the best indicator of the long-term cumulative effects of under nutrition in
childhood development and we therefore define a child ldquostuntedrdquo or chronically malnourished
if her if her score is below ldquo-2rdquo The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a health child For a child of 36 months this translates into approximately 36 cm
difference1
Over the last decade Malawi experienced a considerable decline in chronic malnutrition
respectively dropping by 7pp (Table 1 and Figure 1) Yet approximately 1 child in 2 is still
stunted Moreover children living in the poorest families and in rural areas show levels of
chronic malnutrition which are higher compared to the national average (Table 1) To mitigate
poverty and seeking to guarantee a better and healthier future to young generations the
government of Malawi embarked in the Social Cash Transfer (SCT) programme
lt FIGURE 1 ABOUT HERE gt
1 2006 WHO manual
6
13
DRAFT NOT FOR CITATION
lt TABLE 1 ABOUT HERE gt
22 Mchinji pilot Social Cash Transfer (SCT) Programme of Malawi
The Social Cash Transfer (SCT) programme targets rural ultra-poor and labour-constrained
households to unconditional cash transfers The estimated transfer size is 78 USD per
household ranging between 4 USD for households with only one member to 13 USD for
families with four or more eligible members Embedded in the transfer the programme
provides a school bonus fee ranging between 13 USD per primary school age child and 26
USD per secondary school age child Participants are selected using a combination of proxy
means implemented at community level The programme covers 116000 individual
beneficiaries (28000 households) and with its current expansion planning to reach 300000
households by 2015
23 Malawi research design The impact evaluation of the Mchinji pilot SCT programme is a randomized longitudinal
design with a baseline household survey conducted in 2007 and two follow-up surveys
conducted respectively after 6 months and 12 months For the evaluation of the Malawi SCT
a one-year pilot of the programme was designed and implemented in the Mchinji District in
central Malawi Specifically four control and four treatment Village Development Groups
(VDCs) forming part of the original programme rollout were randomly selected to be part of
the evaluation These eight VDCs correspond to 23 villages Within each village Community
Social Protection Committees (CSPCs) identified eligible households according to the
national eligibility criteria and then ranked them in order to select the bottom 10 per cent for
inclusion to the programme
3 Theoretical framework As we mentioned in the introduction the aim of this paper is to test whether families enrolled
in the STC programme increased (i) significantly increased the home production of food such
as dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
7
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Introduction
Mitigating poverty and improving standard of living of future generations is at the core of
social protection policies in Least Developed Countries (LDCs) In fact the entry point of
such type of interventions is human capital relying on the assumption that poverty is
transmitted across generations mostly due to the inability of poor parents to invest in their
childrenrsquo human capital (Maluccio 2010) Cash transfer programmes have been widely used in
a growing number of countries to secure the poorest from adverse consequences of poverty
and market failures while promoting human capital enhancing behaviour
The positive effect of CT interventions over households receiving the program usually the
poorest living in rural areas spans a range of human development indicators including the
increase in household consumption expenditure higher levels of child school enrolment
nutritional status and access to health services (See reviews in Fiszbein and Schady 2009 and
Handa and Davis 2006)
A smaller number of studies looked at CT interventions as an engine of growth to help the
ldquohardcorerdquo poor a segment of the population which is usually difficult to reach with anti-
poverty programmes (Alatas et al 2012 and Banerjee et al 2011) In fact since the majority
of the programmes rolled-out in rural areas cash transfers intervention might have the
potential to promote income gains across smallholder farmers whose livelihood rely upon
agriculture as well as livestock activities From a theoretical perspective since consumption
and production choices of agriculture smallholder families are non-separable due to market
failures as liquidity and credit constraints CT programmes should positively influence
consumption choices while they could (positively or negatively) affect production choices
More recently a new stream of studies provides a mix of evidence on the same topic
Maluccio (2010) found that the lionrsquos share of the Nicaraguan Red De Proteccion Social was
mostly related to improvements in child human capital while long term increases in
2
13
DRAFT NOT FOR CITATION
consumption as a result of other forms of increased investment was limited On the contrary
Gertler et al (2012) found that the Mexican ProgresaOportunidades safety net programme
increase agricultural income of poor Mexican smallholders by 10 per cent after 18 months the
programme was implemented In addition they found that the intervention significantly
increased the likelihood of initiating an agricultural business as well as expanding pre-existing
activities with respect to the implementation programme rolled-out In the African context
Covarrubias et al (2012) Boone et al (forthcoming) found positive and significant impacts
on own production of foods and investments in agricultural activities under the Social Cash
Transfer intervention unconditional cash transfer pilot programme conducted in Malawi
between 2007-2008 Asfaw et al (2012) found the CT-OVC in Kenya had a positive impact
on agricultural assets ownership while in Ethiopia Hodinott et al (2012) reported that
significant increase in the use of fertilizer along with higher level of agriculture investments
only occurred for those households receiving joint transfer from different social protection
intervention
By and large this evidence suggests that CT programmes has the potential to increase
agricultural production and thus have positive effects on the livelihood of the poorest which
were not initially foreseen to occur following the CT implementation Moreover if
investments in agricultural activities as well as food own production raise due to the CT
programmes such type of interventions could improve the livelihood of targeted families not
only by loosening household expenditure constraints ndash households can purchase more food -
but they can also consume more and better food as they shift toward on-farm activities and
therefore moving toward a more sustainable path of development revolving around
agricultural growth
Since part of the CTs has been found effective in enhancing activities related to home
production of food a first question is to investigate on how those enhancements in food
3
13
DRAFT NOT FOR CITATION
production interact with the nutritional status of household members Also given the nature of
the policy interventions ndash CT programmes aiming at mitigating poverty in the short run while
improving the well being of the new generationsndash it is important to conduct the analysis over
the health status (eg height) of young individuals residing in targeted families Given these
premises in this paper we seek to answer the following question do children residing in
families that shift towards on-farm activities due to the transfer intervention also experience
larger benefit in terms of their nutritional status compared to counterfactual children where the
absence of the transfer did not generate a significant shift toward inndashfarm activities If we find
that poor households tend to produce a larger variety of nutrients (particularly protein based
food) out of own production and thus they shift toward on-farm activities we would also
expect household members children age 0-5 in our study to be better nourished and display
some significant improvement in health outcomes indicators as resilience to disease or more
interestingly height of treated individuals compared to the counterfactual Although
households could be net-sellers of food commodities this is unlikely to be the case in our
samples as the families targeted to the programme were selected across the poorest 10 per cent
of the population facing labour constraints and mainly involved with subsistence farming
While mitigating child malnutrition cannot be exclusively achieved through a raise in income
(Alderman et al 2003) the empirical evidence from Latin American programmes provides a
mix of evidence Mostly children of 0-36 months significantly benefited from CCT
interventions and that the impacts were larger for those programmes in which the transfer
value consisted of approximately 17-25 per cent or more of the pre-treatment household
income and older household members mostly mothers attended nutritional counselling
(Fiszbein et al 2009 Manley et al 2013 Leroy et al 2009 Gertler et al 2004 Maluccio et al
2005 among others) For example Maluccio et al (2005) found that children in the
Nicaraguan Red De Proteccion Social were 017 standard deviation (SD) taller compared the
4
13
DRAFT NOT FOR CITATION
counterfactual while Attanasio et al (2005) found that the Colombian Familias en accioacuten
significantly improved the height-for-age scores by 016 SD In Africa the empirical evidence
chiefly revolves around the Old Age Pension (OAP) programme and the Child Support Grants
(CSG) both implemented in South Africa Girls of age 0-5 living in households taking up the
OAP intervention were 223 cm significantly taller after two years the programme was rolled
out (Duflo 2003) However little empirical evidence exists on the link between agricultural
production and health status To the best of our knowledge one of the few studies looking at
the ability of agricultural production to fulfil the nutritional gap over the most vulnerable
families living in remote and rural areas is proposed by Muller (2008) Based on a sample of
Rwandan farmers the Author found that several food groups have a diverse effect on adult
health status However no studies try to directly disentangle such type of nexus under CT
programme framework
In an attempt to explore this relation our study uses data from the impact evaluation of the
Mchinji cash transfer pilot programme in Malawi conducted between 2007-2008 By
combining double difference (DD) method with Inverse Probability Weighting (IPW) to
restore pre-treatment balancing conditions we test the hypothesis that (i) households
participating in the programme significantly increased the home production of food such as
dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally test to test if and to what extent consumption out of home production of foods can
influence child outcomes By carrying out this type of analysis we conclude that households
boosting their home production (and consumption) of protein-rich foods as meat and fish due
to the transfer intervention also experience larger benefit in terms of child nutritional status
compared to those households which did not shift toward on-farm activities
5
13
DRAFT NOT FOR CITATION
The remainder of the paper is structured as follows section 2 gives a picture of malnutrition
in Malawi and describes the Mchinji SCT pilot programme and the research design section 3
describes the theoretical economic framework underlying the analysis section describes 4 the
empirical specification and the econometric techniques used to carry out the analysis section
5 discusses household and child level attrition in the sample section 6 presents the household
and child level characteristics section 7 provides the results and section 8 concludes
2 Malnutrition in Malawi and the Social Cash Transfer (SCT) programme in Malawi 21 Malnutrition in Malawi Chronic malnutrition is usually measured using standard linear growth index called the HAZ
index The HAZ is calculated by comparing the height for age of a child with a reference
population of well-nourished children We only focus on this nutrition outcome since it is
considered the best indicator of the long-term cumulative effects of under nutrition in
childhood development and we therefore define a child ldquostuntedrdquo or chronically malnourished
if her if her score is below ldquo-2rdquo The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a health child For a child of 36 months this translates into approximately 36 cm
difference1
Over the last decade Malawi experienced a considerable decline in chronic malnutrition
respectively dropping by 7pp (Table 1 and Figure 1) Yet approximately 1 child in 2 is still
stunted Moreover children living in the poorest families and in rural areas show levels of
chronic malnutrition which are higher compared to the national average (Table 1) To mitigate
poverty and seeking to guarantee a better and healthier future to young generations the
government of Malawi embarked in the Social Cash Transfer (SCT) programme
lt FIGURE 1 ABOUT HERE gt
1 2006 WHO manual
6
13
DRAFT NOT FOR CITATION
lt TABLE 1 ABOUT HERE gt
22 Mchinji pilot Social Cash Transfer (SCT) Programme of Malawi
The Social Cash Transfer (SCT) programme targets rural ultra-poor and labour-constrained
households to unconditional cash transfers The estimated transfer size is 78 USD per
household ranging between 4 USD for households with only one member to 13 USD for
families with four or more eligible members Embedded in the transfer the programme
provides a school bonus fee ranging between 13 USD per primary school age child and 26
USD per secondary school age child Participants are selected using a combination of proxy
means implemented at community level The programme covers 116000 individual
beneficiaries (28000 households) and with its current expansion planning to reach 300000
households by 2015
23 Malawi research design The impact evaluation of the Mchinji pilot SCT programme is a randomized longitudinal
design with a baseline household survey conducted in 2007 and two follow-up surveys
conducted respectively after 6 months and 12 months For the evaluation of the Malawi SCT
a one-year pilot of the programme was designed and implemented in the Mchinji District in
central Malawi Specifically four control and four treatment Village Development Groups
(VDCs) forming part of the original programme rollout were randomly selected to be part of
the evaluation These eight VDCs correspond to 23 villages Within each village Community
Social Protection Committees (CSPCs) identified eligible households according to the
national eligibility criteria and then ranked them in order to select the bottom 10 per cent for
inclusion to the programme
3 Theoretical framework As we mentioned in the introduction the aim of this paper is to test whether families enrolled
in the STC programme increased (i) significantly increased the home production of food such
as dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
7
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
consumption as a result of other forms of increased investment was limited On the contrary
