Rwanda Green Leaf Price Reform Impact Evaluation Baseline...
Transcript of Rwanda Green Leaf Price Reform Impact Evaluation Baseline...
Rwanda Green Leaf Price Reform Impact Evaluation
Baseline Report
March 2015
Investment Climate Impact Program
Contents
1. Overview ................................................................................................................................................. 1
1.1 Introduction ............................................................................................................................ 1
1.2 Tea in Rwanda ......................................................................................................................... 1
1.2.1 Use of evidence base by Rwanda authorities .................................................................... 3
1.3 World Bank Group Support of the Tea Sector Reforms .......................................................... 4
1.4 Objectives................................................................................................................................ 5
1.4.1 Objectives of the impact evaluation ........................................................................... 5
1.4.2 Policy perspective ....................................................................................................... 6
2. Methodology ........................................................................................................................................... 7
2.1 Randomization and Sample Design ........................................................................................ 7
2.2 Sample Size and Strategy ........................................................................................................ 8
2.3 Instruments for Data Collection and Data Quality Assurance ................................................ 8
2.4 Data Storage, Management, and Access Policy ...................................................................... 9
3. Sample Representativeness and External Validity of the Survey ........................................................... 9
3.1 Comparison of the 2013 and 2004 Tea Sector Surveys .................................................... 10
4. Findings of the Tea Sector Baseline Survey .......................................................................................... 13
4.1 Demographic Profiles of Tea and Non-tea Farmer from 2013 Baseline Survey ............... 13
4.2 Land Holdings and Size ...................................................................................................... 15
4.3 Livestock Ownership, Crop Activity, and Labor Usage of Farmers .................................... 16
4.4 The Pattern of Baseline Tea Prices .................................................................................... 19
4.5 The Pattern of Baseline Tea Production ............................................................................ 20
4.6 Conclusion ......................................................................................................................... 21
5. Internal Validity of the Study ................................................................................................................ 21
5.1 Intra-correlation Coefficients and Design Effects ............................................................. 22
5.2 The Measurement of Land Size ......................................................................................... 23
5.3 Respondent Responses for the Non-heads of Household ................................................ 24
6. Other Descriptive Findings from Baseline Survey: Requested Sector Evidence by GoR ...................... 24
6.1 Household Income Distributions ..................................................................................... 24
6.2 Household Income Determination of Rwandan Farmers in the Tea-growing Areas ....... 29
6.3 Tea Reforms and the Non-tea Producers ........................................................................ 33
6.4 Subjective Measures of Household Welfare for Rwandan Farmers ................................ 37
6.5 Some Concluding Remarks .............................................................................................. 40
7. Recommendations and Suggestions for Follow-up Survey .................................................................. 41
Annex 1: Tea Growing Districts ........................................................................................................ 43
Investment Climate Impact Program
Annex 2: Tea Growing Sectors Covered by the Survey .................................................................... 44
Annex 3: Mean Differences between Co-op and non-Co-op Thé Villageois Farmers...................... 45
Annex 4: Household Income Determinants for Female Farmers .................................................... 46
Annex 5: Household Income Determinants for Farmers with Health Disability .............................. 48
Annex 6: Probit Models Explaining “Yes” Outcomes in Table 6.8 ................................................... 50
List of Figures:
Figure 1 Tea Production in Rwanda (1961–2012) …………………………………………………………………………….. 1
Figure 2 Tea Area Harvested (1961-2012) .............................................................................................. 1
Figure 3 Tea Yield in Rwanda vs. in Kenya and Uganda .......................................................................... 3
Figure 4 Estimated Revenues for Smallholder Crops on 0.25 ha (RWF ‘000s, 2011) ............................. 4
Figure 5 Theory of Change for Rwanda’s Tea Sector Price Reform ........................................................ 5
List of Tables:
Table 3.1: Main Economic Activity of Head of Household by Tea Sector Survey ................................. 10
Table 3.2: Demographic Profiles of Head of Household by Tea Sector Survey .................................... 10
Table 3.3: Total Land Size Holdings of Farmers (in Hectares) ............................................................... 12
Table 3.4: Household Incomes for Rwanda Farmers (RWF at 2010 Prices) .......................................... 12
Table 4.1: Summary of Economic Activity of the Head of Household .................................................. 13
Table 4.2: Summary of Main Occupational Activity of the Head of Household ................................... 14
Table 4.3: Demographic Profile of Tea and Non-tea Farmers (Main Occupation) ............................... 14
Table 4.4: Land Size Profile for Tea and Non-tea Farmers .................................................................... 16
Table 4.5: Livestock Ownership of Farmers 12 Months Prior to Interview .......................................... 17
Table 4.6: Crop Activity of Farmers 12 Months Prior to Interview ....................................................... 18
Table 4.7: Labor Usage by Farmers ....................................................................................................... 19
Table 4.8: Average Baseline Price per Kilogram for Green Leaf Tea ..................................................... 19
Table 4.9: Baseline Price per Kilogram by Tea Holding Type ................................................................ 20
Table 4.10: Baseline Tea Production Yield ............................................................................................ 21
Table 5.1: Intra-cluster Correlations for Selected Variables for the 2013 Tea Survey ......................... 22
Table 5.2: Propensity Score Matching for Main Plot Size Measured in Log Hectares .......................... 23
Table 6.1: Household Income and Wealth Profile of Tea and Non-tea Farmers .................................. 25
Table 6.2: Real Household Incomes in the Tea Growing Areas of Rwanda 2004 and 2013 ................. 29
Table 6.3: Determinants of Log Annual Household Income for Tea and Non-tea Farmers ................. 31
Table 6.4: Quantile Regression Estimates for Tea Farmer Effect ......................................................... 32
Table 6.5: Attitudes & Perceptions of Tea Reforms among Non-tea Farmers ..................................... 34
Table 6.6: Minimum Green Leaf Price per Kilogram Required to Encourage Non-tea Farmers to
Cultivate Tea ......................................................................................................................................... 34
Table 6.7: Probit Model for Non-tea Farmers’ Reluctance to Cultivate Tea ........................................ 34
Investment Climate Impact Program
Table 6.8: Factors Preventing Engagement of Non-tea Farmers in Tea Cultivation ............................. 36
Table 6.9: Subjective Perceptions of Farmer Well-being ...................................................................... 37
Table 6.10: Ordered Probit Model of Farmer Satisfaction with Living Standards ................................ 38
Investment Climate Impact Program
Executive Summary
The Government of Rwanda (GoR) views the tea sector as central to the country’s economic
development across a number of key dimensions, including its potential to raise smallholder
farmer incomes (and thus reduce poverty), provide good returns and investment
opportunities for private investors, and assist the country in meeting its balance of
payments targets. Indeed, the tea sector is now the third largest employer in Rwanda and
tea is one of the country’s principal exports. In response to the constraints identified in the
first Tea Sector Strategy, specifically those related to farmers’ perceived low level of
remuneration for their green tea leaves, the GoR reformed the pricing mechanism for green
leaf from a factory-cost basis to a more transparent pricing mechanism that directly links
the prices paid to farmers to the international market price for “made tea.”
The World Bank Group (WBG) is conducting an evaluation study with the Ministry of
Agriculture and Animal Resources (MINAGRI), its National Agriculture Export Board (NAEB),
with the assistance of the National Institute of Statistics (NISR). The study aims to rigorously
evaluate the effects of the reform of the green leaf team pricing mechanism on tea
production, tea quality, and the living standards of Rwandan smallholder tea farmers. The
study exploits household survey data—baseline and follow-on in late 2015—to inform a set
of research questions around the impact of the reform. The baseline household survey was
conducted in October 2013 and this report summarizes the process and outcomes of the
baseline survey as well as additional data analysis on determinants of observed baseline
patterns and farmer perceptions of a number of key variables flagged by MINAGRI.
Key findings of the 2013 Rwanda Tea Sector Baseline Survey for the impact evaluation focus
on a comparison of profiles and outcomes between tea farmers and non-tea farmers to
provide a sense of differences or commonalities across these two groups. Analysis of the
baseline survey data suggests that Rwandan tea farmers are, on average, a more affluent
group compared to their non-tea growing counterparts, in terms of household income,
household assets, livestock holdings and land holdings. Tea farmers also tend to be more
diversified in terms of their cropping activities and they use 20 percent more hired labor
than non-tea counterparts. The majority of tea farmers are in cooperatives and there are
sizeable variations in revenues across these cooperatives in the data. There is a strong
gender dimension to tea revenues with female–headed households receiving 24 percent
less than male-headed households, on average. The analysis looks at some of the
determinants of these observed patterns and also examines farmer perceptions of the tea
sub-sector and the reforms. A key finding is the lack of any knowledge of the price reforms
among the overwhelming majority of non-tea farmers. The majority of non-tea farmers
surveyed are reluctant to engage in tea cultivation at any price, citing access to land, a lack
of tea production expertise, and other more profitable crops as key explanations for their
reluctance. Finally, the baseline report also yields key recommendations for the 2015 follow
on survey on which the final impact evaluation will be based.
1
1. Overview
1.1 Introduction The impact evaluation study aims to rigorously evaluate the effects of the reform of the
green leaf team pricing mechanism in Rwanda on tea production, tea quality, and the living
standards of Rwandan smallholder tea farmers. The evaluation will assess whether, and to
what extent, the tea pricing reforms recently implemented in Rwanda have impacted tea
revenues, farmer incomes, farmer wealth assets, productivity, and expansion. The study
exploits household survey data to inform a set of research questions relevant to the above
themes. The baseline household survey was conducted in October 2013 and will be
complemented by an end-line survey in December 2015.
1.2 Tea in Rwanda Tea cultivation was introduced in Rwanda in 1952. Since then, production has increased
steadily from a mere 160 tons in 1961 to 22,503 tons in 2012, with the exception of a
production decline during the 1994 genocide. Growth in production mirrors growth in areas
under cultivation, as well as moderate yield gains over this same period (FAOSTAT) (Figures
1 and 2). Over 90 percent of production is exported, making tea one of the country’s
principal exports. The tea sector is now the third largest employer in Rwanda, behind coffee
and the public sector, and currently employs about 60,000 people. The Government of
Rwanda (GoR) views the tea sector as central to the country’s economic development
across a number of key dimensions, including its potential to raise smallholder farmer
incomes (and thus reduce poverty), provide good returns and investment opportunities for
private investors, and assist the country in meeting its balance of payments targets.
Figure 1 Tea Production in Rwanda (1961–2012) Figure 2 Tea Area Harvested (1961-2012)
Source: FAOStat
Despite its importance to the economy, Rwandan tea accounts for only a small share of the
total volume traded globally, making Rwanda a price taker on the international market.
Currently, tea is grown on 14 estates comprising a total area of approximately 14,000
hectares. An estate is a tea-producing unit including a factory, a plantation, private tea
plots, and an associated forest to provide fuel wood to the factory for tea processing.
-5
0
5
10
15
20
25
30
Pro
du
ctio
n (
00
0'
Ton
s)
Tea Production (000' Tonnes)
0.00
2.00
4.00
6.00
8.00
10.00
12.00
14.00
16.00
1961 1968 1975 1982 1989 1996 2003 2010
Area harvested (000' Ha)
Source: FAOstat
2
However, not all estates have all of these components; for instance, some do not own
plantations. All tea plantations are located near factories because of the perishable nature
of tea, which requires processing within a few hours of harvesting. Tea production is
primarily organized into three types of farming systems:
1. Industrial blocks: these are large plantations integrated into a processing plant. In total, they account for 31 percent of the area under tea cultivation. They employ wage labor;
2. Coopthés: These are also plantations, but owned by cooperatives. They employ a mixture of family and wage labor. They account for 8 percent of cultivated area. Cooperatives play a key management role in the tea cultivation and production processes. They distribute fertilizers, collect and deliver tea leaves to the factory, pay pluckers and the growers themselves, and redistribute surplus earnings to members;
3. Thé villageois: These are smallholder tea plots essentially relying on family labor. They account for 61 percent of the total area under tea cultivation. The majority of private smallholders are also organized under cooperatives. As with the coopthés, these cooperatives organize importation and distribution of fertilizers, and they facilitate access to credits and transportation of tea leaves. Production decisions, however, reside completely in the hands of the individual smallholder households.
Prior to 2003, all tea factories and the associated plantations were owned by the state.
OCIR-Thé, a parastatal directly responsible for the production, processing, and marketing of
tea, was in charge of managing the 10 state-owned factories. An eleventh estate and
factory, established in 1972, has always been privately owned by SORWATHE.
Since 2003, the GoR has gradually privatized the tea industry and restructured OCIR-Thé.
Two new (private) factories were established. OCIR-Thé assumed a new institutional role,
and many of its capacity constraints have been addressed. By 2007, the industry had
succeeded in raising productivity, particularly through improved fertilizer application. During
the same year, the size of the tea industry increased to $32 million compared to $23 million
in 2003, representing a 39 percent increase.1 The privatization process was finally
completed in 2012 with the privatization of the two remaining state-owned factories.
Some of the problems facing the sector prior to privatization, however, remained relevant.
Key challenges flagged by the first Tea Sector Strategy for Rwanda (2003), still remained
unresolved by the end of the privatization in 2012. In general, the industry was considered
to have remained in a low-quantity-low-quality trap. The factors underpinning this were
identified as:2
Tea farmers’ perception that they were not adequately remunerated for the quality of their green leaf delivered to the tea factories;
Tea farmers’ production of low-quality green leaf because of poor farming practices;
1 Ministry of Agriculture and Animal Resources. 2008 . A Revised Tea Sector Strategy for Rwanda –
Transforming Rwanda’s Tea Industry (MINAGRI: Kigali). 2 Ibid.
3
Tea farmers’ limited funds to invest in working capital for fertilizers and transportation;
Low capacity in tea husbandry and plantation management; Tea factories at maximum processing capacity, both qualitatively and quantitatively; Lack of investment in marketing for direct sales and in diversification of Rwanda’s
tea product portfolio.
The issue of price incentives to farmers formed the primary motivation for the
government’s reform of the green leaf tea pricing policy—the object of this impact
evaluation. At the time of the baseline report, the remainder of the above challenges still
remain, although to differing degrees. For instance, the yield in Rwanda remains low
compared to other producing countries in Africa, as shown in Figure 3.
Figure 3 Tea Yield in Rwanda vs. in Kenya and Uganda
1.2.1 Use of evidence base by Rwanda authorities The Ministry of Agriculture and Animal Resources (MINAGRI) has an interest in, and is
committed to, evidence-based policy formation. This is clear from the depth and variety of
impact evaluations conducted for the agricultural sector in Rwanda. MINAGRI is currently
conducting rigorous evaluations of agricultural financial products, extension services, feeder
roads, and it uses impact evaluation methodologies to measure the causal effects of its large
Land Husbandry, Water Harvesting, and Hillside Irrigation project. Most of the impact
evaluations are ongoing, but lessons learned and preliminary results have been taken into
account in project scale-up decisions, which have been shared with other stakeholders in
the sector. Furthermore, MINAGRI was very keen to use the baseline data of this impact
evaluation on the green leaf tea price reforms, to inform their current tea expansion
strategy. This led to the survey instrument including questions on farmer perceptions of the
reforms and their responses to different levels of price incentive. MINAGRI further
requested cross-sectional analysis based on these data to feed into its policy decisions
around the tea sector strategy in 2014, and that analysis forms a part of the deliverable to
the government, in addition to this baseline report.
0.0
0.5
1.0
1.5
2.0
2.5
3.0
19
61
19
65
19
69
19
73
19
77
19
81
19
85
19
89
19
93
19
97
20
01
20
05
20
09
Yie
ld (
ton
s/ h
a)
Kenya Rwanda Uganda
Source: FAOstat
4
1.3 World Bank Group Support of the Tea Sector Reforms There have been four distinct phases in IFC Advisory Services’ support of the tea sector in
Rwanda, as part of broader World Bank Group (WBG) support of the sub-sector (tea) and
the sector (agriculture) at large: (1) an investor conference in November 2010; (2) the
privatization strategy note for the two publicly owned factories remaining in 2011; (3) the
green leaf pricing reform of July 2012; and (4) the collaboration with the World Bank on a
WBG concept note in support of the GoR’s further tea sector expansion program,
envisioning 18,000 additional hectares of tea cultivation under private management. The IFC
interventions were delivered under the second phase of the Rwanda Investment Climate
Reform Program, which aims to improve agribusiness market efficiency through market,
legal, and regulatory remedies. This second phase of the program drew on the recent gains
from extensive Doing Business reforms,3 and it is a part of a broader WBG program to
increase exports and investment in Rwandan agribusiness. These efforts include
considerable World Bank support to the GoR’s investments in productivity, capacity, and
irrigation infrastructure across the country. The impact evaluation looks specifically at the
green leaf tea pricing reforms for which WBG assistance was requested by the GoR.
In response to the constraints identified in the first Tea Sector Strategy, specifically those
related to farmers’ perceived low level of remuneration for their green tea leaves, MINAGRI
undertook to reform the pricing mechanism for green leaf. Up to that time, the price for
green leaf was determined annually, and the amount paid to farmers for their green leaf tea
was based on the self-declared costs of a factory or cooperative. The system was generally
viewed as cumbersome and opaque. The price paid to farmers, furthermore, was not linked
to the market price and was low by comparison to the benchmark provided by other tea-
producing countries. This had the effect of discouraging farmers from planting and
harvesting tea and investing in their plots and the quality of their output, given higher
returns could be obtained from cultivating alternative crops (Figure 4).
Figure 4 Estimated Revenues for Smallholder Crops on 0.25 ha (RWF ‘000s, 2011)
Source: FAOStat; Rwanda National Agricultural Export Board (tea)
3 In FY10, under Phase 1, Rwanda became the global top Doing Business reformer and made Doing Business
reform history by improving 76 ranks in one year. In the 2014 report, Rwanda is ranked 67th
of 183 countries. In addition to ease of doing business, impressive gains were made in important areas such as reducing the time and cost to trade, obtaining business licenses, registering a business, and the overall strengthening of the legal, regulatory, and policy framework for special economic zone developments in the country.
-
200
400
600
800
1,000
Tho
usa
nd
s
5
The GoR requested technical assistance from the WBG’s IFC Advisory Services to inform and
structure the reform of the green leaf pricing mechanism. In July 2012, the GoR passed a
radical reform to the pricing mechanism with this technical support. The reform consists of a
more transparent pricing mechanism that directly links the prices paid to farmers to the
market price for “made tea.” The made-tea price is now based on a weighted average price
of all Rwandan tea sales on the market, of which farmers receive a fixed percentage
(currently 30 percent), as opposed to the factory-cost model described above. These prices
are now set every six months rather than on an annual basis as previously. In addition, in
order to encourage farmers to improve the quality of their harvested green leaves, farmers
also receive a bonus depending on whether or not their tea exceeds a quality threshold.
Thus, the policy is designed to increase both the quantity and quality of tea produced by
farmers. The key implementers are MINAGRI and the Rwanda National Agricultural Export
Board (NAEB).
1.4 Objectives 1.4.1 Objectives of the impact evaluation
The objectives of the impact evaluation are linked to the objectives of the reform. The green
leaf price reform was undertaken by the GoR with the objectives of increasing farmers’
incomes and enhancing incentives to raise the productivity and the quality of the raw
material utilized by the tea factory. In addition to the government’s rural poverty alleviation
objectives, price reform was seen as critical to improving the efficiency and competitiveness
of Rwandan tea factories and increasing tea export revenues (MINAGRI, 2012).4
The purpose of the impact evaluation study is to evaluate the impact of tea price reform
measures on tea production, farmer tea revenues, and a range of farmer welfare metrics.
The impact evaluation is conducted for the introduction of a market-based green leaf pricing
mechanism for farmers, which was passed in Cabinet in July 2012 and became effective in
September 2012 after clearing the legislative process. The reform is expected to have the
intermediate outcomes of improving prices, quality, productivity, and output volumes. The
intermediate outcomes in turn are expected to lead to higher farmer revenues, growth in
exports and jobs, and greater food security (Error! Reference source not found., in black,
evaluation aims).
Figure 5 Theory of Change for Rwanda’s Tea Sector Price Reform
4 MINAGRI. 2012. Cabinet Briefing Paper: New Green Leaf Tea Pricing Model to Tea Farmers in Rwanda.
Activities
•green leaf price reform
Intermediate Outcomes
•improved prices
•improved quality
•improved productivity
•increased output
Higher Order Outcomes
•higher farmer revenues
•growth in exports
•growth in jobs
•greater food security
6
The green tea price reform activities are expected to result in intermediate outcomes and
ultimately, higher order outcomes by driving the following impacts:
Direct impact: Assuming that the price reform is implemented as planned, it will have a
direct impact in increasing farmer revenues by raising the prices received by farmers. The
assumption is that the old factory-cost model suppressed farmer returns and also isolated
farmers from market signals.
Indirect impacts: Increased tea prices following the reform are expected to affect revenue
indirectly through the incentive effect, encouraging greater on-farm investment.
Investments in tea farms (for example, in labor, land husbandry, improved inputs) are in
turn expected to improve farm productivity, thereby impacting farmer revenues through
increased quantities produced. Furthermore, the fact that prices are now made to reflect
differences in quality is expected to incentivize farmers to invest more in quality
improvement. This in turn is expected to bring additional higher prices for farmers, thus
providing a third channel to increasing farmer revenues. The final channel is through the
incentive effect anticipated that higher prices will induce new farmers to start producing
tea.
While the direct impact of the price reform is expected to affect all types of farmers in the
same manner, the indirect impact may work differently for independent smallholder
farmers (thé villageois) and farmers under cooperatives (coopthés). The structure of
production decision-making is different across these two groups of tea farmers, and the
response to incentives may differ as well. As part of the cross-sectional analysis promised to
the GoR from the baseline data, the report also informs on differences in price and
production by tea-farming system. Given the hypothesis of change and resulting impacts
following the reforms, and the additional analysis requested by GoR, the impact evaluation
study aims to answer the following specific questions:
- Are farmers actually being paid higher prices compared to the price levels before the reform, and to what extent have these raised household revenues?
- Are there any changes in production levels before and after the reform? Are these due to improvements in productivity or expansion in cultivation, or both?
- Did the price reform lead to improvements in productivity? - Are productivity improvements due to the expected increase in on-farm
investments?
The design of the survey instrument and the impact evaluation study to answer these
questions is discussed in the next section.
1.4.2 Policy perspective The study will contribute to GoR’s objectives for the tea sector (as well as resource
allocation) by supporting evidence-based policy making. The results of the impact
evaluation of Rwanda’s tea sector reforms are intended, at the government’s request, to
inform the GoR’s agricultural policy around pricing regulation and incentives for different
commodities. This is true of the final results, as well as for the immediate baseline study. As
mentioned, further to the impact evaluation questions, the baseline data will help to inform
7
key questions posited by the GoR in its current tea policy work. These include: what is the
level of awareness of reforms among farmers; what price level would induce farmers to stop
producing tea; what price level would induce them to increase production; what other
factors will motivate farmer participation in a potential tea expansion; and, are there
existing cross sectional differences between farmers according to type of farming system or
cooperative arrangement that affect incentives even at baseline? Results on the
demography of tea producing households and their use of labor through the production
cycle can inform on rural labor markets effects, impacting broader policies in the sector.