Gertler et al (2012) found that the Mexican ProgresaOportunidades safety net programme
increase agricultural income of poor Mexican smallholders by 10 per cent after 18 months the
programme was implemented In addition they found that the intervention significantly
increased the likelihood of initiating an agricultural business as well as expanding pre-existing
activities with respect to the implementation programme rolled-out In the African context
Covarrubias et al (2012) Boone et al (forthcoming) found positive and significant impacts
on own production of foods and investments in agricultural activities under the Social Cash
Transfer intervention unconditional cash transfer pilot programme conducted in Malawi
between 2007-2008 Asfaw et al (2012) found the CT-OVC in Kenya had a positive impact
on agricultural assets ownership while in Ethiopia Hodinott et al (2012) reported that
significant increase in the use of fertilizer along with higher level of agriculture investments
only occurred for those households receiving joint transfer from different social protection
intervention
By and large this evidence suggests that CT programmes has the potential to increase
agricultural production and thus have positive effects on the livelihood of the poorest which
were not initially foreseen to occur following the CT implementation Moreover if
investments in agricultural activities as well as food own production raise due to the CT
programmes such type of interventions could improve the livelihood of targeted families not
only by loosening household expenditure constraints ndash households can purchase more food -
but they can also consume more and better food as they shift toward on-farm activities and
therefore moving toward a more sustainable path of development revolving around
agricultural growth
Since part of the CTs has been found effective in enhancing activities related to home
production of food a first question is to investigate on how those enhancements in food
3
13
DRAFT NOT FOR CITATION
production interact with the nutritional status of household members Also given the nature of
the policy interventions ndash CT programmes aiming at mitigating poverty in the short run while
improving the well being of the new generationsndash it is important to conduct the analysis over
the health status (eg height) of young individuals residing in targeted families Given these
premises in this paper we seek to answer the following question do children residing in
families that shift towards on-farm activities due to the transfer intervention also experience
larger benefit in terms of their nutritional status compared to counterfactual children where the
absence of the transfer did not generate a significant shift toward inndashfarm activities If we find
that poor households tend to produce a larger variety of nutrients (particularly protein based
food) out of own production and thus they shift toward on-farm activities we would also
expect household members children age 0-5 in our study to be better nourished and display
some significant improvement in health outcomes indicators as resilience to disease or more
interestingly height of treated individuals compared to the counterfactual Although
households could be net-sellers of food commodities this is unlikely to be the case in our
samples as the families targeted to the programme were selected across the poorest 10 per cent
of the population facing labour constraints and mainly involved with subsistence farming
While mitigating child malnutrition cannot be exclusively achieved through a raise in income
(Alderman et al 2003) the empirical evidence from Latin American programmes provides a
mix of evidence Mostly children of 0-36 months significantly benefited from CCT
interventions and that the impacts were larger for those programmes in which the transfer
value consisted of approximately 17-25 per cent or more of the pre-treatment household
income and older household members mostly mothers attended nutritional counselling
(Fiszbein et al 2009 Manley et al 2013 Leroy et al 2009 Gertler et al 2004 Maluccio et al
2005 among others) For example Maluccio et al (2005) found that children in the
Nicaraguan Red De Proteccion Social were 017 standard deviation (SD) taller compared the
4
13
DRAFT NOT FOR CITATION
counterfactual while Attanasio et al (2005) found that the Colombian Familias en accioacuten
significantly improved the height-for-age scores by 016 SD In Africa the empirical evidence
chiefly revolves around the Old Age Pension (OAP) programme and the Child Support Grants
(CSG) both implemented in South Africa Girls of age 0-5 living in households taking up the
OAP intervention were 223 cm significantly taller after two years the programme was rolled
out (Duflo 2003) However little empirical evidence exists on the link between agricultural
production and health status To the best of our knowledge one of the few studies looking at
the ability of agricultural production to fulfil the nutritional gap over the most vulnerable
families living in remote and rural areas is proposed by Muller (2008) Based on a sample of
Rwandan farmers the Author found that several food groups have a diverse effect on adult
health status However no studies try to directly disentangle such type of nexus under CT
programme framework
In an attempt to explore this relation our study uses data from the impact evaluation of the
Mchinji cash transfer pilot programme in Malawi conducted between 2007-2008 By
combining double difference (DD) method with Inverse Probability Weighting (IPW) to
restore pre-treatment balancing conditions we test the hypothesis that (i) households
participating in the programme significantly increased the home production of food such as
dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally test to test if and to what extent consumption out of home production of foods can
influence child outcomes By carrying out this type of analysis we conclude that households
boosting their home production (and consumption) of protein-rich foods as meat and fish due
to the transfer intervention also experience larger benefit in terms of child nutritional status
compared to those households which did not shift toward on-farm activities
5
13
DRAFT NOT FOR CITATION
The remainder of the paper is structured as follows section 2 gives a picture of malnutrition
in Malawi and describes the Mchinji SCT pilot programme and the research design section 3
describes the theoretical economic framework underlying the analysis section describes 4 the
empirical specification and the econometric techniques used to carry out the analysis section
5 discusses household and child level attrition in the sample section 6 presents the household
and child level characteristics section 7 provides the results and section 8 concludes
2 Malnutrition in Malawi and the Social Cash Transfer (SCT) programme in Malawi 21 Malnutrition in Malawi Chronic malnutrition is usually measured using standard linear growth index called the HAZ
index The HAZ is calculated by comparing the height for age of a child with a reference
population of well-nourished children We only focus on this nutrition outcome since it is
considered the best indicator of the long-term cumulative effects of under nutrition in
childhood development and we therefore define a child ldquostuntedrdquo or chronically malnourished
if her if her score is below ldquo-2rdquo The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a health child For a child of 36 months this translates into approximately 36 cm
difference1
Over the last decade Malawi experienced a considerable decline in chronic malnutrition
respectively dropping by 7pp (Table 1 and Figure 1) Yet approximately 1 child in 2 is still
stunted Moreover children living in the poorest families and in rural areas show levels of
chronic malnutrition which are higher compared to the national average (Table 1) To mitigate
poverty and seeking to guarantee a better and healthier future to young generations the
government of Malawi embarked in the Social Cash Transfer (SCT) programme
lt FIGURE 1 ABOUT HERE gt
1 2006 WHO manual
6
13
DRAFT NOT FOR CITATION
lt TABLE 1 ABOUT HERE gt
22 Mchinji pilot Social Cash Transfer (SCT) Programme of Malawi
The Social Cash Transfer (SCT) programme targets rural ultra-poor and labour-constrained
households to unconditional cash transfers The estimated transfer size is 78 USD per
household ranging between 4 USD for households with only one member to 13 USD for
families with four or more eligible members Embedded in the transfer the programme
provides a school bonus fee ranging between 13 USD per primary school age child and 26
USD per secondary school age child Participants are selected using a combination of proxy
means implemented at community level The programme covers 116000 individual
beneficiaries (28000 households) and with its current expansion planning to reach 300000
households by 2015
23 Malawi research design The impact evaluation of the Mchinji pilot SCT programme is a randomized longitudinal
design with a baseline household survey conducted in 2007 and two follow-up surveys
conducted respectively after 6 months and 12 months For the evaluation of the Malawi SCT
a one-year pilot of the programme was designed and implemented in the Mchinji District in
central Malawi Specifically four control and four treatment Village Development Groups
(VDCs) forming part of the original programme rollout were randomly selected to be part of
the evaluation These eight VDCs correspond to 23 villages Within each village Community
Social Protection Committees (CSPCs) identified eligible households according to the
national eligibility criteria and then ranked them in order to select the bottom 10 per cent for
inclusion to the programme
3 Theoretical framework As we mentioned in the introduction the aim of this paper is to test whether families enrolled
in the STC programme increased (i) significantly increased the home production of food such
as dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
7
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
production interact with the nutritional status of household members Also given the nature of
the policy interventions ndash CT programmes aiming at mitigating poverty in the short run while
improving the well being of the new generationsndash it is important to conduct the analysis over
the health status (eg height) of young individuals residing in targeted families Given these
premises in this paper we seek to answer the following question do children residing in
families that shift towards on-farm activities due to the transfer intervention also experience
larger benefit in terms of their nutritional status compared to counterfactual children where the
absence of the transfer did not generate a significant shift toward inndashfarm activities If we find
that poor households tend to produce a larger variety of nutrients (particularly protein based
food) out of own production and thus they shift toward on-farm activities we would also
expect household members children age 0-5 in our study to be better nourished and display
some significant improvement in health outcomes indicators as resilience to disease or more
interestingly height of treated individuals compared to the counterfactual Although
households could be net-sellers of food commodities this is unlikely to be the case in our
samples as the families targeted to the programme were selected across the poorest 10 per cent
of the population facing labour constraints and mainly involved with subsistence farming
While mitigating child malnutrition cannot be exclusively achieved through a raise in income
(Alderman et al 2003) the empirical evidence from Latin American programmes provides a
mix of evidence Mostly children of 0-36 months significantly benefited from CCT
interventions and that the impacts were larger for those programmes in which the transfer
value consisted of approximately 17-25 per cent or more of the pre-treatment household
income and older household members mostly mothers attended nutritional counselling
(Fiszbein et al 2009 Manley et al 2013 Leroy et al 2009 Gertler et al 2004 Maluccio et al
2005 among others) For example Maluccio et al (2005) found that children in the
Nicaraguan Red De Proteccion Social were 017 standard deviation (SD) taller compared the
4
13
DRAFT NOT FOR CITATION
counterfactual while Attanasio et al (2005) found that the Colombian Familias en accioacuten
significantly improved the height-for-age scores by 016 SD In Africa the empirical evidence
chiefly revolves around the Old Age Pension (OAP) programme and the Child Support Grants
(CSG) both implemented in South Africa Girls of age 0-5 living in households taking up the
OAP intervention were 223 cm significantly taller after two years the programme was rolled
out (Duflo 2003) However little empirical evidence exists on the link between agricultural
production and health status To the best of our knowledge one of the few studies looking at
the ability of agricultural production to fulfil the nutritional gap over the most vulnerable
families living in remote and rural areas is proposed by Muller (2008) Based on a sample of
Rwandan farmers the Author found that several food groups have a diverse effect on adult
health status However no studies try to directly disentangle such type of nexus under CT
programme framework
In an attempt to explore this relation our study uses data from the impact evaluation of the
Mchinji cash transfer pilot programme in Malawi conducted between 2007-2008 By
combining double difference (DD) method with Inverse Probability Weighting (IPW) to
restore pre-treatment balancing conditions we test the hypothesis that (i) households
participating in the programme significantly increased the home production of food such as
dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally test to test if and to what extent consumption out of home production of foods can
influence child outcomes By carrying out this type of analysis we conclude that households
boosting their home production (and consumption) of protein-rich foods as meat and fish due
to the transfer intervention also experience larger benefit in terms of child nutritional status
compared to those households which did not shift toward on-farm activities
5
13
DRAFT NOT FOR CITATION
The remainder of the paper is structured as follows section 2 gives a picture of malnutrition
in Malawi and describes the Mchinji SCT pilot programme and the research design section 3
describes the theoretical economic framework underlying the analysis section describes 4 the
empirical specification and the econometric techniques used to carry out the analysis section
5 discusses household and child level attrition in the sample section 6 presents the household
and child level characteristics section 7 provides the results and section 8 concludes
2 Malnutrition in Malawi and the Social Cash Transfer (SCT) programme in Malawi 21 Malnutrition in Malawi Chronic malnutrition is usually measured using standard linear growth index called the HAZ
index The HAZ is calculated by comparing the height for age of a child with a reference
population of well-nourished children We only focus on this nutrition outcome since it is
considered the best indicator of the long-term cumulative effects of under nutrition in
childhood development and we therefore define a child ldquostuntedrdquo or chronically malnourished
if her if her score is below ldquo-2rdquo The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a health child For a child of 36 months this translates into approximately 36 cm
difference1
Over the last decade Malawi experienced a considerable decline in chronic malnutrition
respectively dropping by 7pp (Table 1 and Figure 1) Yet approximately 1 child in 2 is still
stunted Moreover children living in the poorest families and in rural areas show levels of
chronic malnutrition which are higher compared to the national average (Table 1) To mitigate
poverty and seeking to guarantee a better and healthier future to young generations the
government of Malawi embarked in the Social Cash Transfer (SCT) programme
lt FIGURE 1 ABOUT HERE gt
1 2006 WHO manual
6
13
DRAFT NOT FOR CITATION
lt TABLE 1 ABOUT HERE gt
22 Mchinji pilot Social Cash Transfer (SCT) Programme of Malawi
The Social Cash Transfer (SCT) programme targets rural ultra-poor and labour-constrained
households to unconditional cash transfers The estimated transfer size is 78 USD per
household ranging between 4 USD for households with only one member to 13 USD for
families with four or more eligible members Embedded in the transfer the programme
provides a school bonus fee ranging between 13 USD per primary school age child and 26
USD per secondary school age child Participants are selected using a combination of proxy
means implemented at community level The programme covers 116000 individual
beneficiaries (28000 households) and with its current expansion planning to reach 300000
households by 2015
23 Malawi research design The impact evaluation of the Mchinji pilot SCT programme is a randomized longitudinal
design with a baseline household survey conducted in 2007 and two follow-up surveys
conducted respectively after 6 months and 12 months For the evaluation of the Malawi SCT
a one-year pilot of the programme was designed and implemented in the Mchinji District in
central Malawi Specifically four control and four treatment Village Development Groups
(VDCs) forming part of the original programme rollout were randomly selected to be part of
the evaluation These eight VDCs correspond to 23 villages Within each village Community
Social Protection Committees (CSPCs) identified eligible households according to the
national eligibility criteria and then ranked them in order to select the bottom 10 per cent for
inclusion to the programme
3 Theoretical framework As we mentioned in the introduction the aim of this paper is to test whether families enrolled
in the STC programme increased (i) significantly increased the home production of food such
as dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
7
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
counterfactual while Attanasio et al (2005) found that the Colombian Familias en accioacuten
significantly improved the height-for-age scores by 016 SD In Africa the empirical evidence
chiefly revolves around the Old Age Pension (OAP) programme and the Child Support Grants
(CSG) both implemented in South Africa Girls of age 0-5 living in households taking up the
OAP intervention were 223 cm significantly taller after two years the programme was rolled
out (Duflo 2003) However little empirical evidence exists on the link between agricultural
production and health status To the best of our knowledge one of the few studies looking at
the ability of agricultural production to fulfil the nutritional gap over the most vulnerable
families living in remote and rural areas is proposed by Muller (2008) Based on a sample of
Rwandan farmers the Author found that several food groups have a diverse effect on adult
health status However no studies try to directly disentangle such type of nexus under CT
programme framework
In an attempt to explore this relation our study uses data from the impact evaluation of the
Mchinji cash transfer pilot programme in Malawi conducted between 2007-2008 By
combining double difference (DD) method with Inverse Probability Weighting (IPW) to
restore pre-treatment balancing conditions we test the hypothesis that (i) households
participating in the programme significantly increased the home production of food such as
dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally test to test if and to what extent consumption out of home production of foods can
influence child outcomes By carrying out this type of analysis we conclude that households
boosting their home production (and consumption) of protein-rich foods as meat and fish due
to the transfer intervention also experience larger benefit in terms of child nutritional status
compared to those households which did not shift toward on-farm activities
5
13
DRAFT NOT FOR CITATION
The remainder of the paper is structured as follows section 2 gives a picture of malnutrition
in Malawi and describes the Mchinji SCT pilot programme and the research design section 3
describes the theoretical economic framework underlying the analysis section describes 4 the
empirical specification and the econometric techniques used to carry out the analysis section
5 discusses household and child level attrition in the sample section 6 presents the household
and child level characteristics section 7 provides the results and section 8 concludes
2 Malnutrition in Malawi and the Social Cash Transfer (SCT) programme in Malawi 21 Malnutrition in Malawi Chronic malnutrition is usually measured using standard linear growth index called the HAZ
index The HAZ is calculated by comparing the height for age of a child with a reference
population of well-nourished children We only focus on this nutrition outcome since it is
considered the best indicator of the long-term cumulative effects of under nutrition in
childhood development and we therefore define a child ldquostuntedrdquo or chronically malnourished
if her if her score is below ldquo-2rdquo The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a health child For a child of 36 months this translates into approximately 36 cm
difference1
Over the last decade Malawi experienced a considerable decline in chronic malnutrition
respectively dropping by 7pp (Table 1 and Figure 1) Yet approximately 1 child in 2 is still
stunted Moreover children living in the poorest families and in rural areas show levels of
chronic malnutrition which are higher compared to the national average (Table 1) To mitigate
poverty and seeking to guarantee a better and healthier future to young generations the
government of Malawi embarked in the Social Cash Transfer (SCT) programme
lt FIGURE 1 ABOUT HERE gt
1 2006 WHO manual
6
13
DRAFT NOT FOR CITATION
lt TABLE 1 ABOUT HERE gt
22 Mchinji pilot Social Cash Transfer (SCT) Programme of Malawi
The Social Cash Transfer (SCT) programme targets rural ultra-poor and labour-constrained
households to unconditional cash transfers The estimated transfer size is 78 USD per
household ranging between 4 USD for households with only one member to 13 USD for
families with four or more eligible members Embedded in the transfer the programme
provides a school bonus fee ranging between 13 USD per primary school age child and 26
USD per secondary school age child Participants are selected using a combination of proxy
means implemented at community level The programme covers 116000 individual
beneficiaries (28000 households) and with its current expansion planning to reach 300000
households by 2015
23 Malawi research design The impact evaluation of the Mchinji pilot SCT programme is a randomized longitudinal
design with a baseline household survey conducted in 2007 and two follow-up surveys
conducted respectively after 6 months and 12 months For the evaluation of the Malawi SCT
a one-year pilot of the programme was designed and implemented in the Mchinji District in
central Malawi Specifically four control and four treatment Village Development Groups
(VDCs) forming part of the original programme rollout were randomly selected to be part of
the evaluation These eight VDCs correspond to 23 villages Within each village Community
Social Protection Committees (CSPCs) identified eligible households according to the
national eligibility criteria and then ranked them in order to select the bottom 10 per cent for
inclusion to the programme
3 Theoretical framework As we mentioned in the introduction the aim of this paper is to test whether families enrolled
in the STC programme increased (i) significantly increased the home production of food such
as dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
7
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
The remainder of the paper is structured as follows section 2 gives a picture of malnutrition
in Malawi and describes the Mchinji SCT pilot programme and the research design section 3
describes the theoretical economic framework underlying the analysis section describes 4 the
empirical specification and the econometric techniques used to carry out the analysis section
5 discusses household and child level attrition in the sample section 6 presents the household
and child level characteristics section 7 provides the results and section 8 concludes
2 Malnutrition in Malawi and the Social Cash Transfer (SCT) programme in Malawi 21 Malnutrition in Malawi Chronic malnutrition is usually measured using standard linear growth index called the HAZ
index The HAZ is calculated by comparing the height for age of a child with a reference
population of well-nourished children We only focus on this nutrition outcome since it is
considered the best indicator of the long-term cumulative effects of under nutrition in
childhood development and we therefore define a child ldquostuntedrdquo or chronically malnourished
if her if her score is below ldquo-2rdquo The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a health child For a child of 36 months this translates into approximately 36 cm
difference1
Over the last decade Malawi experienced a considerable decline in chronic malnutrition
respectively dropping by 7pp (Table 1 and Figure 1) Yet approximately 1 child in 2 is still
stunted Moreover children living in the poorest families and in rural areas show levels of
chronic malnutrition which are higher compared to the national average (Table 1) To mitigate
poverty and seeking to guarantee a better and healthier future to young generations the
government of Malawi embarked in the Social Cash Transfer (SCT) programme
lt FIGURE 1 ABOUT HERE gt
1 2006 WHO manual
6
13
DRAFT NOT FOR CITATION
lt TABLE 1 ABOUT HERE gt
22 Mchinji pilot Social Cash Transfer (SCT) Programme of Malawi
The Social Cash Transfer (SCT) programme targets rural ultra-poor and labour-constrained
households to unconditional cash transfers The estimated transfer size is 78 USD per
household ranging between 4 USD for households with only one member to 13 USD for
families with four or more eligible members Embedded in the transfer the programme
provides a school bonus fee ranging between 13 USD per primary school age child and 26
USD per secondary school age child Participants are selected using a combination of proxy
means implemented at community level The programme covers 116000 individual
beneficiaries (28000 households) and with its current expansion planning to reach 300000
households by 2015
23 Malawi research design The impact evaluation of the Mchinji pilot SCT programme is a randomized longitudinal
design with a baseline household survey conducted in 2007 and two follow-up surveys
conducted respectively after 6 months and 12 months For the evaluation of the Malawi SCT
a one-year pilot of the programme was designed and implemented in the Mchinji District in
central Malawi Specifically four control and four treatment Village Development Groups
(VDCs) forming part of the original programme rollout were randomly selected to be part of
the evaluation These eight VDCs correspond to 23 villages Within each village Community
Social Protection Committees (CSPCs) identified eligible households according to the
national eligibility criteria and then ranked them in order to select the bottom 10 per cent for
inclusion to the programme
3 Theoretical framework As we mentioned in the introduction the aim of this paper is to test whether families enrolled
in the STC programme increased (i) significantly increased the home production of food such
as dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
7
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
lt TABLE 1 ABOUT HERE gt
22 Mchinji pilot Social Cash Transfer (SCT) Programme of Malawi
The Social Cash Transfer (SCT) programme targets rural ultra-poor and labour-constrained
households to unconditional cash transfers The estimated transfer size is 78 USD per
household ranging between 4 USD for households with only one member to 13 USD for
families with four or more eligible members Embedded in the transfer the programme
provides a school bonus fee ranging between 13 USD per primary school age child and 26
USD per secondary school age child Participants are selected using a combination of proxy
means implemented at community level The programme covers 116000 individual
beneficiaries (28000 households) and with its current expansion planning to reach 300000
households by 2015
23 Malawi research design The impact evaluation of the Mchinji pilot SCT programme is a randomized longitudinal
design with a baseline household survey conducted in 2007 and two follow-up surveys
conducted respectively after 6 months and 12 months For the evaluation of the Malawi SCT
a one-year pilot of the programme was designed and implemented in the Mchinji District in
central Malawi Specifically four control and four treatment Village Development Groups
(VDCs) forming part of the original programme rollout were randomly selected to be part of
the evaluation These eight VDCs correspond to 23 villages Within each village Community
Social Protection Committees (CSPCs) identified eligible households according to the
national eligibility criteria and then ranked them in order to select the bottom 10 per cent for
inclusion to the programme
3 Theoretical framework As we mentioned in the introduction the aim of this paper is to test whether families enrolled
in the STC programme increased (i) significantly increased the home production of food such
as dairy products or meat and fish (ii) cash transfer programmes can significantly improve the
7
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
health status of young children measure as height-for-age z-scores and stunting rates and (iii)
finally to test if and to what extent consumption out of home production and thus agricultural
production can influence child outcomes These hypotheses can be explored using economic
models synthesizing household andor individual behaviour described through utility
maximization processes and economic constraints Particularly in the case of the agricultural
production choices we make use of an agricultural household model where households are
both utility-maximizing consumers of agricultural goods and profit-maximizing producers of
those goods (Singh et al 1986) On the other hand when we move into the child level
analysis we employ the health production function model (HPF) that seeks to identify the