Beyond Rwanda, the outcomes of the impact evaluation can inform and motivate similar
price mechanism reforms in agricultural commodity markets where outmoded mechanisms
persist. The vast scope for reform in agricultural prices has been extensively documented,
with a long history of price distortion in every region of the world, spanning from farm gate
to export prices. 5 The factory-cost model reformed in Rwanda is common to many
regulated commodity price regimes. The results of the impact evaluation can therefore add
to the evidence basis for the reform of similar types of pricing regulations elsewhere in the
world.
2. Methodology
2.1 Randomization and Sample Design The study relies on a random sample of households within the tea-growing areas of
Rwanda. As each household within the target population had an equal chance of selection,
the sample is representative of the underlying population in the tea-growing areas
(including both tea and non-tea producers).
The study used the 2004 Rwanda tea sector survey Enquête Quantitative de Base auprѐs des
Ménages des zones Théicole (EQBT) as the template in designing the sampling strategy for
the 2013 baseline survey. Accordingly, the baseline closely follows the geographic coverage,
sample size, and sampling strategy of the 2004 survey. The target population is all rural
households (both tea-producing and non-tea producing) in tea-growing areas within the
production radius of Rwanda’s 14 tea factories. The EQBT covered 12 of these 14 factories.
The two new tea growing areas correspond to the two new tea factories opened after 2004
(see Annexes 1 and 2).
Multi-stage cluster sampling was used to randomly select 2,062 households. To start, all tea-
growing administrative sectors corresponding to each of the 14 tea factories were
purposively selected based on a list provided by NEAB, for a total of 71 sectors of which 63
(or 89 percent) were drawn randomly.6 Within each of these sectors, cells, villages and
households were selected using multi-stage randomization. In the first stage of the
randomization, administrative cells were selected randomly from each sector, such that the
number of cells is proportionate to the size of the sector (on average, two cells were 5For a recent landmark study, see Kym Andersen, ed. 2009. Distortions to Agricultural Incentives: A Global
Perspective, 1955–2007. (Washington, DC: Palgrave-MacMillan and The World Bank). 6 The plan was to cover the entire list of 71 tea-growing sectors provided by NEAB; however, the size had to be
reduced to 63 to accommodate logistical considerations. Still this covers about 89 percent of the total list of tea-growing sectors provided by NEAB.
8
selected from each sector). In the second stage, two administrative villages were randomly
selected from each of the selected cells. In the third stage, eight households were selected
randomly from each of the selected villages. Household selection was done using a walking
pattern where, in a given village, the supervisor assigned a specific walking direction to each
enumerator in the team by using a random reference point in the village. Each enumerator
then used five intervals in selecting a household. Within this walking pattern, specific
protocols were provided for replacing households where either respondents were not
available or refused to be interviewed.
In 55 of the selected cells, both villages randomly selected were found not to grow tea. In
these cases, field teams worked with village officials to categorize all villages in the cell as
either tea-growing or non-tea-growing. One village was then randomly selected from each
list, to replace each of the two originally sampled villages.
2.2 Sample Size and Strategy The overall sample size for the current study was based on power calculations detailed in
Section 5, below. Given the clustered sampling design described above, 2,000 households is
more than sufficient to statistically distinguish effects in the outcomes in which the
evaluation is most interested. Unsurprisingly, given the deliberate similarities in sampling
frame and desired outcomes, the overall sample size is similar to that of the EQBT. (The
implications of the design effects associated with the sample size are explored in more
detail in Section 5.)
2.3 Instruments for Data Collection and Data Quality Assurance The time frame for the analysis will span the effective introduction of the price reforms in
October 2012 to end-2015, when a follow-up survey is envisaged. The Cabinet decision of
July 2012 only cleared the legislative process in September 2012, with implementation
following shortly thereafter. This implied a one-year recall period for the surveyed
households for impact evaluation baseline conducted in 2013. The 2013 baseline survey was
specifically designed to collect recall data from the period before the reform
implementation started. A one-year lag in recall is common in annual agriculture surveys
(which require recall over the past seasons), and research shows little evidence of recall bias
affecting data quality.7 However, the impact evaluation will also use factory-level data on
quality, quantities, and prices to triangulate the survey data described here.
The questionnaire was designed to ensure that adequate information was collected from
each household to facilitate the empirical analysis for the impact evaluation, but that the
length was short enough to minimize interviewee fatigue. The questionnaire comprised
eight separate sections. The first section contained household and interview identification
information necessary for a follow-up survey. The second section contained household
roster information. A third section contained questions on household location, housing
quality, and household assets. A fourth section asked questions on household income from
7 Beegle, Kathleen; Carletto, Calogero; Himelein, Kristen. 2011. Reliability of recall in agricultural data. Policy Research
working paper ; no. WPS 5671. Washington, DC: World Bank. http://documents.worldbank.org/curated/en/2011/06/14286199/reliability-recall-agricultural-data
9
different sources and household expenditure activity, both for the last month and the last
year. Section five solicited information on household land holdings and agricultural
activities. Section six only applied to tea farmers and contained questions on various
aspects of their tea production activity. Section seven asked questions of all farmers on the
recent privatization and price reforms, while a final section asked respondents for an
assessment of their household living standards, both currently and in terms of changes over
the last 12 months. The average duration of interviews for non-tea-famers was about 45
minutes and 75 minutes for tea farmers.
The data were collected by the interviewers electronically using hand-held phones. This
ensured a high degree of quality control in that this method eliminates transcription errors
and ensures the correct sequence of questions is being asked by the interviewer. The
resultant data were transmitted back to the survey team coordinators at the end of each
day. The research team was forwarded data collected in the field on a weekly basis, and
conference calls were scheduled regularly to review any quality or other issues that
emerged in the data collection process.
2.4 Data Storage, Management, and Access Policy All household survey data collected for the impact evaluation will be filed in the World Bank
Microdata library (microdata.worldbank.org), in accordance with the World Bank’s Open
Data Policy. All personally identifying information will be removed from the datasets before
publication to the Microdata library. Once filed, the data will be publicly accessible.
3. Sample Representativeness and External Validity of the Survey
The 2004 EQBT and this baseline survey relied on the same sample frame, all smallholder
tea farmers in the sectors associated with tea factories.8 Comparisons between the two
surveys therefore yield useful descriptive evidence on changes in the sector which help to
inform the analysis. In addition, comparisons of selected household demographic profiles in
the tea-growing areas allow an assessment of the extent to which they are either
comparable or have evolved. Thus, while the EQBT is not directly exploited as part of the
impact evaluation for the price reforms, it does provide some valuable external information
against which our data can be compared and selected outcomes validated. In particular, it
can be used to validate the study’s use of recall data as a baseline, because there are few
observable changes between the two surveys that can’t be explained by general
demographic shifts in Rwanda.
To test the comparability of the populations, we first examine the degree to which the
sampling design described in section 2.1 yielded a comparable share of tea farmers to that
obtained in 2004. Second, we assess the size of landing holdings reported in the tea growing
areas. As the size of such holdings has moved glacially in Rwanda overall, little change would
be anticipated across these two surveys. Third, we investigate how farmers’ real incomes
have evolved over the last nine years and assess how this compares with the national trends
in agricultural incomes reported for Rwanda over the same period. Finally, we use selected
8 Essama-Nassah et al. (2008) provides a description of the EQBT survey. The design effects of the two surveys are
compared in section five.
10
outcomes from the 2004 survey to compare with some preliminary descriptive analysis
undertaken in section six.
3.1 Comparison of the 2013 and 2004 Tea Sector Surveys Table 3.1 reports the proportion of the sample across three aggregate economic activity
categories for the two years. The more recent sample comprises a slightly larger share of
households that are directly engaged in tea sector activities, broadly defined.
Table 3.1: Main Economic Activity of Head of Household by Tea Sector Survey
Main Activity of the Household Head Tea Sector Survey 2004
Tea Sector Survey 2013
Tea Related Agricultural Activity 0.185 0.234
Non-tea Related Agriculture Activity 0.545 0.610
All Other Activities 0.270 0.156
Sample Sizes 2,062 2,108 Notes to table 3.1: ‘All Other Activities’ in 2004 include ‘not in work’, which represents more than one-half of this category.
This category did not feature in the 2013 survey.
Table 3.2 compares selected demographic profiles of the heads of household across the two
surveys. The average household size has remained relatively stable over the two years. The
average head of household is 1.4 years older in the later compared to the earlier year and
the average differential is statistically significant (|t| = 3.0). The proportion of female-
headed households has fallen by almost five percentage points, with the reduction again
found to be statistically significant (|z| = 3.4). The marital status pattern has also altered
sharply with fewer heads now reported as widowed. In addition, the proportion of
household heads reporting no education at all has fallen by nine percentage points over the
relevant period. The differences noted in these average outcomes most likely reflect
changes in outcomes with respect to these variables within the underlying population rather
than sampling procedures. In particular, the average age, gender, and marital status
patterns reported are to be anticipated given the country’s recent history of genocide, while
the growth in human capital levels is also reflective of the widening access to education
evident in Rwanda in more recent decades.
Table 3.2: Demographic Profiles of Head of Household by Tea Sector Survey
Characteristics Tea Sector Survey 2004
Tea Sector Survey 2013
Age in years of Head of Household (HoH) 45.7 (15.8)
47.14 (15.3)
HoH Female 0.289 0.241
Marital Status of HoH:
Married 0.697 0.779
Widowed 0.257 0.179
Other 0.046 0.042
Educational Background of HoH:
11
Characteristics Tea Sector Survey 2004
Tea Sector Survey 2013
No Education 0.424 0.334
Complete or Some Primary Education 0.491 0.561
Other (including Secondary & Higher) 0.095 0.105
Household Demographics:
Number of Household Members 5.16 (2.26)
5.40 (2.26)
Sample Sizes 2,062 2,108
Table 3.3 reports the total landholdings for farmers who own their own land measured in
hectares. The 2013 survey validated respondent estimates with information from land title
deeds, which contain accurate information on plot sizes, in about one-half of the
interviews.9 The analysis reported in section 5.2 below provides some tentative evidence of
a tendency for respondents to over-estimate their land holdings so those measurements not
triangulated by the land documentation may be over-estimated. Land sizes reported in the
2004 survey were entirely self-reported estimates provided by respondents, so that the
current baseline survey employed more robust data collection on this variable. It should be
noted, however, that in the case of 15 percent of famers, no size holdings for any plots were
reported and this is reflected in the reduced sample sizes reported in Table 3.3, below.
Nevertheless, there is close concordance in land holding patterns across the two surveys. In
particular, average and median land size holdings remained fairly stable over this period.
The average farm in both years had approximately three-quarters of a hectare, while the
median was 0.4. There does appear to have been a decline in the inequality of land holdings
among farmers with both the upper/lower decile ratio falling. However, the difference is
not found to be statistically significant. Therefore, on balance, there appears to have been
little movement in either average land holding size or its dispersion.
9 Approximately half of the respondents did not provide interviewers access to land title deeds.
12
Table 3.3: Total Land Size Holdings of Farmers (in Hectares)
Mean & Quantiles Tea Sector Survey 2004
Tea Sector Survey 2013
Average 0.773 (2.44)
0.760 (1.26)
10th Percentile 0.050 0.077
25th Percentile 0.150 0.177
50th Percentile 0.390 0.400
75th Percentile 0.880 0.855
90th Percentile 1.500 1.756
90th/10th Percentiles 30.0 22.8
Gini Coefficient 0.621 (0.031)
0.593 (0.013)
Sample Sizes 1,463 1,443 Notes to table 3.3: Standard deviations/errors are reported in parentheses.