underlying mechanism through which a household produces better health outcomes given a
set of (health) inputs the available technology and household budget constraints (Becker
1965 Stanford 1995 CEBU study 1995 Rosenzweig et al 1983 and Handa et al 2012)
In these contexts a common entry point to analyze household decision making function is to
make use of an agricultural household model in which rural households are both utility-
maximizing consumers of agricultural goods and profit-maximizing producers of those goods
while potentially facing market constraints (Singh et al 1986)
The model assumes that households are price takers prices of goods are exogenously
determined and they are not affected by labour credit or transportation constraints and they
are able to hire labour at the current market wage andor obtain credit with no restriction In
the absence of market failures production and consumption choices are ldquoseparablerdquo and first
the households maximize quantities to be produced given market prices and then maximize
their consumption
However rural households in developing countries often face significant market failures
which limit their capability of agricultural production and productivity resulting in poor living
conditions and inadequate food energy and nutrient intakes Examples of market failures are
8
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
liquidity and credit constraints two of the main factors limiting poor agricultural households
from investing optimally (Rosenzweig and Wolpin 1993 Fenwick and Lyne 1999 Lopez
and Romano 2000 Barrett et al 2001 Winter-Nelson and Temu 2005) Without access to
adequate credit markets or insurance agricultural households are likely to undergo low-risk
low-return strategies either in production or the diversification of income sources Therefore
agricultural households may then make decisions to ensure that they have enough food to eat
but not necessarily what would be the most profitable or more relevant for the sake of our
analysis the most nutrient ones For example they could decide to produce more of staple
crops as cereals or grains while limiting the production of dairy products or meat and fish In
the face of such constraints the production and consumption decisions of agricultural
households can be viewed as ldquonon-separablerdquo in the sense they are jointly determined If
household production and consumption decisions are non-separable cash transfers may be
able to help overcome several of these constraints Particularly transfers are a regular (eg
monthly or bimonthly) and predictable source of income allowing poor smallholders in
addressing part of the market constraints they are usually affected by and move into more
valuable (both in terms of monetary and nutritional value) home production of foods In short
we would expect households enrolled in the STC programmes significantly diversifying home
production of foods and thus moving into production of ldquobetterrdquo foods from a nutritional
perspective As a consequence we would also expect health status of younger household
members ndash children of age 0-5 years in our sample ndash to improve as now those families
receiving the transfer have access to more and better quality of calories coming from on-farm
activities and that are fundamental for child growth
In the standard microeconomic theory a relation between child health outputs and health
inputs (eg energy macronutrient and micronutrient intakes) are modelled through the Health
Production Function (HPF) This framework is similar to the household production model
9
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
introduced by Becker (1965) and seeks to identify the underlying mechanism through which a
household produces better health outcomes given a set of (health) inputs the available
technology and household budget constraints (Stanford 1995 CEBU study 1995 Rosenzweig
et al 1983 and Handa et al 2012) To determine the optimal level of inputs for each childs
health production process a family undergoes a utility maximization process in which child
nutritional outputs (eg height or weight) are shaped as cumulative variables resulting from a
dynamic interaction between ldquostockrdquo and ldquoflowrdquo components Stock components are factors
that depend on accumulation process and whose realization is determined over a certain
period of time Examples of stock variables are resilience to disease (genetic endowment)
birth at weight or lagged values of individual height On the other hand flow components
such as calorie macronutrient and micronutrient intakes are produced with current inputs and
consumed in the current period (Handa et al 2012) In our case part of the food intake would
be obtained through household expenditure while part of it through home production of foods
ndash what we are mostly interested in the present article Along with calorie intakes other
factors such as parental education or the education of the caregiver orphan status and
children birth order preventive health check-ups and pre-natal care have been found
determinants of childrenrsquo health In the context of the Mchinji district poverty has hit hardest
through limited access and poor quality of community infrastructures low level of parental
education (particularly maternal education) and insufficient pre-natal health care vaccinations
and preventive health check-ups all of which represent developmental short-falls childrenrsquos
everyday lives As a demand side intervention the SCT does not directly address issues
related to inefficient and poor quality of community level infrastructure or influencing
maternal education2 On the other hand as mentioned above we would expect the transfer
altering production choices and shifting toward better nutrients Last because any type of
2 Also notice that the data were collected between 2007 to 2008 therefore we are not able to detect long-run changes in standard of living
10
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
human health indicator is influenced by biological characteristics genetic endowments enter
in the equation as an unobservable characteristic3
4 Empirical specification
41 First difference and double difference estimator Our empirical strategy is based on a first difference (FD) and double-difference (DD)
methodology depending on the outcome of interest We made use of the FD when we
analyzed the impact of the SCT on agricultural production since information on consumption
out of own production were only collected at follow-up whereas we employed the DD
methodology when analyzing the child linear growth Following Dehejia (2004) the simplest
version of the both estimators can be written as
(1) τ = E( =1)- E( =0)
Which can be estimated using the following regression model
(2) = + τ +
In the context of experimentally designed evaluations a random allocation of the treatment
would lead to unbiased estimates of the programme impact since (i) the potential outcomes
are independent from the treatment (Y1i Yoi perp Ti) and (ii) observationsrsquo characteristics are
independent from the treatment as well Xi(Xi perp Ti) This implies that if the SCT recipient
and the counterfactual were truly randomly selected we would observe a low level of
covariate unbalancing between the treatment and the counterfactual group However
significant differences highlighted in Table 2 raise concerns on the reliability of the control
group as a ldquogoodrdquo counterfactual
A first approach to removing potential bias arising from the misallocation of the SCT is to
control for a vector ldquoXrdquo of baseline characteristics such as household demographics gender
and head of the household head etc In this case we can expand (2) as
3 As we will later explain double difference methodology helps to remove unobserved characteristicswhich are constant over time as well as genetic trait effects
11
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
(3) = + τ + 13 +
Equation (3) is used to test (i) whether treated households would experience a boost in the
agricultural sector by producing more and nutritionally richer foods and (ii) childrenrsquo s height
status gains from the transfer programme As we have already mentioned the only but
substantial difference is that in the case of the DD estimator the outcomes of interest (child
height and stunting) are observed in two periods of time before and after the SCT rolled out
over the treatment group By taking the difference between the treatment group outcomes
before and after the households receive the cash transfer and subtracting from it the
difference in the control outcomes we obtain the DD estimate
Our ultimate goal however is to explore the nexus between child health status with
household level agricultural production As mentioned in section 3 child health status can be
shaped as shaped as cumulative variables resulting from a dynamic interaction between
ldquostockrdquo and ldquoflowrdquo components Flow components as calories macronutrient and
micronutrient intakes are produced with current inputs and consumed in the current period In
the context of poor and rural households engaged with subsistence farming nutritional inputs
are produced at household level and consumed by the same families To detect if agriculture
production can truly influence child height outcomes we should interact it with the treatment
status which is equivalent to rewrite equation (3) as follows
(4) i=ao + τTi + 13Xi+13Zi + 13TZ13i + Ei Where the Z vector is a set of variables representing current consumption out of own
agricultural production of foods such as cereals legumes dairy products and meat and fish
collected at follow-up and TZ1 is the interaction term between the treatment status and
consumption out of home production of the ldquonrdquo food group for example home production of
dairy and eggs If through equation (3) we observed positive and significant impacts of the
SCT on consumption out of own production and child health status and subsequently we
found positive and significant impacts in equation (4) larger then what previously resulting
12
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
from the treatment status in the child level equation we would be able to link the agricultural
production with the cash transfer ain fulfilling the nutritional gap of the youngest In addition
we would expect some foods groups such as meat and fish ndash rich in high quality protein -
more valuable than other food groups in improving linear growth of young children
One might argue that by including collected agriculture production variables collected at
follow-up we would pollute the model with endogeneity as adults in charge of the child
feeding process would adjust their production of foods by observing child height (Stanford
1995 CEBU study 1995 Rosenzweig et al 1983) In addition genetic endowment rather than
the treatment could lead to diverse growth trajectories which would ultimately bias our
estimate However by using DD approach in the child level analysis we ruled out these
hypotheses since the child outcome of interest is not height per se but rather the height
variation which we believe it would be hardly observed by adults in the households The same
argument apply to child stunting measured at different cut-off points of the child height
distribution4 In addition the DD method allow to remove unobservables as genetic traits by
taking the difference of treatment and control group between baseline and follow-up
42 Propensity score and inverse probability weighting
Since the randomization was stratified at VDC level and then within each geographical area
the targeting process relied on community based criteria some concerns still remain on the
ability of the DD estimator in producing unbiased estimates of the SCT programme In
addition when the data are affected by error measurement or missing values in the variables
as it is the case in the child sample the reliability of the DD is further weakened (Hirano and
Imbens 2001) even when in presence of an optimal treatment randomization
4 The definition13 of stunting is based on13 a cut-shy‐off point obtained13 comparing13 a child13 standardized13 height to13 apopulation13 of healthy children If the HAZ score of a child is below ldquo-shy‐2rdquo she would13 be defined13 as stunted or13 chronically malnourished Knowing whether a child is stunted or not requires clinical heath check-shy‐upsperiodically attended by the family which were very unlikely to happen13 at the time the data were collected In addition as13 we measured the variation in stunting child status it is13 very unlikely that olderfamily members in charge of13 feeding process had these information
13
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
In cases in which the data analysed are affected by both covariate unbalancing and missing
data Rosenbaum and Rubin (1983) first showed that unbiased estimates of the treatment
allocation can still be independent from the outcomes of interests if conditioned on
observational characteristics (unconfoundness assumption)
(5) perp |
And that condition n ldquoXrdquo is equivalent to condition on p(X) the estimated probability of
joining the treatment In formulas equation (4) is equivalent to
(6) perp | ( )
Usually the p(X) is modelled over both treated and control observations using a vector of X
variables as individual or household level characteristics Conditioning on the propensity
score nets out bias from impact estimates as long as the p(X) eliminates mean differences at
baseline that is it restores covariate balancing Indeed Rosenbaum (2010) states that
ldquopropensity score is a mean to an endrdquo to balance observed covariates
In Figure 2 Panels A B and C show Kernel density estimates of the propensity score by the
SCT and the counterfactual group over the 3 data sets used in the analysis In each case the
mean PS difference between the T and C group is always statistically significant (1 level)
implying some level of unbalancing5 This is particularly true in the case of the child sample
in which the wedge between the treatment PS and the control PS is considerable Hence
simply conditioning on X would not correctly identify the estimate due to heterogeneous
effect Also in the case of the child data some of the observations are off common support
which limits the use of the PS
To address this estimation problem we use Inverse Probability Weighting (IPW) proposed by
Hirano and Imbens (2001) The core of the IPW method consists of using the inverse of the
5 As shown by Rubin (1983) and Rosenbaum (2010) PS helps to resolve the curse of dimensionality issue and if so can be considered a reliable summary13 statistic to13 evaluate if covariate balancing is an issue in theanalysed sample PS13 mean values by13 treatment status are available upon request
14
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
estimated PS as a weight in the FDDD estimator Weighting by the inverse of the estimated
propensity score can also achieve covariate balance and in contrast to matching and
stratificationblocking uses all of the observations in the sample (Sacerdote 2004 and Todd et
al 2009) Generally the treatment estimator weighted by the IPW takes the form of
lowast lowast (7) = =minus ( )ndash
( )
in which p(Xi ) is the estimated propensity score The consistency of the IPW estimator relies