The household income measure used for our comparative analysis is the sum of gross
incomes earned by the household in the last 12 months from (i) own farm activity, (ii)
providing labor services to other farms, and (iii) non-farm enterprise activity (including from
any salaried employment). We can directly compare this gross income measure across the
two surveys. There has been a real increase in this income measure of about 20 percent
over this nine-year period (almost 2 percent growth per annum). This aligns with similar
estimates by the Thematic Report on Income for Rwanda10 from 2012, which reported an
increase of about 23 percent nationally in agricultural incomes between 2005/6 and
2010/11 [see National Institute of Statistics of Rwanda (2012)]. The magnitude of income
inequality observed in the income data in Table 3.4 appears to have remained relatively
stable over this nine-year period.
Table 3.4: Household Incomes for Rwanda Farmers (RWF at 2010 Prices)
Mean & Quantiles Tea Sector Survey 2004
Tea Sector Survey 2013
Average 146515.3 (325301.8)
174607.3 (259850.1)
10th Percentile 13834.6 17086.1
25th Percentile 40133.3 42715.3
50th Percentile 92783.1 102516.8
75th Percentile 173248.9 211868.0
90th Percentile 306069.2 375895.9
90th/10th Percentile Household Income 22.1 22.0
Gini coefficient for Household Income 0.554 0.550
10
The recent Thematic Report on Income for Rwanda based on the EICV3 conducted in 2010/11 gives more complex net income estimates, but proved to be very data intensive for the baseline survey purposes. The baseline gross income estimates do not include any estimate of the home consumption of food and non-food products as reported in the EICV3 report. Therefore, the income estimates reported here are not directly comparable to those contained in the above report.
13
Mean & Quantiles Tea Sector Survey 2004
Tea Sector Survey 2013
(0.066) (0.029)
Sample Sizes 1,463 1,604 Notes to Table 3.4: (a) The household income are all expressed in 2010 prices using the Rwanda consumer price index; (b)
the sample is comprised of all farmers in both years; (c) standard deviations/errors are reported in parentheses; (d) the
sample sizes are different relative to Table 3.3 given the number of missing values for land size.
4. Findings of the Tea Sector Baseline Survey
This section summarizes some of the key findings of the 2013 Rwanda Tea Sector Baseline
Survey for the impact evaluation. The emphasis in this section is on a comparison of profiles
and outcomes between tea farmers and non-tea farmers to provide a sense of any
differences or commonalities across these two groups. As mentioned, the survey was
conducted after the introduction of the green-leaf price reform and relies on recall data to
describe household-level and other outcomes which pre-date the reform. These are either
outcomes that are immutable over such a short period of time or relate to time periods
prior to the implementation of the price reforms.
The next sub-section provides demographic profiles of these two farmer types (4.1), which
is followed by five sub-sections that focus in turn on land holdings (4.2), agricultural activity
profiles (4.3), baseline (pre-reform) tea prices both overall and by co-operative (4.4),
baseline tea production yields (4.5), and a summary of conclusions (4.6).
4.1 Demographic Profiles of Tea and Non-tea Farmer from 2013 Baseline Survey Table 4.1 provides the frequency distribution for the main economic activities of the heads
of household as reported in the 2013 baseline survey. Over 86 percent of surveyed
households report the head as engaged in agricultural activities either on their own or
someone else’s farm. This compares to about 90 percent nationally as reported in the EICV3
conducted in 2010/11. The sample sizes for the responses in the other non-agricultural
activities are negligible.
Table 4.1: Summary of Economic Activity of the Head of Household
Main Economic Activity of the Head of Household Sample Proportion
Number of Observations
Works on Own Farm 0.7676 1618
Works on Someone Else’s Farm 0.0977 206
Self-employed Vendor or Trades Person 0.0237 50
Salaried Employment in the Public Sector 0.0304 64
Salaried Employment in the Private Sector 0.0209 44
All Other Activities 0.0597 126
Total 1.0000 2,108
Table 4.2 reports the main occupational activities of these heads and reveals that the
majority working in agriculture are farmers (94 percent) with one-quarter of these recorded
as tea farmers.
14
Table 4.2: Summary of Main Occupational Activity of the Head of Household
Main Occupational Activity of the Head of Household Sample Proportion
Number of Observations
Farmer Engaged in Tea Cultivation 0.2035 429
Tea Plucking 0.0242 51
Casual Labourer on Tea Farms 0.0066 14
Farmer Engaged in Non-tea Cultivation 0.6106 1287
Working on Someone Else’s Farm 0.0066 14
Working in Non-agricultural Sectors 0.1485 313
Total 1.0000 2,108 Notes to Table 4.2: The non-agricultural sectors include those working in manufacturing, mining, construction,
transportation, commerce and retail trade, education and health, public sector administration, and other sectors.
The sample of all farmers provides the basis for computing the descriptive statistics
informing the demographic outcomes reported in Table 4.3 below. Given their immutable
nature over the short period between the price reforms and the survey implementation,
these profiles are taken to represent the baseline demographics obtaining prior to these
reforms.
The profiles are delineated across tea and non-tea producing farmers to provide insights on
the extent of any differences across these two groups. On average, tea farmers are almost
seven years older than non-tea farmers and are six percentage points more likely to report a
physical disability. About one-quarter of households are female-headed in both sub-
samples, with more widows reported among tea farmers. Approximately one-third of all
farmers report having no formal education, and the distribution of educational attainment
between the two samples (tea and non-tea farmers in the baseline) is not found to be
statistically different on the basis of a goodness-of-fit chi-squared test. The number of
household members is statistically higher for the sub-sample of tea farmers, although the
number of children under five in the household is statistically lower than that of non-tea
farmers. This latter finding is unsurprising given the older age profile of the tea farmers. The
overwhelming majority of farmers own their residential dwellings, though the ownership
rate is statistically higher for the tea farmers. The physical attributes of residential
properties, however, exhibit no statistical differences across the two sub-samples in terms
of either the number of bedrooms or the presence of electricity.
Table 4.3: Demographic Profile of Tea and Non-tea Farmers (Main Occupation)
Characteristics Tea Farmers Non-tea Farmers
Test of Difference
Age of Head of Household (HoH) in years 53.2 (14.5)
46.3 (14.9)
|t| = 8.2; pv=0.00
HoH Female (=1) 0.2634 0.2595 |z| = 0.2; pv=0.87
HoH Physical Health Disability (=1) 0.2004 0.1445 |z| = 2.7; pv=0.00
Marital Status of HoH:
Married – monogamous (=1) 0.6760 0.7312 ‡
Married – polygamous (=1) 0.0629 0.0350 ‡
15
Widowed (=1) 0.2214 0.1888 ‡
Divorced (=1) 0.0210 0.0256 ‡
Single (=1) 0.0186 0.0194 ‡
χ2
4 = 9.0; pv=0.05
Educational Background of HoH:
No Education (=1) 0.3263 0.3675 ‡
Some Primary Education (=1) 0.3357 0.3364 ‡
Complete Primary Education (=1) 0.2564 0.2222 ‡
Some Secondary Education (=1) 0.0536 0.0490 ‡
Complete Secondary Education (=1) 0.0047 0.0062 ‡
Other (including Higher) Education (=1) 0.0233 0.0186 ‡
χ2
5 = 3.8; pv=0.58
Household Demographics:
Number of Household Members 5.72 (2.48)
5.28 (2.15)
|t| = 3.5; pv=0.00
Number of Female Adults Aged 15+ 1.94 (1.08)
1.61 (0.93)
|t| = 6.2; pv=0.00
Number of Male Adults Aged 15+ 1.65 (1.10)
1.38 (0.98)
|t| = 4.7; pv=0.00
Number of Children Aged < 5 0.67 (0.89)
0.87 (0.90)
|t| = 3.9; pv=0.00
Number of Children Aged 6 to 14 1.50 (1.44)
1.45 (1.34)
|t| = 0.7; pv=0.47
Dwelling Characteristics:
Own House (=1) 0.9790 0.9594 |z| = 2.2; pv=0.00
The Number of Bedrooms (=1) 2.86 (0.95)
2.79 (1.04)
|t| = 1.3; pv=0.19
Electricity (=1) 0.0606 0.0816 |z| = 1.4; pv=0.16
Sample Sizes 429 1287 Notes to Table 4.3: (a) The summary statistics reported here relate only to households whose main activity is reported as
farming; (b) Standard deviations are reported in parentheses; (c) Goodness-of-fit chi-squared tests are used to test for the
differences in variables with mutually exclusive groups; (d) |z| denotes the absolute z-score, |t| denotes the absolute t-
value, and pv denotes prob-value.
4.2 Land Holdings and Size Table 4.4 reports the number of plots and the size of land holdings by tea and non-tea
farmers at baseline. The average number of plots owned by tea farmers is about 1.5 times
higher than that for non-tea farmers. The average tea farmer allocates 1.5 plots to tea
cultivation, which represents about one-third of the average. Both the average and median
tea farmer plot sizes, as measured in hectares, are almost twice the corresponding figures
for the sample of non-tea farmers, with the point differentials well determined in all cases.
The size distribution of land holdings is more dispersed among non-tea farmers.
Table 4.4 also reveals that the average land size devoted to tea cultivation is about 0.6
hectares, though the median is about one-half of this, suggesting that a small number of the
larger sized holdings are impacting the mean. Overall, the average and median tea farmers
devote roughly about one-half of their holdings to tea cultivation. There is a greater
dispersion in the distribution in the land size used by tea farmers for tea production
16
compared to their non-tea activities with the Gini land coefficient computed at 0.56 for the
latter (not reported in Table 4.4) compared to 0.62 estimate reported in this table.
Table 4.4: Land Size Profile for Tea and Non-tea Farmers
Tea Farmers
Non-tea Farmers
Test of Difference
Average Number of Plots Owned 5.232 (3.873)
3.555 (2.647)
|t| = 9.9; pv=0.00
Average Number of Tea Plots Owned 1.527 (1.347)
‡ ‡
Average Total Land Holdings (Hectares) 1.128 (1.592)
0.607 (1.053)
|t| = 7.3; pv=0.00
10th Percentile Land Holdings (Hectares) 0.174 0.060 |t| = 11.4 pv=0.00
25th Percentile Land Holdings (Hectares) 0.358 0.133 |t| = 14.5;pv=0.00
50th Percentile Land Holdings (Hectares) 0.680 0.305 |t| = 11.7; pv=0.00
75th Percentile Land Holdings (Hectares) 1.288 0.641 |t| = 9.4; pv=0.00
90th Percentile Land Holdings (Hectares) 2.382 1.352 |t| = 5.2; pv=0.00
90th/10th Percentile Land Holdings 13.69 22.53 ‡
Gini Coefficient for Land Holdings 0.529 (0.027)
0.601 (0.014)
|t| = 2.4; pv=0.00
Average Tea Holdings (Hectares) 0.586 (1.005)
‡ ‡
10th Percentile Tea Holdings (Hectares) 0.054 ‡ ‡
25th Percentile Tea Holdings (Hectares) 0.120 ‡ ‡
50th Percentile Tea Holdings (Hectares) 0.279 ‡ ‡
75th Percentile Tea Holdings (Hectares) 0.633 ‡ ‡
90th Percentile Tea Holdings (Hectares) 1.208 ‡ ‡
90th/10th Percentile Tea Holdings 22.37
Gini Coefficient for Tea Holdings 0.620 (0.025)
‡ ‡
Sample Size for Plots 427 1221 ‡
Sample Size for All Land Holdings 405 1090 ‡
Sample Size for Tea Land Holdings 348 ‡ ‡ Notes to Table 4.4: (a) The data used are based on those who own their own land and where households where farming is
the main activity; (b) standard deviations are reported in parentheses; (c) the statistical tests for the different quantiles are
based on the t-test for a tea-farmer dummy in quantile regression; (d) ‡ denotes not applicable for calculation; (e) the
differing sample sizes relates to missing values for the measured land variable.
4.3 Livestock Ownership, Crop Activity, and Labor Usage of Farmers Following recall methodology of agricultural surveys, Table 4.5 summarizes the nature of
livestock ownership and crop activity for both farmer types as of 12 months prior to the
interview (thus before the implementation of the price reforms). The table reports the
baseline ownership rate and number for seven separate livestock categories. In general, the
livestock ownership rate is statistically higher for tea compared to non-tea farmers for all
but one livestock group (rabbits). The sample averages, conditional on livestock ownership,
only differ statistically for the case of cattle. In this case, the tea farmers own a slightly
higher average number of cattle.
17
There are three growing seasons in Rwanda. These span September to January (Season A),
February to June (Season B), and July to August (Season C). The crop growing activities in
Seasons A and B cover September 2012 to June 2013, prior to the full implementation of the
green leaf price reform. These two seasons are the most intensive growing seasons for
Rwandan farmers with only 47 percent of all farmers in the survey reported as engaged in
any kind of cultivating activity in Season C. This is unsurprising because only farmers with
access to valley bottom plots have sufficient water to cultivate in Season C, and most
farmers have hillside plots.
Table 4.5: Livestock Ownership of Farmers 12 Months Prior to Interview
Tea Farmers
Non-tea Farmers
Test of Difference
Own Cattle (=1) 0.684 0.549 |z| = 4.8; pv=0.00
Average Number of Cattle (conditional on ownership)
1.77 (1.17)
1.60 (0.99)
|t| = 2.2; pv=0.02
Own Goats (=1) 0.464 0.350 |z| = 4.2; pv=0.02
Average Number of Goats (conditional on ownership)
2.42 (1.40)
2.30 (2.02)
|t| = 0.7; pv=0.48
Own Sheep (=1) 0.288
0.220 |z| = 2.8; pv=0.00
Average Number of Sheep (conditional on ownership)
2.22 (1.55)
2.06 (1.36)
|t| = 1.1; pv=0.29
Own Pigs (=1)
0.330 0.231 |z| = 4.1; pv=0.00
Average Number of Pigs (conditional on ownership)
1.65 (1.54)
1.81 (2.08)
|t| = 0.8; pv=0.44
Own Poultry (=1) 0.391 0.297
|z| = 3.6; pv=0.00
Average Number of Poultry (conditional on ownership)
3.16 (2.51)
3.45 (4.39)
|t| = 0.8; pv=0.21
Own Rabbits (=1)
0.145 0.118 |z| = 1.4; pv=0.15
Average Number of Rabbits (conditional on ownership)
4.94 (4.28)
4.61 (4.25)
|t| = 0.6; pv=0.58
Notes to Table 4.5: (a) standard deviations are reported in parentheses for continuous variables; (b) the statistical tests for
the continuous variables are based on the t-test, while the z-score is used for the proportional variables; (c) pv denotes the
prob-value.
Table 4.6 reports the average number of crops grown in each of the A and B Seasons by
farmer type. Given tea farmers possess both a larger number of plots and larger sized
holdings compared to their non-tea counterparts, it is not surprising that the average
number of crops grown in each of the two seasons is also, on average, statistically higher.
Tea farmers grow an average of one more crop than the non-tea producing farmers in
18
Season A, which is the most intensive growing season. The averages are also statistically
higher for the sample of tea farmers in Season B.
The four most popular crops grown in each of the two seasons are climbing beans, maize,
sweet potato and Irish potato. In all cases other than climbing beans in Season A, tea
farmers record statistically higher activity rates in the other three crops. For Season B, the
rates across all four crops are comparable. Overall, tea farmers appear more diversified
across crops and have higher levels of engagement than non-tea farmers in the four most
popular crops produced in the tea growing areas of Rwanda.
Table 4.6: Crop Activity of Farmers 12 Months Prior to Interview
Tea Farmers
Non-tea Farmers
Test of Difference
Crop Activity:
Average Number of Crops Cultivated (Season A)
5.012 (3.144)
4.063 (2.769)
|t| = 5.9; pv=0.00
Average Number of Crops Cultivated (Season B)
4.255 (2.469)
3.666 (2.535)
|t| = 4.1; pv=0.00
Most Popular Crop Type – Season A:
Climbing Beans (Haricot) 0.749 0.753 |z| = 0.2; pv=0.88
Maize 0.700 0.603 |z| = 3.6; pv=0.00
Sweet Potato 0.515 0.460 |z| = 1.96; pv=0.05
Irish Potato 0.424 0.333 |z| = 3.4; pv=0.00
Most Popular Crop Type – Season B:
Climbing Beans (Haricot) 0.562 0.536 |z| = 0.92; pv=0.88
Maize 0.522 0.553 |z| = 1.1; pv=0.26
Sweet Potato 0.451 0.482 |z| = 1.1; pv=0.26
Irish Potato 0.375 0.337 |z| = 1.4; pv=0.16
Sample Size
427 1,221
Notes to Table 4.6: (a) standard deviations are reported in parentheses for continuous variables; (b) the statistical tests for
the continuous variables are based on the t-test, while the z-score is used for the proportional variables; (c) pv denotes the
prob-value.
Table 4.7 reveals there is a high reliance on household members for harvesting and non-
harvesting chores by both farmer types throughout Seasons A and B. The average number
of days worked by household members is also statistically higher for tea farmers reflecting,
among other things, the larger average number of household members for this group. In
contrast, the use of hired labour is statistically higher for tea farmers across both seasons.
In A and B almost one-half of tea farmers hired in outside labour, while the figure for non-
tea farmers is less than one-third. This may be attributable to the fact that tea farmers own
more plots and have larger average holdings than non-tea farmers, as well as the labor
intensive nature of tea as a crop (plucking). The average number of days worked by hired
labour on farms owned by tea farmers is about 20 percent higher compared to the group of
19
non-tea farmers in Seasons A and B. The median number of days worked by hired labour
across the two seasons is 25 percent higher.
Table 4.7: Labor Usage by Farmers
Tea Farmers
Non-tea Farmers
Test of Difference
Use Household Labour Season A (=1) 0.972 0.971 |z|=0.14; pv=0.88
Use Household Labour Season B (=1)
0.972 0.960 |z|=0.27; pv=0.61
Average Days Worked (Seasons A & B) [Conditional on using HH members)
138.8 (69.9)
121.6 (71.9)
|t|=4.16; pv=0.00
Median Days Worked (Seasons A & B) [Conditional on using HH members)
140 121 |z|=4.4; pv=0.00
Use Hired Labour Season A (=1) 0.491 0.319 |z|=6.44; pv=0.00
Use Hired Labour Season B (=1)
0.456 0.300 |z|=5.92; pv=0.61
Average Days Worked (Seasons A & B) [Conditional on using Hired Labour]
44.4 (49.2)
37.2 (46.7)
|t|=1.66; pv=0.09
Median Days Worked (Seasons A & B) [Conditional on using Hired Labour]
22.5 18 |z|=2.9; pv=0.00
Sample Sizes 421 1218 Notes to Table 4.7: (a) standard deviations are reported in parentheses for continuous variables; (b) the statistical tests for
the continuous variables are based on the t-test, while the z-score is used for the proportional variables; (c) pv denotes the
prob-value.
4.4 The Pattern of Baseline Tea Prices This section examines the pattern of tea prices received by tea farmers prior to the green
leaf price reform. Table 4.8 reports the average price per kilogram of green leaf tea paid to
farmers during a pre-reform period (averaged over the 12-month period prior to the
implementation of the government’s price-reform). The average price per kilogram received
by tea farmers prior to the reforms was RWF 55.8. However, there was substantial variation
in the average price paid across the different co-operatives, with the average differences
found to be statistically significant across the fifteen that feature here. However, the small
sample sizes available for some of the co-operatives merits some interpretational caution
when focussing down on some of the individual co-operatives.
Table 4.8: Average Baseline Price per Kilogram for Green Leaf Tea
Co-operative Sample Proportion
Price per Kilogram in RWF
Overall Average 1.00 55.8 (22.5)
20
Of which:
COOP_1 0.013 26.0
COOP_2 0.096 55.7
COOP_3 0.022 71.3
COOP_4 0.132 39.2
COOP_5 0.041 56.9
COOP_6 0.013 71.3
COOP_7 0.039 52.1
COOP_8 0.080 64.2
COOP_9 0.106 73.5
COOP_10 0.124 66.6
COOP_11 0.013 100.7
COOP_12 0.078 39.0
COOP_13 0.072 35.9
COOP_14 0.017 76.0
COOP_17 0.026 63.7
Non-member 0.128 55.2
F-test for Differences in Co-op Means ‡ 16.25 pv=0.00
Sample Size ‡ 431 Notes to Table 4.8: (a) the baseline prices are the average in the 12 months prior to the reform; (b) pv denotes prob-value;
(c) the F-test for a difference in cooperative means is based ANOVA; (d) ‡ denotes not applicable.
Table 4.9 re-examines this issue from a different perspective through dividing the sample
between Thѐ villageois producers and those producers on co-operative lands. The pre-
reform price received by the former is about 20 percent higher with the difference
statistically significant. This differential is intuitive given that tea producers on co-operative
lands incur deductions for inputs used and services provided on these lands.
Table 4.9: Baseline Price per Kilogram by Tea Holding Type
Tea Producer Type Sample Proportion
Average Price per Kilogram in RWF
Cooperative Lands 0.20 48.3 (24.8)
Thѐ Villageois 0.80 57.6 (21.7)
Test for Difference ‡ |t| =3.5 (pv=0.00)
Sample Size ‡ 431 Notes to Table 4.9: (a) the baseline prices are the average in the 12 months prior to the reform; (b) |t| denotes the
absolute t-value for a difference in means; (c) pv denotes prob-value; (d) ‡ denotes not applicable.
4.5 The Pattern of Baseline Tea Production The baseline figures for farmer tea production are based on the output produced by tea
farmers in the full year prior to the survey based on the factory determined weight. Table
4.10 reports the baseline yield figures. The average yield generated per hectare is 6,394.7
kilograms of tea. However, this masks differences across the distribution with the median
21
reported at a more modest 3,362 kilograms per hectare. In addition, the decile ratio is
about 47, suggesting fairly substantial variation in yield between those producers at the
bottom and the top end of the yield distribution. Finally, Table 4.10 also reports yield per
tea plot, with the average plot producing 1,227.9 kilograms of tea, though the median is
reported at 780.5 kilograms.
Table 4.10: Baseline Tea Production Yield
Mean & Quantiles Kilograms per Hectare Kilograms per Tea Plot
6,394.7 (7902.7)
1,227.9 (1880.8)
10th percentile 347.6 147.2
25th percentile 1,063.1 320.75
50th percentile 3,361.7 780.5
75th percentile 8,498.2 1,465.75
90th percentile 16,297.0 2,519.3
90th/10th percentile ratio 46.9 17.1
Sample Size 272 272 Notes to Table 4.10: (a) standard deviations are reported in parentheses; (b) the calculations are based on all tea farmers
with a maximum of five tea plots; (c) the data are trimmed at the top end to exclude the impact of a small number of
extreme outliers on the calculations reported here. The data trimming comprised about 2% of the sample of tea
producers.
4.6 Conclusion These baseline data provide us with a platform to assess the research questions articulated
in Section 1. In conjunction with the data to be obtained from the planned follow-up survey
in 2015, the impact evaluation will investigate how the green leaf tea price reform has
impacted farmer productivity, revenues, the area under tea cultivation, and farmer non-
income wealth assets including livestock holdings and so on. The impact evaluation of the
price reforms on farmer incomes presents a greater challenge. However, the availability of
benchmark tea revenue data will allow us re-calibrate tea farmer incomes to reflect the
baseline levels that prevailed prior to the reforms and use this baseline for the purposes of
the impact evaluation on farmer incomes. Finally, since all green leaf has to be processed in
factories, the factory level data reviewed by the impact evaluation team provides baseline
and impact data to triangulate quantities (production) and factory-adjudicated quality.
5. Internal Validity of the Study
This section reviews a number of issues related to the internal validity of the baseline study.