on its ability to restore balancing conditions as we showed in Table 2 Once the controls are
balanced virtually all baseline differences are eliminated in all the 3 samples Also the
distribution of the PS tends to uniformly overlap after we weight it by the IPW (Fig 2 Panels
B and D)
lt FIGURE 2 ABOUT HERE gt
In our case we are interested in the Average Treatment across the Treated (ATT) Therefore
we will weigh the DD estimator by6
p X(8) w T X = 1 minus T lowast ---
In which p X is the estimated propensity score and w T X is the IPW weight which
depends both on the treatment status and the X set of covariates used to estimate the PS
Intuitively the IPW estimator put more emphasis on those counterfactual observations having
an estimated PS similar to that of the treatment group while underweighting those for which
the PS is relatively small In Tables 2 and 3 we show unweighted and IPW weighted control
mean differences between the treat T and C group which will be later used in the
econometrics analysis In the case of Malawi the IPW virtually remove all the baseline mean
differences between the treatment and the control group as shown in Table 2 so we conclude
that the IPW technique worked in balancing observable controls
6 When T=1 the IPW13 reduces to 1 so treated observations are not weighted Also notice that the IPW13 isvalid until the13 propensity13 score is13 bounded in the following interval (0ltp(X)1) If so the IPW is13 valid untilthe propensity score does not13 take either13 value ldquo1rdquo or13 ldquo0rdquo
15
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Last although child level attrition is not an issue from a strict statistical perspective we think
it is more appropriate to consider the estimated treatment effect as the Sample Average
Treatment Effect on the Treated (SATT) rather than the Population Average Treatment Effect
(PATT) and therefore not giving any external validity to our results
5 Household and child level attrition in the samples
Since non-random attrition in the household or child data could severely bias our estimates
we run a set of tests to determine this eventuality In fact the central concern in the analyses
of attrition ndash and missing data in general ndash is selection bias that is a distortion of the
estimation due to non-random patterns of attrition (Alderman et al 2001)
51 Household level attrition
In Malawi the baseline survey contained 402 and 419 intervention and control households
respectively The 2008 follow-up contained 365 treatment households and 386 control
households Again a comparison of household characteristics between the two waves
indicates that attrition is random Covarrubias et al (2012) and Boone et al (forthcoming)
also confirm our findings
52 Child level attrition Sample households (751) contained 563 children below age 5 239 living in the control
households and 315 living in the treatment households We excluded child observations with
misreported dates of birth negative height growth more than 30 cm growth in 12 months and
having a HAZ scores above ldquo6rdquo or below ldquo- 6rdquo We remained with 280 observations at
baseline while 273 at follow-up Of those observations only data for 208 children could be
used to construct a panel data set7 (T = 106 C = 102) residing respectively in 77 treatment
households and 76 control households Because of high attrition in both samples we
performed some analysis to check whether dropping child observations at baseline and
follow-up could bias our estimates Particularly we conducted t-tests for mean differences in
7 Eligible children13 in13 Malawi are those of age 0-shy‐6013 months at baseline while age 12-shy‐7213 at follow-shy‐up that is after one13 year the13 programme13 rolled-shy‐out
16
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
z-scores for children retained in the panel and those dropped out at baseline as well as at
follow-up on the whole sample and then by treatment group8 We found no significant non-
random panel attrition within the Mchinji pilot programme in Malawi Attrition analysis
results are shown in the Appendix
53 Data sets used in the analysis
Finally for the sake of the analysis we used 3 data sets First we utilize the full sample to
investigate the impact of the SCT over levels and shares of household consumption
expenditure Second we restrict the expenditure analysis only to that segment of the
households 153 families in total hosting children for which we have reliable height
measures We refer to the second data set as the Small Household sample (SHH) Last we
analyze the impact of the SCT over child health indicators controlling for households
characteristics obtained from the SHH sample We refer to this child level data set as the
Child Household sample (CHH)
6 Characteristics of the data
61 Household characteristics
We start the analysis by providing a description of the household characteristics and the
outcomes of interest which will be the focus of the study In Table 2 we report descriptive
statistics for variables linked to the eligibility criteria household characteristics vulnerability
to shocks participation in safety net interventions and child characteristics Descriptive
statistics for the outcome indicators are in Table 3 Panel A presents characteristics of the full
household sample while panel C reports data from the SHH sample namely the sample we
obtain when restricting the analysis only to those families for which we have reliable
information on child anthropometrics and that will be used later on to assess the impact of the
SCT on child health outcomes The data are also disaggregated by treatment status to show
mean differences between groups at baseline We use T-tests to identify any significant mean
8 We found a similar procedure used in Handa et al 2012
17
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
differences over the variables of interest between the treatment and the counterfactual group
Panel B and Panel D reports the same variablesindicators weighted by the IPW
As expected rural households in the survey are ultra-poor and food insecure With a reported
average monthly per capita expenditure of 19243 MLS9 the SCT programme targeted the
poorest of the poor in Malawi
lt TABLE 2 ABOUT HERE gt
Over half of the households had at most one meal in the day prior the interview and over half
owned 1 or fewer assets One of the main criteria to select the households targeted to the SCT
programme was ldquolabour constrainedrdquo defined as having no available household labour or as
having a dependency ratio larger than three10 In fact we found that 72 per cent of the
households have a dependency ration larger than ldquo3rdquo
The household head characteristics along with the household composition and the orphan
status of the children reflect the demographic profile of a segment of the population heavily
affected by HIVAIDS pandemic Heads of households are primarily elderly single females
with few years of schooling and approximately 16 per cent being disabled or affected by a
chronic disease The average household is made up of 4 members half of whom are children
under 15 Households in the sample are vulnerable to shocks approximately 60 per cent of
the families experienced a natural or financial shock or faced a sudden increase in commodity
prices Risk coping strategies adopted by households to mitigate the impact of hunger and
severe deprivation includes ganyu labour11 and begging for food and money (38 per cent) The
data show signs of previous policy interventions aiming at improving food security in the
Mchinji district Over a quarter of households had benefitted in the past by free food and
9 The exchange rate is 140 MLS per 1 USD 10 For those households with13 no adult members we replaced the denominator with 01 and then13 we calculated the dependency ratio 11 Ganyu13 labor is a rural low wage informal activity utilized by poor households in Malawi to cope with negative shocks13 and during the hungry season in Malawi Covarrubias13 et al (2012)13 documented that13 the average number13 ofdays worked13 in the sample was 75 at13 baseline which was significantly reduced after the SCT was implemented
18
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
maize distribution and 42 per cent had received other type of subsides12
Overall demographic characteristics of the SHH sample (Panel C) are in line with the full
sample matching the profile of ultra-poor and labour constrained families SHH households
have on average lower monthly per capita expenditure The number of children is almost as
twice as large compared to the full sample the household heads are considerably younger and
slightly more educated and these households are less likely to have at least one household
member who is able to work Children age 0-5 are on average 36 months old while girls make
up roughly half of the children
62 Child characteristics
As mentioned in section 2 we focused on the height to detect cumulative patterns in the child
health status The most popular approach to calculate stunting rates relies on standardized
indexes which compare the observations in the sample of interest with a reference population
of well-nourished children In particular we calculated a linear growth outcome the Height-
for-Age Z-score (HAZ) index The interpretation of the HAZ is given in terms of Standard
Deviation (SD) distance from the mean population level For example if the HAZ associated
to an individual is ldquo-1rdquo a child will be 1 SD smaller compared to what we would normally
observe for a healthy child of that age For a child of 36 months this translates in being
approximately 3813 cm shorter compared to an healthy child Once the HAZ is calculated a
child is defined moderately stunted if her HAZ score is below ldquo-1rdquo stunted if her HAZ score
is below ldquo-2rdquo while severely stunted if her HAZ is below ldquo- 3rdquo By considering different cut-
off points that take into account the severity of the malnutrition status we seek to detect and
gauge transfer impact over the most critical segments of the HAZ distribution rather than only
focusing on the ldquocanonicalrdquo stunting definition which is usually based on the value ldquo-2rdquo the
cut-off point used to determined whether a child is stunted or not
12 I is important13 to note that13 at13 the time the baseline survey was conducted neither control13 nor treatment13 familieswere not enrolled in other type of safety nets13 programmes13 (for13 more details13 see Miller13 et al 2011)13 See 199 ldquoWHO Global Database13 on Child Growth13 and13 Malnutritionrdquo manual for detailed information
19
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
lt TABLE 3 ABOUT HERE gt
At baseline the mean HAZ value was -187 meaning that on average a child in the sample
was -187 standard deviation compared to the reference population For a child of 36 months
this translates into approximately 7 cm difference (38 cm per standard deviation) In terms of
stunting we found respectively that 3 children in 4 are moderately stunted (737 per cent) 1
child in 2 (452 per cent) is chronically malnourished14 while 1 in 4 is severely stunted (25 per
cent) In addition the child nutritional status drops rapidly by age group and then it tapers off
(Figure 3) as it is has been commonly observed over poor and vulnerable children
lt FIGURE 3 ABOUT HERE gt
Table 2 also presents descriptive statistics by treatment status and level of statistical
significance in order to assess the validity of the control group for the impact analysis The
two groups are virtually identical when compared over the set of outcome indicators With the
exception of purchase of other foods consumption out of expenditure incidence of vegetables
food purchased from street vendors and non alcoholic beverages no other significant
differences exist between the treatment and the control The same holds in the SHH sample
whereby differences arise only on few items Conversely as mentioned earlier we observe
disparities in the eligibility criteria and the demographic characteristics Households in the
full as well as SHH sample show significant mean differences in the demographic
characteristics and eligibility criteria depending on the treatment arm and in both samples the
SCT families are on average worse-off
For the sake of our analysis the outcome indicators do not need to be balanced as potential
differences at baseline are removed by the DD method However baseline characteristics
should be similar across the two arms of the sample As described earlier in an attempt to
mitigate shortfalls in the experiment design we made use of the IPW technique The next step
entails testing the efficiency of IPW in reweighting the baseline controls Overall the IPW
14 As shown in Table 1 last DHS estimate of stunting are respectively 482 in rural areas and 555 in the bottomwealth quintile suggesting that the Mchinji data are in line with other data source
20
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
does well in restoring covariate balancing In fact virtually all the mean baseline differences
between the two arms of the experiment are removed when weighted (Table 2) In fact after
reweighting the data only the number of days spent in ganyu labour in the SHH sample
presents a significant mean difference
62 Consumption of food out of own production
The questionnaire collected information on the primary source of specific types of food
consumption naming own production purchases or gifts received as the possible sources
lt TABLE 3 ABOUT HERE gt
However these questions were not elicited at baseline (only household purchases were
collected) and therefore we cannot present baseline values of consumption out of own
production In addition the questionnaire did not collect quantities consumed by the
households but rather only the amount of the foods consumed This prevented us to calculate
calories and nutrients using the available data However we could calculate per capita
monetary values of foods home produced and consumed which we believe are sufficient to
approximate In fact other things equal the larger the per capita value of any food group
consumed out of home production the more likely that each individual in the household
would consume more of that food group15 As shown in Table 3 we disaggregated foods in
seven groups in order to highlight their nutritional value (carbohydrates proteins vitamins
etcetc) as well as well as to count for heterogeneity consumption out of home production (i)
cereals (ii) roots and tubers (iii) pulses (iv) vegetables (v) fruits (vi) meat and fish and
(vii) dairy products While we cannot conclude that the SCT significantly improved on-farm
agricultural production we see that the value of foods consumed and home produced is higher
in the treatment compared to the control group in both the full sample and the SHH sample
15 Assuming a unitary household model in which resources are equally distributed While this is an unrealistic assumption in13 our regression13 models we control13 for household composition and otherdemographics13 to count13 for13 within household heterogeneity in the allocation of resources with respect13 tothe outcome of interest
21
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
For example in the full sample the treatment group per capita home production of meat and