The first sub-section explores the magnitude of the sample design effects ex-post. A
subsequent sub-section exploits propensity score matching techniques to get a sense of the
potential problem associated with the measurement of the land size variable. The final sub-
section covers some observations on potential quality differences depending on whether
the head of household or someone deputed by the head of household was the primary
interviewee.
22
5.1 Intra-correlation Coefficients and Design Effects The sampling undertaken for this survey was described earlier in section 2 above and
exploited randomization at a cluster level. The final number of administrative villages
clusters selected within the tea-growing areas of Rwanda was 126 and the average number
of interviewed households within each cluster was approximately 16 households. Clustered
samples are not as statistically efficient as those drawn from simple random sampling. In
particular, similarities among units in clusters (intra-cluster correlation, ) can reduce the
variability of responses from a cluster compared to that drawn from a simple random
sample.
Specifically, a positive correlation between units in the same cluster inflates the variance of
the sample mean. In the extreme theoretical case where = 0, and there is no correlation
in responses within clusters, then all of the variation is within clusters and the sampling is
equivalent to simple random sampling. In the alternative extreme case where = 1, all
responses within a cluster are identical and the effective sample size is simply the number of
clusters. In addition, the design effect, which allows computation of the effective sample
size, is positively related to both the intra-cluster correlation coefficient and the number of
units sampled in each cluster.
In order to examine this issue for the current baseline survey we compute intra-cluster
correlations for the sample of farmers for a selected array of continuous variables in the
survey. The purpose of this exercise is largely illustrative and is intended to demonstrate the
possible magnitude of design effects present in these data. Table 5.1 reports the intra-
cluster correlation coefficients for four separate continuous variables for the sample of
farmers in the survey.
Table 5.1: Intra-cluster Correlations for Selected Variables for the 2013 Tea Survey
Variable Intra-cluster Correlation
Sample Size Number of Clusters
Household Income 0.0563 1716 126
Land Size Holdings 0.0387 1496 126
Household Size 0.0505 1716 126
Age of Head of Household 0.0146 1703 126 Notes to table 5.1: The number of units per cluster is 16.
In general, the reported correlation coefficients are modest. As a further illustration, we
compute the design effect using the household gross income variable introduced in the last
section. The following formula can be used: D = 1)ρ - (n + 1 k where nk is the number of
sampled units per cluster. We assume nk = 16 for this calculation. In empirical analysis
using the income (or land size) measure, for example, standard errors are likely to be 1.36
(1.26) times higher than they would be if we had used a simple random sampling approach.
The foregoing calculations provide sufficient confidence that we have adequate power to
detect any effects in regard to the income (or land size) variable.
23
5.2 The Measurement of Land Size An issue of concern for the study relates to the size of land holdings and the self-reported
estimates provided by respondents. A primary objective of the survey was to ensure that
interviewers consulted land title deeds to obtain an accurate measurement for the size of a
farmer’s land holdings. Only in about one-half of the cases, however, where title deeds
were held by farmers, was the interviewer able to consult the land title deeds.
In order to investigate the implications of this issue further we focus down on the size of the
respondent’s main plot and separate the sample into those for whom a land title deed was
available and seen by the interviewer and all others. We exploit a propensity score matching
(PSM) technique to match those farm households with and without the land title deed
information. Once we have matched the households, using an array of characteristics, we
compute the difference in reported plot sizes in order to get a sense of the magnitude of
differences between these two groups for the main plot. For this exercise, farmers who had
and showed the land title are considered the ”treatment” group, with the remaining
farmers providing the “control” group.
The variables used for both the propensity score ( the probability of the ”treatment”) and
the outcome ( the size of the farm’s main plot11) include: head of household characteristics
(for example, age, gender, marital status, human capital level, and health), the log of
household income and household size, agricultural activity (for example, the number of
crops cultivated in each season, whether the farmer is a tea farmer, the number of different
livestock), the location of the plot ( top or bottom of a hill), and the district within which the
farm is located. A logit model used for the treatment assignment equation validates the
methodology: the constructed propensity score is balanced across treatment and
comparison groups, as are the covariates.
Table 5.2 reports the average treatment effect for the treated in this case. There were 803
treated farms in the common support and these were matched with 707 from the control
group. The ATT estimate suggests the average size of those main plots for which the size
was obtained directly from the title deed is about 20 percent smaller than those for which
this was not the case.12 Therefore, the main plots, where title deeds were used to provide
the plot size measurement, appear to be systematically smaller, on average, suggesting the
area size of the main plots for which self-reported estimates were supplied (and not
obtained from the title deed directly) is over-estimated by close to one-fifth, on average.
Table 5.2: Propensity Score Matching for Main Plot Size Measured in Log Hectares
Number of Treated Number of Controls Average Treatment of the Treated (ATT)
t-ratio
803 707 -0.222 (0.073)
-3.053
11
The outcome variable, size of the farm’s main plot, is expressed as log of hectares and relies on a kernel matching method based on the Epanechnikov density. 12
Bootstrapped standard errors were computed for the average treatment effect using 100 replications.
24
Notes to Table 5.2: The standard error in parentheses is bootstrapped using 200 replications.
Although the magnitude of this finding is best interpretable as suggestive, it does highlight a
potential caveat for the conduct of empirical work using the land size data reported in this
survey given some measurements provided are based on self-reporting. For example, the
work of Carletto, Gourlay and Winters (2013) emphasize the problems associated with self-
reported estimates of land area for empirical work. It is clear that additional work is
required to try and address the informational deficit on land size and this is discussed
further in the recommendations section.
5.3 Respondent Responses for the Non-heads of Household Another issue encountered by interviewers was the fact that about one-quarter of the
interviewees for the farming households were not the head of household. This appears to
have had implications for access to important information regarding the land title deeds (as
already discussed in section 5.2 above), which was seen as crucial in securing accurate
information on the size of farmer land holdings as well as other household characteristics.
About 15 percent of farm households reported having no land title deeds for their main
plot. In only 65 percent of cases where respondents reported having land title deeds were
the documents viewed by the interviewer. There was a large differential in interviewer
viewing rates between respondents who were heads of household and those who were not.
The viewing rate for the former was 68.9 percent and 57.4 percent for the latter. The
differential is found to be statistically significant with a z-score of 4.29 (pv=0.00).
A similar problem emerged for tea farmers in regard to access to the farmer’s tea factory
notebook recording the volume of green tea leaves delivered to the factory and the prices
paid. Only 18 percent reported not having such a book. The viewing rate by interviewers
was 72.9 percent when heads of households were present for interview and 61.4 percent
when not. Again the differential is statistically significant with a z-score of 2.05 (pv=0.02).
Overall, the foregoing highlights the informational benefits to interviewing heads of
household rather than those deputed by the head to provide such information.
6. Other Descriptive Findings from Baseline Survey: Requested Sector Evidence by GoR
6.1 Household Income Distributions The household income measure used for this part of the analysis, as described earlier, is the
sum of gross annual income earned by the household in the 12 months prior to the
interview date (hence, prior to the reform) from (i) own farm activity, (ii) providing labour
services to other farms, and (iii) non-farm enterprise activity (including from any salaried
employment). No imputations for own consumption or deductions for inputs are
undertaken for this measure.
Table 6.1 provides more detail on the distributional differences in household income
between the two sub-samples of tea and non-tea farmers. The average household income
for tea farmers is 60 percent higher to that of non-tea farmers, while the median is 85
percent higher. Annual household income is greater (and statistically so) at all reported
percentiles of the distribution. The income gap is widest for the poorest farmers (230
percent at the 10th percentile) and narrowest for the wealthiest farmers (39 percent at the
25
90th percentile). The average tea farmer’s gross household income level is higher than
approximately 82% of non-tea farmer households.
Table 6.1 also reveals a wide income distribution across the sample, i.e. there is a large
difference between the lowest and highest income households for both tea farmers and
non-tea farmers. The income dispersion is considerably more pronounced among the non-
tea farmers. The Gini coefficient, a commonly used measure of income inequality, is nearly 7
points higher for the non-tea farmers, and the difference is statistically significant. Similarly,
the ratio of income between the 90th and 10th percentile of non-tea farmers is double that
of tea farmers.
Table 6.1: Household Income and Wealth Profile of Tea and Non-tea Farmers
Tea Farmers
Non-tea Farmers
Test of Difference
Average Household Income (RWF) 282568.5 (381388.0)
176291.6 (265730.4)
|t| = 6.2; pv=0.00
10th Percentile Household Income (RWF) 50000.0 15000.0 |t| = 10.3 pv=0.00
25th Percentile Household Income (RWF) 95311.5 40000.0 |t| = 11.3; pv=0.00
50th Percentile Household Income (RWF) 117170.0 95500.0 |t| = 9.4; pv=0.00
75th Percentile Household Income (RWF) 337185.0 200000.0 |t| = 8.1; pv=0.00
90th Percentile Household Income (RWF) 555000.0 400000.0 |t| = 3.0; pv=0.00
Average Log Household Income 12.0340 (1.0807)
11.3571 (1.2916)
|t| = 9.6; pv=0.00
90th/10th Percentile Household Income 11.1 26.7 ‡
Gini coefficient for Household Income 0.514 (0.025)
0.583 (0.013)
|t| = 2.5; pv=0.00
PCA Wealth Index 0.258 (1.501)
-0.178 (1.441)
|t| = 5.3; pv=0.00
Correlation Coefficient between Household Income and Wealth Index
0.526; pv=0.00
0.459; pv=0.00
‡
Sample Sizes 424 1180 Notes to Table 6.1: (a) The data used are based on non-zero income for those households where farming is the main
activity; (b) standard deviations are reported in parentheses; (c) the statistical tests for the different quantiles are based on
the t-test for a tea-farmer dummy in quantile regression; (d) ‡ denotes not applicable for calculation.
Figure 5 plots the income distribution for all farmers in the sample and reveals a left-skewed
distribution. Figure 6 shows the income distributions by farmer type and reveals that the
central location of the distribution for tea farmers lies to the right of the non-tea farmer log
income distribution, suggesting both a higher mean and median income level for the sample
of tea farmers.
26
Figure 5 Kernel Density of the Log of Annual Farmer Household Income
Figure 6 Kernel Density of the Log of Annual Farmer Household Income by Farmer Type
0.1
.2.3
.4
De
nsity
6 8 10 12 14 16The Log of Annual Household Income
0.1
.2.3
.4
De
nsity
6 8 10 12 14 16ly
Tea-farmers
Non-tea farmers
kernel = epanechnikov, bandwidth = 0.2493
Kernel density estimate
27
An asset index, based on the household’s non-income wealth, is also constructed from the
survey data using Principal Components Analysis (PCA). The assets include those relating to
housing quality (i.e., the number of bedrooms, whether or not the residence has electricity),
the number of radios, the presence of a television set, mobile phones, bikes and motorbikes
owned by the household, and the numbers of different livestock owned by the household
(i.e., the number of cattle, sheep, goats, poultry). The distribution of the asset index broadly
follows the same pattern as household income with tea farmers possessing more assets
than non-tea farmers (Figures 7 and 8). On the basis of Table 6.1, the average PCA value for
tea farmers is about 0.44 of a standard deviation higher than the average value reported for
the non-tea farmers, with the differential between the two groups statistically significant.
Figure 7 Kernel Density of the PCA Index of Household Assets – All Farmers
0.1
.2.3
De
nsity
-5 0 5 10kernel = epanechnikov, bandwidth = 0.2638
28
Figure 8 Kernel Density of the PCA Index of Household Assets by Farmer Type
Figure 9 plots the PCA and the log household gross income (expressed as a standardized
variable) for all farmers and reveals a reasonably close relationship between the two
distributions. The Pearson product moment correlation coefficient computed for the
household income and the wealth index provides a statistically significant coefficient of
0.487 suggesting a reasonable degree of correlation between the two measures and
comparable to what is generally found in the literature for developing countries. The
estimated correlation coefficient is found to be higher for the tea farmers (0.53) compared
to the non-tea farmers (0.46).
0.1
.2.3
.4
De
nsity
-5 0 5 10Scores for component 1
Tea-farmers
Non-tea farmers
kernel = epanechnikov, bandwidth = 0.3327
Kernel density estimate
29
Figure 9 The Kernel Densities for Log Household Income and the Wealth Index
We now briefly draw again on the data from the 2004 survey to place the gap in income
between tea and non-tea farmers in a recent historical context. Table 6.2 below reveals
that in 2004 tea farmers already enjoyed a sizeable income advantage over their non-tea
growing counterparts. However, this had widened sharply by 2013. In 2004, the average
tea farmer’s income was 33 percent higher than non-tea farmers and by 2013, this gap had
almost doubled. Thus, at first blush, tea farmer incomes appear to have risen significantly
over this period but the explanation for this requires further and more detailed empirical
analysis.
Table 6.2: Real Household Incomes in the Tea Growing Areas of Rwanda 2004 and 2013
2004 2013
Ratio of Average Tea Farmer to Non-tea Farmer Household Income
1.34 1.60
Sample Size 1,505 1,604
6.2 Household Income Determination of Rwandan Farmers in the Tea-growing Areas We now explore the key determinants of annual gross household income of farmers in the
2013 sample. The logarithm of annual gross income introduced in section 6.1 provides the
metric for the dependent variable and the OLS procedure is used in estimation. The sample
is pooled across both farmer types. The reported goodness-of-fit of the regression model is
adequate and suggests that almost 30 percent of the variation in the log of household
income is explained by the set of covariates (including controls for 12 districts) in the
specification reported in Table 6.3.
The log of household income is found to be insensitive to the age of the head of household.
Incomes in households headed by women are subject to a penalty of almost 29 percent
0.1
.2.3
.4
De
nsity
-5 0 5 10lyx
Log Income
Wealth Asset
kernel = epanechnikov, bandwidth = 0.1916
Kernel density estimate
30
compared to those headed by men, on average and ceteris paribus, while a physical
disability to the household head is associated with a 22 percent lower household income.
There are significant effects for the formal human capital measures included, with
households headed by those without any formal education receiving around 58 percent less
than those with a head reporting secondary level or better. Household income is also
positively related to the number of household members (both adults and children), and an
additional member engaged in paid work (in the previous month) raises the household
income by 11 percent, on average and ceteris paribus.
The estimated income elasticity with respect to land size holdings is well determined, but
fairly inelastic. In particular, a 10 percent increase in land holdings raises household income
by a modest 2 percent.13 There is a positive relationship between the number of crops
grown in season A (September to January) and household income. An additional crop grown
in this season, relative to the average, raises annual household income by about 5 percent,
on average and ceteris paribus, while a one crop increase in the number grown in season C
(July and August) actually reduces household income by approximately 3 percent. Recent
infrastructural improvements in the local roads accessing markets are associated with
higher household incomes. In particular, those households where such improvements have
occurred in the last five years enjoy a 12.8 percent income premium compared to
households in areas that have not benefitted from such recent infrastructural
enhancements, on average and ceteris paribus.
As discussed, Table 6.1 revealed a wide disparity in household income between tea and non-
tea farmers. The ceteris paribus income differential reported in Table 6.3 is also found to be
sizeable. Specifically, farmers engaged primarily in the production of tea receive, on
average, 68 percent more in household income compared to their non-tea counterparts. In
addition, the quantile regression estimates reported in Table 6.4 confirm the income
advantage across selected quantiles of the conditional log household income distribution as
originally exhibited in the raw data in Table 6.1. The pattern reported in Table 6.4 is one of
a declining differential across the conditional distribution with a statistically significant
difference recorded between the 90th and the 10th percentiles. However, the magnitude of
these estimated effects are attenuated with the inclusion of other covariates. Nevertheless,
the quantile regression findings confirm that tea farmers at the bottom end of the
conditional distribution fare significantly better in income generation terms than their non-
tea counterparts also located at this part of their corresponding distribution.
13
An inelastic response is also found if the number of plots is used instead of the total land size. The estimated elasticity in this case is found to be 0.248, which is dimensionally comparable to the land elasticity reported in table 6.3.
31
Table 6.3: Determinants of Log Annual Household Income for Tea and Non-tea Farmers
Characteristics Sample Mean
OLS Estimates
Constant
1.0000 11.9308*** (0.3558)
Age of Head of Household (years) 48.27 (14.93)
0.0002 (0.0126)
Age of Head of Household Squared (years) 2553.1 (1545.8)
-0.0001 (0.0001)
HoH Female (=1) 0.2493 -0.2542** (0.1268)
HoH Physical Health Disability (=1) 0.1652 -0.2422*** (0.0855)
Marital Status of HoH:
Married – monogamous (=1) 0.7263 -0.0562 (0.1349)
Married – polygamous (=1) 0.0417 0.1439 (0.1780)
Divorced (=1) 0.0201 -0.2050 (0.1995)
Single (=1) 0.0201 -0.5272* (0.2853)
Widowed (=1) 0.1918 ‡
Educational Background of HoH:
No Education (=1) 0.3355 -0.8604*** (0.1333)
Some Primary Education (=1) 0.3384 -0.6987*** (0.1299)
Complete Primary Education (=1) 0.2442 -0.4552*** (0.1322)
All Other Education Greater than Primary (=1) 0.0819 ‡
Household Demographics:
Number of Adult Household Members 3.18 (1.55)
0.0824*** (0.0230)
Number of Children Household Members 2.31 (1.68)
0.0442** (0.0195)
Number of Household Members Engaged in Paid Work (Last Month)
0.82 (0.89)
0.1052*** (0.0406)
Land Holdings & Agricultural Activities:
Log of Total Land Size (Hectares) -0.9743 (1.2154)
0.2016*** (0.0272)
Number of Crops Grown in Season A 4.31 (2.89)
0.0499*** (0.0163)
Number of Crops Grown in Season B 3.79 (2.48)
0.0052 (0.0191)
32
Number of Crops Grown in Season C 1.23 (1.84)
-0.0291* (0.0177)
Non-tea Farmer (=1)
0.7126 ‡
Tea Farmer Primary Activity (=1) 0.2874 0.5194*** (0.0667)
Tea Farmer Secondary Activity (=1) 0.0151 0.0577 (0.2230)
Community Attributes & District Controls:
Improved Roads to Markets in the Last 5 Years (=1) 0.7055
0.1209* (0.0711)
Twelve District Controls Included
not reported
Yes
Adjusted R-Squared
n/a 0.2809
Sample Size 1,392 1,392 Notes to Table 6.3: (a) standard deviations are reported in parentheses for the sample means; (b) robust standard errors
are reported in parentheses for the OLS estimates; (c) ***, **, and * denote statistical significance at the 0.01,0.05 and 0.1
level of statistical significance respectively; (d) ‡ denote base group in OLS estimation; (e) n/a denotes not applicable.
Table 6.4: Quantile Regression Estimates for Tea Farmer Effect
10th Percentile
25th Percentile
50th Percentile
75th Percentile
90th Percentile
Tea Farmer 0.7949*** (0.1504)
0.6842*** (0.0916)
0.5076*** (0.0856)
0.4022*** (0.0785)
0.1564*** (0.0867)
Notes to Table 6.4: (a) The regression model contains all the variables included in the specification reported in table 6.3; (b) bootstrapped standard errors are computed using 200 replications; (c) ***, **, and * denote statistical significance at the 0.01,0.05 and 0.1 level of statistical significance respectively.
The empirical analysis reported in table 6.3 was developed further to investigate whether there was a gender dimension in the income returns to tea-growing activity. A comparable specification to the one reported in table 6.3 was re-estimated using only the sub-sample of female farmers. The results of the exercise are reported in Annex Four. A key finding of this analysis is that female tea farmers enjoy a sizeable income advantage over female non-tea farmers. Specifically, on average and ceteris paribus, the household income advantage is of the order of about 88%. However, the advantage enjoyed by female tea farmers relative to their non-tea farmer counterparts is not statistically different to the income advantage earned by male tea farmers relative to their male non-tea farming counterparts. The computed t-ratio for this latter differential is 0.87 with a prob-value of 0.38.
Finally, we also investigated whether there was evidence of differences in the income rewards for those tea farmers with health disabilities compared to non-tea farmers with such disabilities. In order to examine this issue we again estimated a comparable specification to that reported in table 6.3 but restricted the sample to those farmers reported to have a physical disability. The results of this exercise are reported in Annex Five but interpretational caution is required here given the small sample size of 230 farmers used here for the analysis. Table 6.3 revealed that, on average, farmers with a physical disability
33
had statistically lower household incomes than their able-bodied counterparts. The estimates from Annex Five suggest that tea farmers with a physical disability have household incomes that are, on average and ceteris paribus, 112% higher compared to non-tea farmers with such disabilities. This differential is also found to be statistically higher than that enjoyed by able-bodied tea farmers compared to their able-bodied non-tea counterparts at the 10% level of significance using a two-tailed test. The computed t-ratio is 1.75 in this case with a prob-value of 0.080. Thus, acknowledging the caveats regarding the sample size used here, the effect of physical disability on income determination is mitigated for the sub-sample of tea farmers.
6.3 Tea Reforms and the Non-tea Producers The future growth of the small farmer tea sector in Rwanda will require, among other
things, attracting existing non-tea famers into tea-growing activities as well as increasing the
amount of land cultivated by current tea farmers. Both will depend on incentives, among
other determinants. This sub-section examines some of the further challenges and
difficulties facing this type of transition as revealed in the survey responses provided by the
large sub-sample of non-tea farmers. This analysis is informed by the questions included at
the request of the GoR on farmer awareness, perceptions and conjectured responses to the
price reform.
We first investigate the awareness of non-tea farmers regarding both the factory
privatization programme and the recent green leaf tea reforms, and their attitude to
engaging in tea cultivation. Table 6.5 reports non-tea farmer awareness of the two most
recent tea sector reforms. Approximately one-quarter of non-tea farmers knew of the
privatization program but only 17 percent were aware of the recent price reforms.
Furthermore, as Table 6.6 reveals, no price, however high, would induce about two-thirds of
non-tea farmers to engage in tea cultivation. In other words, their reservation price to
engage in tea cultivation is set to infinity in this case.
In order to explore this particular issue further, we use a probit regression model to
interrogate the determinants of this reluctance among non-tea farmers. The dependent
variable assumes a value of one for the two-thirds of non-tea farmers who would not
engage in tea production at any price, and zero otherwise. Table 6.7 provides a description
of the explanatory variables, their sample average values, and the corresponding probit
marginal and impact effects.
The gender and health status of the head of household emerge as the most important
determinants. Female headed households are 13.3 percentage points more likely to be
reluctant to engage in tea production at any price, on average and ceteris paribus. If the
head of household is afflicted by a physical disability, the reluctance probability increases by
9.6 percentage points. There are no statistically significant effects for the key marital status
variables and human capital effects are also found to be absent. In addition, none of the
three welfare or asset wealth metrics included in the specification register statistical
significance at a conventional level and this result is invariant to excluding either land,
34
income or the wealth index in turn. In addition, recent improvements in road infrastructure
also fail to yield a statistically significant effect in this regression model.
On the other hand, farmers who have past experience cultivating tea, or know of either the
privatization or the price reform are significantly less likely to be reluctant to cultivate tea at
any price. A former tea farmer is 24.2 percentage points less likely to be reluctant to engage
in tea production compared to a farmer who has never engaged in tea production, on
average and ceteris paribus. In addition, knowledge of the factory privatization programme
reduces the probability of being reluctant to engage in tea cultivation at any price by 7.7
percentage points. However, the strongest effect is reserved for those non-tea farmers who
are aware of the price reform policy. These farmers are 15.5 percentage points less likely to
be reluctant to engage in tea production activity, on average and ceteris paribus.