fish is equal to 11 MWK while in the control group is approximately (Table 3)
7 Results
We first start the analysis by presenting impacts of the SCT over own production of food
measured through monthly per capita value expressed in MWK Then we move to the child
level analysis in which we gauge the impact of the transfer programmes over HAZ z-scores as
well as stunting rates Table 4 presents SCT impact estimates over food consumption out of
home production Table 5 show child level impact estimates on health outcomes while Table 6
when including food consumption controls out of own production Table 7 presents
heterogeneity analysis by interacting the treatment status with each of food consumption food
group All the regression results are controlled for the baseline characteristics listed in Table
2 and when we tested the hypothesis food consumption out own production would influence
the nutritional status of children age 0-5 at baseline we also included additional controls that
were used as dependent variables in the household level analysis Coefficient estimates are
always weighted by the IPW In the Appendix we also presented the propensity score
estimation computed using a probit model
71 Impact of the SCT on agricultural on household consumption out of own production
Regression results show a significant impact on the incidence of own production of food
group all of which are key determinants of human nutritional status particularly at early stage
of life Boone et al (forthcoming) analyzed the impact of the SCT over ownership of
agricultural inputs number of crops harvested and labour activities concluding that the
programme led to higher agriculture production cause by an increase in both existing family
labour and the return of additional labour (Soares et al 2012)
22
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
lt TABLE 4 ABOUT HERE gt
Household level analysis of consumption out of own production corroborate this hypothesis
Under the STC programme families experienced a gain in each of the seven food groups we
analyzed For example per capita home consumption of cereals increased by 564 MWK
(pgt001) while production of pulses by 305 MWK (pgt001) Moreover consumption out of
home production of meat and fish increased by 102 MWK (pgt001) When we restrict the
analysis to the SHH sample we found similar results Overall these findings show that at
end-line families targeted to the transfer programme experienced a rise on agricultural
production leading to a significant increase in the food consumption out of own production In
this context what is the net effect over nutritional status of small children We explore it the
next sub-section
72 Impacts of the transfer programmes on child nutritional status
Given the large effect on consumption out of food own produced a natural question is if the
SCT impacted child nutritional status Early developmental shortfalls in childrenrsquo s living
condition contribute substantially to the intergenerational transmission of poverty through
reduced cognitive skills employability productivity and overall well-being in later life
(Alderman 2011) The advantage of using height-for-age scores is that the height measure is
standardized with respect to a reference healthy population given the age in months and the
gender of the children The flipside of the z-score measures is that variation in the height is
measured through standard deviation and thus it must be reinterpreted after the estimation is
carried out As we mentioned in section 2 ldquo1rdquo SD difference for a child of age 36 months
would approximately corresponds to 36 cm in the reference population The Mchinji SCT had
a significant impact on linear growth We first discuss the effect of the transfer on linear
growth and then we analyze SCT impacts on stunting at different cut-off points A year after
the programme rolled out the HAZ index rose significantly (5 level) by 0221 SD across
23
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
children living in beneficiary households For a child of age 36 months this would translate in
an increase of approximately 083 cm
lt TABLE 5 ABOUT HEREgt
The result is consistent with some of the earlier evidence of social protection programme in
Latin America and South Africa (see Maluccio et al 2006and Duflo 2003) Following the
improvement in linear growth children in the treatment group compared to the counterfactual
are respectively 806pp (pgt005) less likely to be moderately stunted 137 pp (pgt005) less
likely to be stunted while 603 pp less likely to be severely stunted yet not in a significant
manner In absolute terms moderate stunting significantly dropped by 657pp (7640086)
while stunting decreased by 621 (4530137) amongst children under the SCT programme
73 Linking agriculture production to child height status
What is the role of consumption from on-farm activities over the linear growth of the
children Given substantial and positive impacts of the transfer programme in both countries
on household consumption out of own production as well as ownership of agricultural assets
we might expect some direct effect on household members nutritional status16 In Table 6 we
included in the regression follow-up controls on food consumption out of home production
while in Table 7 we reported the interaction between the treatment and the food groups We
start the analysis treating the problem as an omitted variable issue Particularly by introducing
in the regressions control variables indicating per capita household consumption out of own
production disaggregated by food groups we should observe a significant change in the
treatment coefficient If household consumption out of agricultural production of matter in
16 We assume that families behave as a single unit ie an equal intra-shy‐household13 allocation of the resources produced and consumed On13 the one hand family members in charge of13 decision related to feed young children could engage in investing type behavior that13 is they would favorite and better13 nourish childrenconsidered already healthier On the contrary they could shift to compensating behavior13 and thus trying13 to13 feed better those children which look disadvantaged and trying to recover their health status We chooseto keep as neutral as possible assumption since the data do not13 directly allow to test13 for13 this hypothesis
24
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
influencing child growth trajectory we should expect the treatment coefficient decrease in
magnitude and possibly coefficient of consumption out of own production should have a
positive sign Apparently when including such type of controls we found some contradictory
results The impact of SCT on HAZ score stunting and severe stunting are almost identical
varying only by few decimal points On the other hand the impact of the SCT on moderate
malnutrition (column (6) of Table 6) decrease from - 0086 (pgt005) to -012 (pgt0001)In
addition the coefficient estimates of consumption out of own agricultural production does not
always go in the hoped direction yet with the exception of consumption out of own
production of dairy products in column (6) of Table 6 they are always highly not significant
lt TABLE 6 ABOUT HEREgt
74 Heterogeneity in the agricultural production and child height
A more appropriate way to seek identifying whether some link exists between agricultural
production and child linear growth is to interact the consumption out home production
variables disaggregated by food groups with the treatment status and thus conducting
heterogeneity analysis17 over the sample of interest
In theory we should find children living in families consuming meat and fish from home
production andor dairy products and eggs taller than their counterparts and since we
previously shown that the transfer contributed to boost home production of such type of foods
we can establish an ldquoindirectrdquo link between the transfer agricultural production and child
health status Our findings corroborate this hypothesis As we shown at the beginning of this
section the ATT impact of the programme on child height status was 021 SD (pgt005)
lt TABLE 7 ABOUT HEREgt
Children living in families consuming vegetables and meat and fish from own production
experienced respectively an improvement in height equal to 0519 SD (pgt001) The gain in
17 See section 4 for more details
25
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
child height is striking and in the case of home production of meat and fish since the impact is
as twice as large compared to the ATT over the full sample We found similar patterns also
when analyzing stunting where we found children living in families consuming meat and fish
from own production 427 percent less likely to be stunted compared to the counterfactual
This corresponds to a 193pp (04270453) reduction in the stunting rate amongst treated
children also living in families consuming meat and fish out of home production Along with
this we also found some positive and significant result when interacting the treatment status
with consumption out of home production of vegetables being the estimated coefficient equal
to equal to 232 SD (pgt01) In addition moderate stunting significantly decreased more
compared to the initial ATT being children in this group 116 pp less likely to be moderately
malnourished Consumption out of own production of eggs and dairy products also seems
significantly but weakly associated with stunting rate of young children compared to the
overall treatment effect over the treated Interestingly we found some statically significant
results over families consuming pulses and cereals and grains out of own production yet the
treatment coefficient narrows down compared to the overall ATT effect
8 Conclusion
In this paper we analyzed the impact of the Mchinji Social Cast Transfer pilot programme on
agriculture production and health status of children age 0-5 We found the social cash transfer
programme of Malawi greatly affecting food consumption out of own production amongst
targeted families For example per capita home consumption of cereals and grains increased
by 564 MWK (pgt001) while production of pulses by 102 MWK (pgt001) Moreover child
linear growth increased by 0221 SD (corresponding approximately to 083 cm) while stunting
rate dropped by 621pp amongst children age 0-5 under the SCT programme If we exclude
the social pension scheme rolled out in South Africa the impact of the SCT on child health
26
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
status ranks across the highest in the CT literature By including in the child level analysis
observables counting for consumption out of own production disaggregated by food groups
and interacting them with the SCT treatment status we found that consumption of meat and
fish greatly influenced child height status and stunting in the sample of interest Children
living in families consuming meat and fish from own production experienced respectively an
improvement in height equal to 0519 SD (pgt001) while stunting decreased by 193 pp We
also found some positive impacts related to consumption of vegetables and dairy products and
eggs Not surprisingly food groups as cereals and grains fulfil the nutritional gap less than
other groups (eg meat and fish) confirming the idea that height status not only depend on the
quantity of energy intakes but rather on the quality and variety of food absorbed by young
children While our results point at large effect of the transfer on agricultural production and
child nutritional status given the small sample size and the fact the data were collected over a
pilot programme carried out only in the Mchinji district of Malawi we cannot expand our
findings to other Malawi regions in which the SCT programme has been currently taking
place and thus we do not believe our results characterized by external validity Nonetheless
our findings are interesting and highlight at the role of cash transfer as engine of agricultural
growth and interventions that complemented with more agricultural oriented policies can
lead to significant mitigation of poverty and economic growth in rural and remote regions of
the LDCs
By and large the transfer programme appear to be a success in mitigating one of the most
abject dimension of poverty child malnutrition while also inducing enhancements in
agricultural production Compared to other transfer programmes the SCT transfer size is large
standing at approximately 30 per cent of the per capita income of targeted families (Boone et
al Forthcoming Miller et al 2008b Osei et al 2012) Based on empirical evidence recent
reviews (Fizbein and Schady 2009 Manley 2012 and Leroy et al 2009) from LAC countries
27
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
show how the transfer size is one of the key factor associated with significant improvements
in child height outcomes and food nutrition as well as poverty reduction and therefore this
could explain the transfer was so successful While the STC needs further investigation to
cross check our findings with a larger evaluation sample we believe that this intervention was
successful in reaching the ldquohard-corerdquo poor a segment of the population which is usually
difficult to be targeted by anti-poverty programmes
References Aguumlero J M Carter MR et al (2007) The Impact of Unconditional Cash Transfers on Nutrition The South African Child Support Grant International Poverty Center Working Paper 30
Ahmed AU et al (2009) Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh International Food Policy Research Institute Research Monograph 163 Washington DC p 1-250
Alderman H Behrman J R Kohler H Maluccio JA Watkins S 2001 Attrition in Longitudinal Household Survey Data Demographic Research Max Planck Institute for Demographic Research Rostock Germany vol 5(4) pages 79-124 November
Alderman H SA Haddad L Song L and Yohannes Y ldquoReducing Child Malnutrition How Far Does Income Growth Take Usrdquo CREDIT Research Papers March 2001 0105
Alderman H JH and Kinsey B ldquoLong term consequences of early childhood malnutritionrdquo Oxf Econ Pap 2006 58(3) p 450-474
Attanasio O Battistin E Fitzsimmons E Mesnard A amp Vera-Hernaacutendez M (2005) How Effective are Conditional Cash Transfers Evidence from Colombia Briefing Note No 54 The Institute for Fiscal Studies Washington DC 10 p
Attanasio O C Meghir et al (2010) Mexicos Conditional Cash Transfer Program Lancet 375 980
Attanasio O L Goacutemez P Heredia amp Vera-Hernaacutendez M (2005) The Short Term Impact of a Conditional Cash Subsidy on Child Health and Nutrition in Colombia Centre for the Evaluation of Development Policies Report Summary Familias 03 The Institute for Fiscal Studies Washington DC 15pp
Baird S McIntosh C amp Oumlzler B (2009) Designing Cost-Effective Cash Transfer Programs to Boost Schooling among Young Women in Sub-Saharan Africa World Bank Policy Research Working Paper 5090 October 44pp
28
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Baird S McIntosh C amp Oumlzler B (2011)ldquoCash or Condition Evidence from a Cash Transfer Experimentrdquo Quarterly Journal of Economics Vol 126(4 November) forthcoming
Balk D A Storeygard et al (2005) Child Hunger in the Developing World An Analysis of Environmental and Social Correlates Food Policy 30 584-611
Barrientos A Delong J ldquoReducing Child Poverty with Cash Transfers A Sure Thingrdquo Development Policy Review 2006 24 (5) p 537-552
Barrientos A amp Nino-Zarazua M (2010) Effects of non-contributory social transfers in developing countries A compendium Geneva International Labor Office 40
Barrientos A Nintildeo-Zarazuacutea M amp Maitrot M (2010) Social Assistance in Developing Countries Database Chronic Poverty Research Centre Version 5 July 137 pp Downloaded June 23 2011 from httpwwwchronicpovertyorgpublicationsdetailssocial-assistance-indeveloping-
Barrios Federico Luis Galeano y Susana Saacutenchez (2008) El impacto del programa Tekoporacirc de Paraguay en la nutricioacuten el consumo y la economiacutea local documento presentado en el tercer seminario internacional Transferencias condicionadas erradicacioacuten del hambre y la desnutricioacuten en tiempos de crisis
Bassett L (2008) Can Conditional Cash Transfer Programs Play a Greater Role in Reducing Child Undernutrition World Bank SP Discussion Paper Series Washington DC World Bank 84
Baulch B (2010) The Medium-Term Impact of the Primary Education Stipend in Rural Bangladesh International Food Policy Research Institute Discussion Paper 00976 Washington DC
Behrman J and J Hoddinott (2005) Program Evaluation with Unobserved Heterogeneity and Selective Implementation The Mexican PROGRESA Impact on Child Nutrition Oxford Bulletin of Economics and Statistics 67(4) 547-569
Behrman J Fernald L Gertler P Neufeld LM amp Parker S (2008) Evaluacioacuten externa del Programa Oportunidades 2008 A diez antildeos intervencioacuten en zonas rurales (1997-2007) Tomo I Efectos de Oportunidades en aacutereas rurales a diez antildeos de intervencioacuten Capiacutetulo 1 Evaluacioacuten de los efectos a diez antildeos de Oportunidades en el desarrollo educacioacuten y nutricioacuten en nintildeos entre 7 y 10 antildeos
de familias incorporadas desde el inicio del Programa Secretary of Social Development Mexico City Mexico 2008
Black R E L H Allen et al (2008) Maternal and child under nutrition global and regional exposures and health consequences Lancet 371(9608) 243-260
Boersma B Wit J ldquoCatch-up Growthrdquo Endocrine Reviews 1997 18(5) p 646-661
29
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Bonvecchio A G H Pelto et al (2007) Maternal Knowledge and Use of a Micronutrient Supplement Was Improved with a Programmatically Feasible Intervention in Mexico The Journal of Nutrition137 440-446
Borenstein M Hedges L Higgins J amp Rothstein H (2009) Introduction to Meta analysis Hoboken NJ J Wiley
Bouillon C P amp Tejerina L (2007) Do We Know WHAT WORKS A Systematic Review of Impact Evaluations of Social Programs in Latin America and the Carribeans Caribbean Washington DC Inter-American Development Bank 144
Cameron N PM Cole TJ ldquoCatch-up Growth or Regression to the Mean Recovery from Stunting Revisitedrdquo American Journal of Human Biology 2005 17 p 412-417
Case A (2001) Does Money Protect Health Status Evidence from South African Pensions NBER Working Paper 8495 Cambridge MA p 1-32
Cebu Study Team (Akin J GD Popkin B) ldquoA child health production function estimated from longitudinal datardquo Journal of Development Economics 1992 38 p 323-351
Cognitive Development in Early Childhood Impact Evaluation Series No 25 Washington DC The World Bank Policy Research Working Paper 4759 1-50
Duflo E (2003) Grandmothers and Granddaughters Old-Age Pensions and Intra household Allocation in South Africa The World Bank Economic Review 17(1) p 1-25
Duncan GJ Morris PA amp Rodrigues C (2011) Does Money Really Matter Estimating Effects of Family Income on Young Childrens Achievement with Data from Random Assignment Experiments Developmental Psychology 47(5) 1263-1279
Fernald L Gertler P et al (2008) Role of cash in conditional cash transfer programs for child health growth and development an analysis of Mexicorsquos Oportunidades The Lancet 371 828-837
Fernald L Gertler P et al (2009) 10-year effect of Oportunidades Mexicorsquos conditional cash transfer program on child growth cognition language and behaviour a longitudinal follow-up study The Lancet 374(9706)
Fiszbein A and N Schady (2009) Conditional cash transfers reducing present and future poverty Washington DC The World Bank 361
Fryer R G J (2010) Financial Incentives and Student Achievement Evidence from Randomized Trials Working Paper NBER 91
Gaarder M Glassman A et al (2010) Conditional cash transfers and health unpacking the causal chain Journal of Development Effectiveness 2(1) 6ndash50
Gertler P (2004) Do conditional cash transfers improve child health Evidence from PROGRESAs controlled randomized experiment American Economic Review 94(2) 336-341
30
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Gitter S J Manley amp Barham B (2010) The Coffee Crisis Early Childhood Development and Conditional Cash Transfers Working Papers Washington DC Inter-American Development Bank 37
Glassman A Gaarder M et al (2006) Demand-side incentives for better health for the poor Conditional cash transfer programs in Latin America and the Caribbean Washington DC Economic and Sector Study Series IADB
Glassman A J Todd et al (2007) Performance-Based Incentives for Health Conditional Cash Transfer Programs in Latin America and the Caribbean CGD Working Paper 120 Washington DC Center for Global Development 1-60
Goncalves-Silva R M Valente J G et al (2005) Household Smoking and Stunting for Children Under Five Years Cadernos de Sauacutede Puacuteblica 21(5) 1540-1549
Grantham-McGregor S Cheung Y B et al (2007) Development Potential in the First 5 Years for Children in Developing Countries Lancet 369 60-70
Grillenberger M Neumann CG Murphy SP (2006) ldquoIntake of Micronutrients High in Animal-Source Foods is Associated with Better Growth in Rural Kenyan Childrenrdquo British Journal of Nutrition 95 379-390
Hand S and Peterman A ldquoIs there Catch-Up Growth Evidence from Three Continentsrdquo Journal of Development Economics 2012 forthcoming
Hanlon J A Barrientos et al (2010) Just give money to the poor the development revolution from the global south Sterling Va Kumarian Press
Higgins JPT Thompson SG Deeks JJ amp Altman DG (2003) Measuring inconsistency in meta-analyses British Medical Journal 327 557ndash560
Himaz R (2008) Welfare Grants and Their Impact on Child Health The Case of Sri Lanka World Development 36(10) p 1843-1857
Hirano K Imbens GW ldquoEstimation of Causal Effects using Propensity Score Weighting An Application to Data on Right Heart Catheterizationrdquo Health Services amp Outcomes Research Methodology 2002 2 p 259-278
Hoddinott J amp Kinsey B (2001) Child Growth in Time of Drought Oxford Bulletin of Economics and Statistics 63(4) 409-436
Hoddinott J amp L Bassett (2008) Conditional Cash Transfer Programs and Nutrition in Latin America Assessment of Impacts and Strategies for Improvement International Food Policy Research Institute Washington DC p 1-34
Hoddinott J amp Skoufias E (2004) The Impact of PROGRESA on Food Consumption Economic development and cultural change53 37-61
31
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Hoddinott J (2010) Nutrition and Conditional Cash Transfer Programs Conditional Cash Transfers in Latin America A Magic Bullet to Reduce Poverty M Adato and J Hoddinott Baltimore Johns Hopkins University Press
Hoddinott J Skoufias E et al (2000) The impact of PROGRESA on consumption a final report Washington DC International Food Policy Research Institute 1-86
in Tackling Child Mortality Save the Children UK 66
Janvry A d F Finan et al (2006) Can Conditional Cash Transfers Serve as Safety Nets to Keep Children at School and out of the Labor Market Journal of Development Economics 79(2) 349-373
Lagarde M A Haines et al (2007) Conditional Cash Transfers for Improving Uptake of Health Interventions in Low- and Middle-Income Countries A Systematic Review Journal of American Medical Association 298(16) 1900- 1910
Leoacuten M amp Younger S (2007) Transfer Payments Mothersrsquo Income and Child Health in Ecuador Journal of Development Studies 43(6) p 1126ndash1143
Leroy J L A Garcia-Guerra et al (2008) The Oportunidades program increases the linear growth of children enrolled at young ages in urban Mexico Journal of Nutrition138(4) 793-798
Leroy J Ruel M amp Verhofstadt E 2009 The impact of conditional cash transfer programmes on child nutrition A review of evidence using a programme theory framework Journal of Development Effectiveness 1(2) 103ndash129
Linnemayr S and Alderman H ldquoAlmost random Evaluating a large-scale randomized nutrition program in the presence of crossoverrdquo Journal of Development Economics 2011 96(1) p 106-114
Macours K Schady N amp Vakis R (2008) Can Conditional Cash Transfer Programs Compensate for Delays in Early Childhood Development Johns Hopkins Univ amp the World Bank Working Paper 46 p
Maitra P A Rammohan et al (2010) Food Consumption Patterns and Malnourished Indian Children Is there a Link Department of Economics Discussion Paper Monash University 40
Maluccio J amp Flores R (2005) ldquoImpact Evaluation of a Conditional Cash Transfer ProgramThe Nicaraguan Red de Proteccioacuten Socialrdquo Washington DC International Food Policy Research Institute Research Report No141 1-78
Maluccio J (2005) ldquoCoping with the ldquoCoffee Crisisrdquo in Central America The Role of the Nicaraguan Red de Proteccioacuten Socialrdquo FCND Discussion Paper No 188 IFPRI Washington DC
Maluccio J (2009) Household Targeting in Practice The Nicaraguan Red de Proteccioacuten Social Journal of International Development 21(1) 1-23
32
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Mann C (1990) Meta-Analysis in the Breech Science 249(4968) 476-480
Miller C Tsoka M amp Reichert K (2009) The Malawi Social Cash Transfer and the impact of $14 per month on child health and growth Health Policy and Planning
Moher D Liberati A Tetzlaff J amp Altman DG (2009) Preferred Reporting Items for Systematic Reviews and Meta-Analyses The PRISMA Statement PLoS Medicine 6(7
Moore C (2008) (2008) Assessing Hondurasrsquo CCT programme PRAF Programa de Asignacioacuten Familiar Expected and unexpected realities Country Study Nordm 15 Brasilia Centro Internacional de Poliacuteticas para el Crecimiento InclusivoPrograma de las Naciones Unidas para el Desarrollo (PNUD) Abril
Morris S et al (2004) Conditional Cash Transfers Are Associated with a Small Reduction in the Rate of Weight Gain of Preschool Children in Northeast Brazil Journal of Nutrition 2004 134 p 2336ndash2341
Outes I Porter C ldquoCatching up from early nutritional deficits Evidence from rural Ethiopiardquo Economics amp Human Biology 2012
Paxson C amp Schady N (2007) Does Money Matter The Effects of Cash Transfers on Child Health and Development in Rural Ecuador World Bank Policy Research Working Paper 4226 Washington DC
Perova E amp Vakis R (2009) Welfare impacts of the JUNTOS Programa in Peru Evidence from a non-experimental evaluation World Bank Working Paper
Prader A TJ von Harnack G ldquoCatch-up growth following illness or Starvation An example of developmental canalizationrdquo in man Journal of Pediatrics 1963 62 p 646-659
Rivera J A C Hotz et al (2003) The Effect of Micronutrient Deficiencies on Child Growth A Review of Results from Community-Based Supplementation Trials Journal of Nutrition 133(11 Supplement 2) 4010Sndash4020S
Rivera J et al (2004) Impact of the Mexican Program for Education Health and Nutrition (PROGRESA) on Rates of Growth and Anemia in Infants and Young Children A Randomized Effectiveness Study Journal of the American Medical Association 291(21) p 2563-2570
Rosenbaum DB RubinPR ldquoThe central role of the propensity score in observational studies Biometrikardquo 1983 701 p 41-55
Rosenzweig MRS T Paul ldquoEstimating a Household Production Function Heterogeneity the Demand for Health Inputs and Their Effects on Birth Weightrdquo Journal of Political Economy University of Chicago Press 1983 Vol 91(5) p 723-746
Schultz T P (2004) School subsidies for the poor evaluating the Mexican PROGRESA poverty program Journal of Development Economics 74(1) 52
33
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Shekar M R Heaver et al (2006) Repositioning nutrition as central to development a strategy for large scale action Washington DC World Bank
Sinha N amp Yoong J (2009) Long-Term Financial Incentives and Investment in Daughters Evidence from Conditional Cash Transfers in North India The World Bank Policy Research Working Paper 4860 Washington DC
Skoufias E amp McClafferty B (2001) Is PROGRESA Working A Summary of the Results of an Evaluation by IFPRI Food Consumption and Nutrition Division Discussion Paper IFPRI
Strauss J amp Thomas D (1998) Health Nutrition and Economic Development Journal of Economic Literature 36(2) 766-817 UNICEF (United Nations International Childrens Emergency Fund) (2010) Narrowing the Gaps to Meet the Goals Downloaded from httpwwwuniceforgnutritionindex_55927html on August 21 2012
Strauss JT DT ldquoHealth Over the Life Courserdquo Ch 54 in TP Schultz amp John Strauss (Ed) Handbook of Development Economics 2007 (Vol 4) Amsterdam North-Holland
Todd J E Winters P Hertz T ldquoConditional Cash Transfers and Agricultural Production Lessons from the Oportunidades Experience in Mexicordquo The Journal of Development Studies Taylor and Francis Journals 2010 vol 46(1) p 39-67
Vera-Hernaacutendez M Attanasio O Goacutemez L Heredia P Romero J (2010) Transferencias monetarias condicionadas y nutricioacuten infantil Colombia Departamento Nacional de Planeacioacuten
Victora C G L Adair et al (2008) Maternal and child undernutrition consequences for adult health and human capital Lancet 371(9609) 340-357
Waddington H B Snilstveit et al (2009) Water Sanitation and Hygiene Interventions to Combat Childhood Diarrhoea in Developing Countries 3ie Synthetic Review August 119 pages
Waterlow J C R Buzina et al (1977) The Presentation and Use of Height and Weight Data for Comparing the Nutritional Status of Groups of Children Under the Age of 10 Years Bulletin of the World Health Organization 55(4) 489-498
WHO (World Health Organization) (2006) Infant and Young Child Nutrition The WHO Multicentre Growth Reference Study Quote is online at httpwwwwhointchildgrowthfaqsapplicableenindexhtml accessed April 20 2007
WHO (World Health Organization) (1995) Physical Status The Use and Interpretation of Anthropometry WHO Technical Report Series No 854 Geneva
Younger S Ponce J amp Hidalgo D (2008) El Impacto de Programas de Transferencias a las Madres de Familia en la Seguridad Alimentaria de los Nintildeos Un anaacutelisis comparado de los casos de Meacutexico y Ecuador Unpublished mimeo
34
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Figure1 Nutritional status of children age 0-5 East African countries
30
35
40
45
50
55
60
Stun
ting
2000 2005 2010 Year
Malawi Burundi Eritrea Ethiopia Kenya Madagascar Mozambique Rwanda Tanzania Uganda Zambia Zimbabwe
Source Development and Health Survey (DHS)
Table 1 Nutritional status of children age 0-5 Malawi
Total Urban Rural Poore s t quintile
2000 546 400 569 NA 2004 525 423 539 539 2010 471 407 482 555
Stunting
Total
244 222 196
Urban Rural
135 261 158 231 203 155
Se ve re s tunting
Poore s t quintile
NA 292 241
Source Development and Health Survey (DHS)
35
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
1
1
13
DRAFT NOT FOR CITATION
Figure2 Estimated propensity score by sample Panel A Propensity score BEFORE using IPW Panel D Propensity score AFTER using IPW all households all households
Treament Control Treament Control 25 25
2 2
15
Densi
ty
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel B Propensity score BEFORE using IPW Panel E Propensity score AFTER using IPW SHH sample SHH sample
Treament Control Treament Control 25 25
2 2
15
Dens
ity
Dens
ity 15
1
5 5
0 02 4 6 8 1
Propensity score 2 4 6 8 1 Propensity score
Panel C Propensity score BEFORE using IPW Panel F Propensity score AFTER using IPW CHH sample (all children) CHH sample (all children)
treatment control treatment control
kden
sity ps
core
0 5
1 15
2
kden
sity p
score
0 5
1 15
2
0 2 4 6 8 1 propensity score 0 2 4 6
propensity score 8 1
Note Propensity score computed using robust probit regressions
36
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Figure 3 Trends in HAZ13 z-shy‐score13 by age in months baseline data-2
2
-2
-18
-1
6
-14
HA
Z z-
scor
e
0 20 40 60 Age in months at baseline
Note Locally13 weighted13 regression with13 bandwidth13 equal to13 08
37
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
DRAFT NOT FOR CITATION Table 2 Household characteristics by control and treatment group
$amp() +amp -++amp) ) +amp -++amp) )
0amp$1$$amp$2)3$$4 ) $amp(amp)amp$+) - 01 - -2 - 32 - 4 - 2 - 32
6789amp7lt9$9+$(ltamp - 420 - 405= = = - 342= = = - 345 - 33 - 342 )amp - -4 - 04 - 320 - 35gt - - 320
A)(ltBamp$C)7)BDampB+ampEamplt+7D$amp 50 343 54 4-3 5 4 23 515 12 44 5 4 Fampamplt+ampltC$)7Gamp$4 - 10 - 104 - 12 - 1 - 14 - 12
D(amp$$)Blt - - 424= = = - 105= = = - gt5 - 22 - 105 D(amp$9CB7+$amplt 0 3gt4 515= = = 0 51= = = 0 51gt 0 512 0 51
5+6(7+amp8) 78) 373$$3(4 ) Hamp()amp - gt - gtgt2 - gt4 - gt05 - gt01 - gt4 Iamp gt gt4 -= = 5 5gt= = 5 313 2 51gt 5 5gt
Jamp)$9amp+DC)7lt 21 0-= = = 550= = = 0 -43 0 -11 550 F7)amp - 0 - 0 - 0- - 0-gt - 0-1 - 0-
Hamp()amp)lt++7)amp - 05 - 00 - 41 - 41 - 42 - 41 K+amp$ - 51 - gt-= = = - 3-= = = - 00 - - - 3 L7ltamp - 10 - 14 - 1 - 1 - gt5 - 1
L7ltamp)lt+9amp()amp - gt0gt - gt3 - gt- - 55 - 54 - gt-
5+6(7+amp8) 3+9+($$+4 ) Mamp(amp$)ampN - 30 - 41gt= - 311= - -4 - 4 - 311 Mamp(amp$)ampN - - 2 - gt4-= = = - 52= = = - 543 - 22gt - 52 Mamp(amp$)ampN - 5 - 15= = = 4= = = 12 0-4 4 Mamp(amp$)ampN 5 4gt - 544= = = 4= = = 4gtgt 424 4 Mamp(amp$Gamp$gt- - 115 - 23= = = - 1-= = = - gt11 - gt34 - 1 ODampB+7Pamp 3 -5 4 3= = = 3 gt1= = = 3 gt2 3 gt3 3 gt1
Q9BDampB+7Pamp 0 -10= = = 41= = = 43 40 41
+) 373$$3(4 Iamp)ltampBDampB+(amp(amp$C)ltR$8 - 355 - 31= = = - 335= = = - 334 - 341 - 335
D(amp$9+)amplt7lt)ltD)$ 1 3 gt 53 2 0 2 014 2 302 2 0 ST lt7ltDampltampE)ampU
amp) A 4 ) $178) ) (gt) ampamp) 7+6(7+amp8()
amp) B 4 ) =$178) (gt ) ampamp) 7+6(7+amp8() )
+amp -++amp) )
- 351 - 4 - 32 - 4-1 - 01gt - 442 - 3 - gtgt= = = - 442= = = 00 2 4 301 4- 25 - gt54 - gt51 - gt22 - - 003= = = - 150= = = 3 414 4 531= = = 3 150= = =
- gt3 - gt1 - gt4gt 32 3gt3 32 32 500 5-2 gt- 0 0-2 - - 3 - -12 - -5 - -15 - -45 - 032 - 041 - 0gt - gt3 - gt2 - gt04 - gt-2 - gt40 - 23
30 4 1 43 40= = = 353= = = 5 -45= 442= gt- 4 gt22 - 43gt - 4gt2 - 40 554 gtgt= = = gt 3gt= = = 135 gtgt5= = = 202= = =
- 0-4 - 0 - 5 5 13 2 gt2 - 21
amp) -4 ) $1778) (gt) 37$amp8) 7+6(7+amp8()
+amp -++amp) )
- 44 - 12 - 32 - 42 - 41 - 442 - 30 - 35= = = - 442= = = 4 0gt5 52 04 4- 25 - 14 - 1gtgt - gt22 - 30 - gt0= = = - 150= = = 3 51gt 44 3 150
- 2 - 3 - gt4gt 31 22 3gt 33gt 32 500 0 0 0 00 0 0-2 - -55 - 2 - -12 - -1 - -14 - -45 - 5gt - 30= - 0gt-= - 0 - 35 - gt04 - 2 - 3gt - 23
21 gt 1 32 44 353 gtgt1= 442= 253 0 -1-= = gt22= = - 011 - 04gt - 40 gt gt5 gt 501 gt 3gt 2gt 22 202
- gt4 - 41 - 5 - 410 5 5gt - 21
amp) lt 4 ) =$178) (gt ) 37$amp8) 7+6(7+amp8() )
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Table 2 Household characteristics by control and treatment group (continued) Shocks affecting the household
Shock risk 0 264 0 220 0 310 0 304 0 298 0 31 Natural 0 609 0 573 0 647 0 638 0 629 0 647 Prices 0 559 0 536 0 584 0 598 0 613 0 584
Financial 0 606 0 585 0 627 0 64 0 653 0 627
Access to other transfer programs before2007
Food subsidy 0 268 0 311 0 222 0 219 0 217 0 222 Input subsidy 0 423 0 378 0 471 0 447 0 423 0 471
Observations 751 386 365 751 386 365
0 229 0 614 0 588 0 647
0 184 0 500 0 592 0 645
0 273 0 727 0 584 0 649
0 283 0 676 0 646 0 697
0 292 0 631 0 699 0 738
0 273 0 727 0 584 0 649
0 15 0 503
153
0 158 0 421 76
0 143 0 584 77
0 13 0 504
153
0 12 0 436 76
0 143 0 584 77
0514 3675 0163 0601 0312 0087
0549 36775 0167 0637 0284 0078
0481 36726 016 0566 034 0094
36807 0511 0158 0588 0324 0088
36903 0546 0156 0613 0306 008
36726 0481 016 0566 034 0094
208 102 106 208 102 106
Demographics of childrenage 0-shy‐5 Girls NA NA NA NA NA NA
Age inmonths NA NA NA NA NA NA Maternal orphan NA NA NA NA NA NA Birthorder first NA NA NA NA NA NA
Birthorder second NA NA NA NA NA NA Birthorder third NA NA NA NA NA NA
Observations NA NA NA NA NA NA
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
Table 3 Outcome indicators child level
Variables Total Control Treatment
HAZ indicator andstunting levels at baselineHZ A score -shy‐1 897 -shy‐1 931 -shy‐1 864
M oderate stunting (HA Z lt -shy‐1) 0 774 0 784 0 764 S tunting (HA Z -shy‐ 2 ) 0 452 0 451 0 453
S evere S tunting (HA Z -shy‐ 3) 0 25 0 245 0 255
Observations 208 102 106
Panel A Unweightehdmeans childTotal Control Treatment
-shy‐1 85 -shy‐1 834 -shy‐1 864 0 771 0 779 0 764 0 435 0 414 0 453 0 242 0 227 0 255
208 102 106
Panel B Weighetedmeans child
Note Mean differences are statistically significant at 10 significant at 5 significant at 1
39
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Table 3 Outcome indicators Monthly per capita value of food consumed out of own production Full sample
All foods Ce re als and grains
Roots and Tube rs
Puls e s Vegetables Fruits
Control 170447 102844 3334 8576 50044 5366 Treatment 295967 146958 9159 38461 74319 14294
Total 231452 124284 6165 23101 61842 9705
SHH sample
All foods Ce re als and
grains Roots and
Tube rs Puls e s Vegetables Fruits
Control 111468 66484 4441 5661 31429 3159 Treatment 228566 114921 6918 29832 49566 14988
Total 171143 91168 5703 17979 40672 9187
Meat and fis h
0216 11003
5459
Meat and fis h
0163 10915
5643
Eggs and dairy
products
0068 1773
0897
Eggs and dairy
products
0131 1425
0791 Note Author analysis of STC data
40
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Table 4 FD impact estimates of the SCT programme on food consumption out of own production full sample and CHH sample
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Variables
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 1444298 145228 56399 57015 5845 6210 31547 30040 29149 30494 10987 9424 8862 10327 1640 1719
(10783) (14031) (10807) (9652) (1557) (1509) (3163) (2944) (5490) (4894) (2193) (1855) (1699) (1640) (0332) (0311)
Constant 258153 277119 158962 173954 9439 8069 12091 12605 70973 72517 6271 9533 -0236 -0432 0653 0873 (58951) (61 812) (38225) (40047) (8754) (8 052) (11419) (12346) (17973) (18098) (5393) (5734) (10698) (11324) (1251) (1354)
Observations 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751 751
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16)
Varia bles
Food home produce d
Food home produced -
IPW Ce re a ls Ce re a ls-
IPW Roots and
tubers Roots and tubers-IPW Pulses Pulses-
IPW Vege table s Ve ge table s-IPW Fruits Fruits-IPW Me a t and
fish Me a t a nd fish-IPW
Da iry products
Da iry products-
IPW
Treatment 112258 1160615 45553 42225 4955 3946 22494 26875 16372 17522 11334 12633 10895 12492 1438 1758 4521 479 (15821) (15492) (2238) (2048) (4561) (4793) (6290) (6316) (3065) (3361) (3177) (2912) (0562) (0607)
Constant 30051 154951 209977 338079 -11636 -16094 -24730 -21172 54543 65594 -15192 -15794 24333 39946 1156 -0656 (0180) (0710) (130668) (155757) (11677) (14043) (24934) (40248) (60139) (93918) (17999) (24802) (21203) (38423) (2556) (4360)
Observations 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208 208
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
Table 5 Impact estimates of the SCT programme on child height status O utcomevariable HAZ Moderate stunting S tunting
DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW DD DD13 controls13
DD-shy‐IPW
Treatment 0 137 0 189 0 221 -shy‐0 047 -shy‐0 087 -shy‐0 086 -shy‐0 077 -shy‐0 140 -shy‐0 137 -shy‐0 064 -shy‐0 041 -shy‐0 063
(0 108) (0 112) (0 112) (0 037) (0 040) (0 035) (0 057) (0 059) (0 061) (0 050) (0 055) (0 052)
O bservations13 208 208 208 208 208 208 208 208 208 208 208 208
S evere S tunting
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
41
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Table 6 DD impact estimates of the SCT on child height controlling for follow-up food consumption out of own production (1) (2) (3) (4) (5) (6) (7) (8)
Variables HAZ HAZ-IPW
M ode rate stunting
M ode rate stunting -
IPW Stunting Stunting -
IPW Severe stunting
Se ve re stunting -
IPW
Treatment 0159 0226 -0108 -0120 -0125 -0136 -0024 -0069 (0131) (0138) (0042) (0043) (0074) (0081) (0059) (0057)
Cereals and grains 0000 0000 0000 0000 -0000 -0000 -0000 -0000 (0001) (0001) (0000) (0000) (0000) (0000) (0000) (0000)
Roots and Tubers 0002 0000 0001 0001 -0000 0003 -0000 0002 (0004) (0004) (0001) (0001) (0003) (0003) (0002) (0002)
Pulsest -0005 -0004 0001 0001 -0000 -0000 0002 0001 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Vegetables 0002 0001 -0001 -0000 0000 0000 0000 0001 (0002) (0002) (0001) (0001) (0001) (0001) (0001) (0001)
Fruits 0001 0002 0001 0000 -0000 0001 -0000 0000 (0003) (0003) (0001) (0001) (0002) (0002) (0001) (0001)
Meat and fish 0008 0007 -0002 -0002 0001 0002 -0002 -0001 (0004) (0004) (0002) (0002) (0002) (0002) (0002) (0002)
Dairy -0015 -0011 0009 0010 -0013 -0020 -0012 -0013 (0016) (0016) (0012) (0012) (0009) (0011) (0010) (0008)
Observations 208 208 208 208 208 208 208 208 R-squared 0268 0328 0264 0330 0233 0282 0246 0299
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
42
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Table 7 Treatment variable interacted with monthly per capita food consumption (MWK) out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0143 0200 -0105 -0110 -0121 -0128 -0026 -0080 (0125) (0129) (0043) (0045) (0072) (0077) (0059) (0055)
Roots and Tubers Treatment 0167 0226 -0169 -0130 -0133 -0136 -0032 -0069 (0137) (0140) (0090) (0086) (0075) (0081) (0058) (0056)
Pulses Treatment 0015 0031 -0052 -0096 -0069 -0085 0019 -0049 (0137) (0152) (0053) (0063) (0074) (0077) (0066) (0071)
Vegetables Treatment 0162 0232 -0104 -0116 -0118 -0127 -0027 -0077 (0130) (0137) (0042) (0044) (0072) (0077) (0059) (0055)
Fruits Treatment 0089 0172 -0051 -0096 -0062 -0082 -0000 -0080 (0131) (0130) (0070) (0069) (0077) (0084) (0064) (0061)
Meat and fish Treatment 0519 0562 0113 0088 -0374 -0427 -0125 -0158 (0187) (0235) (0206) (0218) (0111) (0131) (0094) (0103)
Dairy Treatment 0115 0211 -0084 0236 -0120 -0146 0017 -0044 (0124) (0131) (0040) (0250) (0077) (0081) (0058) (0056)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
43
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
APPENDIX
Child level analysis Treatment variable interacted with shares of food consumption out of home production (1) (2) (3) (4) (5) (6) (7) (8)
Interaction terms HAZ HAZ-IPW Moderate stunting
M oderate stunting -
IPW Stunting
Stunting -IPW
Severe stunting
Severe stunting -
IPW
Cereals and grains Treatment 0182 0202 -0097 -0098 -0157 -0162 -0057 -0077 (0116) (0116) (0040) (0039) (0065) (0069) (0055) (0053)
Roots and Tubers Treatment 0177 0192 -0086 -0093 -0171 -0171 -0060 -0076 (0122) (0119) (0040) (0039) (0066) (0069) (0057) (0052)
Pulses Treatment 0177 0187 -0096 -0094 -0158 -0163 -0059 -0079 (0115) (0116) (0040) (0038) (0065) (0068) (0056) (0053)
Vegetables Treatment 0182 0190 -0097 -0094 -0158 -0165 -0059 -0079 (0114) (0114) (0039) (0037) (0065) (0070) (0055) (0053)
Fruits Treatment 0163 0175 -0089 -0084 -0148 -0138 -0057 -0079 (0118) (0114) (0040) (0037) (0064) (0066) (0055) (0052)
Meat and fish Treatment 0237 0260 -0082 -0083 -0215 -0231 -0076 -0100 (0120) (0128) (0040) (0036) (0068) (0074) (0056) (0055)
Dairy Treatment 0173 0193 -0093 -0093 -0158 -0167 -0059 -0079 (0116) (0116) (0039) (0038) (0065) (0069) (0056) (0053)
Note Observations clustered at household level using robust variance-covariance matrix Standard error in parentheses Coefficient are statistically significant at 10 significant at 5 significant at 1
44
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
DRAFT NOT FOR CITATION
Appendix child level attrition analysis $amp ( amp) +-amp$00amp1-+amp23$4amp$5$amp46amp0$amp-$amp 7$amp ) ( amp) $$amp234$+amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt ) =amp02-+amp $amp ( amp) gt +amp-1amp-0+5-0amp $ amp $
7$amp ( amp8 3$4+amp$5Aamp amp+4amp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ $+( )) ( gt +amp-1amp-0+5-0amp + (
7$amp8 ( amp8 3$4+amp$5Aamp amp2-+-$ampamp3-03-$40amp
8 3$4+amp +4ampamp 0$amp
8 3$4+amp 4+-994ampamp 0$amp
-$amp
lt)=amp02-+ ) amp )+ gt +amp-1amp-0+5-0amp amp
0amp 5$amp
0amp 5$amp
(
0amp 5$amp
+amp
Note Values in bold are statistically significant at least at 10 level
$amp(amp) +-amp$00amp1-+amp23$4amp$5$amp46amp1-$$-789amp-$amp $amp) (amp) $$amp234$+amp
lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+amp $amp $() 9+amp-1amp-0+5-0amp + ) $(
$ampA (amplt 3$4+amp$5Bampamp+4amp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ amp (amp 9+amp-1amp-0+5-0amp (
$amplt (amplt 3$4+amp$5Bampamp2-+-$ampamp3-903-$40amp lt 3$4+amp +4ampamp 0$amp
lt 3$4+amp 4+-4ampamp 0$amp
-$amp
=) gtamp02-+ +$( )amp amp 9+amp-1amp-0+5-0amp
80amp 5$9amp
)+(
80amp 5$9amp
$+
80amp 5$9amp
+
Note Values in bold are statistically significant at least at 10 level
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46
13
DRAFT NOT FOR CITATION
Variable s All
houshe olds
1 or less meals per day 0161 (0099)
Risk of begging for food or money 0344 (0174)
No assets -0056 (0107)
Total per capita expenditure 0000 (0000)
Dependency ratio over 3 -0006 (0165)
Number orphans 0067 (0045)
Number of children 0075 (0134)
HH Female -0083 (0359)
HH Age 0008 (0006)
HH Years of education 0063 (0023)
HH Disable -0264 (0202)
HH Female and disable 0532 (0247)
HH Elder -0185 (0231)
HH Single 0268 (0218)
HH Single and female -0227 (0404)
Members age 1-5 0053 (0158)
Members age 5-10 0284 (0163)
Members age 15-59 0246 (0164)
Members over 60 0295 (0108)
Household size 0130 (0175)
Log of the househld size -0526 (0260)
No household members can work 0070 (0173)
Participation in ganyu labor -0006 (0005)
Shock risk -0041 (0212)
Natural 0071 (0127)
Prices 0091 (0232)
Financial -0155 (0234)
Food subsidy -0131 (0114)
Input subsidy 0169 (0105)
Girl -1276 (0396)
Age in months
Maternal orphan
Birth order first
Birth order second
Birth order third
Constant -1276 (0396)
Observations 751
Child le ve l
0296 (0226) -0107 (0316)
-0879 (0240) 0000
(0001) -0151 (0335) 0249 (0099) 0304
(0272) -0063 (0126) 0667
(0516) 0018
(0012) 0052
(0036) -0035 (0443) 0475
(0594) 0576
(0526) -0992 (0707) 0164
(0211) 0243
(0154) -0013 (0194) -0736 (0394) -0473 (0430) 2266
(2147) 0117
(0395) 0002
(0009) 0518
(0469) -0321 (0263) -0344 (0480) -0024 (0478) -0396 (0303) 0352
(0217) -0055 (0209) 0012
(0009) -0293 (0358) 0383
(0482) 0166
(0388)
-4358 (2290)
208
Note significant at 10 significant at 5 significant at 1
46