Table 6.5: Attitudes & Perceptions of Tea Reforms among Non-tea Farmers
Question Yes (%)
‘Are you aware that tea factories were recently privatized?’
26.5
‘Are you aware of the government’s recent green leaf price reforms?’
16.9
Sample Size 1,168
Table 6.6: Minimum Green Leaf Price per Kilogram Required to Encourage Non-tea Farmers to Cultivate Tea
Question Responses (%)
‘What is the minimum price per kilogram (in RWF) that would encourage you to engage in tea production?’
Between 100 RWF and 160 RWF 8.0
Between 161 RWF and 190 RWF 1.2
Between 191 RWF and 230 RWF 5.3
Greater than 230 RWF 18.2
No price would encourage tea production 67.3
Sample Size 1,287
Table 6.7: Probit Model for Non-tea Farmers’ Reluctance to Cultivate Tea
Characteristics Sample Mean
Marginal/ Impact Effects
Age of Head of Household (years) 46.49 (14.70)
0.0016 (0.0014)
HoH Female (=1) 0.2503 0.1326** (0.0575)
HoH Physical Health Disability (=1) 0.1504 0.0960** (0.0823)
35
Marital Status of HoH:
Married (=1) 0.7745 0.0696 (0.0751)
Divorced (=1) 0.0196 -0.2215* (0.1354)
Single (=1) 0.0206 -0.0250 (0.1355)
Widowed (=1) 0.1853 ‡
Educational Background of HoH:
No Education (=1) 0.3471 -0.0755 (0.0703)
Some Primary Education (=1) 0.3357 -0.0778 (0.0691)
Complete Primary Education (=1) 0.2379 -0.0116 (0.0684)
All Other Education Greater than Primary (=1) 0.0793 ‡
Household Demographics:
Number of Adult Household Members 3.02 (1.46)
-0.0143 (0.0131)
Number of Children Household Members 2.35 (1.63)
-0.0224** (0.0105)
Household Income & Wealth Measures:
Log of Annual Household Income 11.38 (1.31)
-0.0057 (0.0148)
Wealth Index based on Principal Components Analysis (PCA)
-0.06 (1.48)
0.0081 (0.0140)
Land Holdings & Agricultural Activities:
Log of Total Land Size (Hectares) -1.202 (1.251)
-0.0085 (0.0144)
Number of Crops Grown in Season A 4.05 (2.75)
-0.0057 (0.0102)
Number of Crops Grown in Season B 3.62 (2.49)
0.0121 (0.0112)
Number of Crops Grown in Season C 1.11 (1.80)
-0.0386*** (0.0104)
Former Tea Farmer (=1) 0.0412 -0.2422*** (0.0881)
Knowledge of Recent Tea Sector Policy Reforms
Aware of Tea Factory Privatizations (=1) 0.2873 -0.0766* (0.0463)
Aware of Green Leaf Price Reform (=1) 0.1812 -0.1549*** (0.0533)
Community Attributes & District Controls:
Improved Roads to Markets in the Last 5 Years (=1) 0.7116
0.0331 (0.0386)
36
District Controls Included
n/a Yes
McFadden R-Squared n/a 0.1512
Sample Size 971 971 Notes to table 6.7: (a) maximum likelihood (asymptotic) standard errors are reported in parentheses; (b) ***, **, and *
denote statistical significance at the 0.01, 0.05 and 0.1 level of statistical significance respectively.
In addition to green leaf tea price, the non-tea farmers were also asked questions on their
perception of other constraints that impact their willingness to engage in tea production.
Table 6.8 provides responses from the sample of non-tea farmers for the three most
popular constraints cited. Over two-thirds cite the absence of adequate land, while close to
one in five flag a lack of expertise in tea growing. A reasonable number (14 percent) of non-
tea farmers also identify the higher profitability of other cash crops as a reason for not
engaging in tea production. A lack of either non-labor inputs, extension services or credit
was quoted by only 2 percent of farmers in each case as representing constraints on their
engagement in tea production.
Table 6.8: Factors Preventing Engagement of Non-tea Farmers in Tea Cultivation
Constraint Yes (%)
(1) Lack of Adequate Land 65.8
(2) No Expertise in Tea Production 18.8
(3) More Profitable to Cultivate Other Crops 14.1
Sample Size 1,287
In an accompanying piece of empirical analysis (not reported here, but see Annex 3 for the
table summarizing results), we separately modelled the three most popular responses as
reported in Table 6.8 again using a probit model. The dependent variables were coded one
for a “yes” response and zero for a “no.” The model included the set of covariates
contained in the specification reported in Table 6.7 in addition to some other variables
depending on the outcome modelled. In the case of the question regarding more profitable
crops, the key crops grown by non-tea farmers found to increase the probability of a ”yes”
response for this question were Irish potatoes and climbing (or haricot) beans. The former
increased the probability of a “yes” response by almost 10 percentage points, while the
latter by a more modest 1.4 percentage points. Both of these crops have exhibited a sharp
rise in popularity among farmers in Rwanda in recent years with the area used for
cultivating the latter rising nationally by 50 percent between 1997 and 2007. There has
been an even sharper increase of about 170 percent over the same period in the total area
used to cultivate Irish potatoes. This crop has shorter production cycles (four months)
compared to many others in Rwanda and generally enjoys high yields. It also currently
boasts strong market demand both domestically and in Uganda. Our survey data do not
permit a comparative analysis of input costs for competing crops and tea production, so the
above only represents a very partial analysis of the role of competitor crops in deterring
non-tea farmers from engagement in tea production.
37
In regard to the question on the absence of adequate land, a significant negative estimate
was registered for the log of land holdings, though the effect was numerically small.
Specifically, the marginal effect suggested that a 5 percent increase in land holdings reduces
the probability of a “yes” response by 0.12 of one percentage point, ceteris paribus and
relative to a mean value of 0.64. However, the quality rather than the quantum of land is
likely to be the more important consideration in animating a “yes” response to this
particular question.
In regard to the final constraint on the lack of expertise in cultivating tea, a gender effect
emerges as statistically significant with a female headed household ten percentage points
more likely to acknowledge a lack of expertise compared to a male headed one. Not
surprisingly, former tea farmers are 14 percentage points less likely to cite a lack of such
expertise as a constraint to engaging in tea production.
6.4 Subjective Measures of Household Welfare for Rwandan Farmers We now explore the responses of farmers to a question posed regarding current level of
satisfaction with their household’s standard of living. Table 6.9 reports the responses to this
question on a five-category ordinal scale. The responses are delineated across tea farmers
and non-tea farmers. There is a very small number of responses in the two extreme
categories with only 7% of all farmers reporting themselves as ‘very dissatisfied’ and 2.5% as
‘very satisfied’. The responses suggest that about one-third of all respondents express
themselves as either ‘satisfied’ or ‘very satisfied’ with their household’s standard of living.
However, there is a sharp contrast in responses between tea farmers and non-tea farmers
with 44.5% of the former placing themselves in these top two categories compared to just
30% for the latter. In addition, the difference in the frequency distribution across
satisfaction categories by farmer type is found to be statistically significant at a conventional
level with a chi-squared value of 55.1 and four degrees of freedom.
Table 6.9: Subjective Perceptions of Farmer Well-being
Question Tea Farmers (%)
Non-tea Farmers (%)
‘How satisfied are you with your household’s standard of living today?’
‘Very dissatisfied’ 3.6 8.1
‘Dissatisfied’ 21.7 36.1
‘Neither satisfied nor dissatisfied’ 30.3 26.2
‘Satisfied’ 41.7 27.0
‘Very satisfied’ 2.8 2.6
Sample Size 429 1287
In order to investigate the determinants of these responses in more detail we employ a five-
category ordered probit regression model. The maximum likelihood (ML) estimates are
reported in Table 6.10. Neither a head of household’s age nor formal human capital levels
matter in determining satisfaction outcomes but both gender and health do. In particular, a
38
female headed household is less likely to report being in the ‘satisfied’ category compared
to a male headed household by six percentage points, and a head of household with a
physical disability is less likely to be reported in this category than an able-bodied one by
four percentage points, on average and ceteris paribus. The marital status of the head of
household exerts no independent effect on reported satisfaction but the number of
household members (both children and adult) is found to lower it. For instance, an
additional adult (child) member reduces the probability of the household being reported in
the ‘satisfied’ category by 3.5 (1.2) percentage points. Further, recent infrastructural
improvements in road access to markets in the last five years is also found to enhance
satisfaction levels.
The three variables included to capture the scale of household wealth along three different
dimensions all yield statistically significant positive effects on household living standard
satisfaction levels. The implicit trade-offs inherent in these particular point estimates
suggest that a 10% increase in household income would be required to off-set a 15%
reduction in land holdings to ensure that an average household’s satisfaction (or utility)
level remained constant.14 In addition, a 10% increase in annual household income would
be required to off-set a 0.08 standard deviation reduction in the wealth asset index.
We now turn to the estimate of the impact of being a tea farmer on satisfaction. This is
found to be positive and extremely well determined. In particular, controlling for the array
of included variables, a tea farmer is eight percentage points more likely, on average, to be
‘satisfied’ compared to a non-tea farmer, and 1.3 percentage points more likely to be ‘very
satisfied’. Thus, as noted in Table 6.9, tea farmers appear more satisfied overall than their
non-tea producing counterparts.
Table 6.10: Ordered Probit Model of Farmer Satisfaction with Living Standards
Characteristics Sample Mean
ML Estimates
Age of Head of Household (years) 48.27 (14.93)
-0.0038 (0.0024)
HoH Female (=1) 0.2493 -0.2031* (0.1183)
HoH Physical Health Disability (=1) 0.1652 -0.1330* (0.0823)
Marital Status of HoH:
Married – monogamous (=1) 0.7263 -0.0733 (0.1299)
Married – polygamous (=1) 0.0417 -0.2081 (0.1903)
Divorced (=1) 0.0201 -0.1741 (0.2152)
Single (=1) 0.0201 0.1953
14
However, the proposition of a unitary relationship cannot be rejected by the data in this case with an absolute asymptotic t-ratio of 0.67 reported for this test.
39
(0.2247)
Widowed (=1) 0.1918 ‡
Educational Background of HoH:
No Education (=1) 0.3355 0.1451 (0.1218)
Some Primary Education (=1) 0.3384 0.1080 (0.1183)
Complete Primary Education (=1) 0.2442 0.0806 (0.1188)
All Other Education Greater than Primary (=1) 0.0819 ‡
Household Demographics:
Number of Adult Household Members 3.18 (1.55)
-0.1117*** (0.0219)
Number of Children Household Members 2.31 (1.68)
-0.0377** (0.0185)
Household Income & Wealth Measures:
Log of Annual Household Income 11.57 (1.28)
0.1166*** (0.0280)
Wealth Index based on Principal Components Analysis (PCA)
0.03 (1.49)
0.1484*** (0.0255)
Land Holdings & Agricultural Activities:
Log of Total Land Size (Hectares) -0.9743 (1.2154)
0.0768*** (0.0265)
Non-tea Farmer (=1)
0.7126 ‡
Tea Farmer Primary Activity (=1) 0.2874 0.2410*** (0.0711)
Tea Farmer Secondary Activity (=1) 0.0151 0.0171 (0.2379)
Community Attributes & District Controls:
Improved Roads to Markets in the Last 5 Years (=1) 0.7055
0.1352* (0.0678)
District Controls Included
n/a Yes
Threshold Parameters:
1
θ n/a -1.1395***
(0.4043)
2
θ n/a 0.2244
(0.4033)
3
θ n/a 1.0372***
(0.4039)
4
θ n/a 2.6792***
(0.4088)
McFadden R-Squared n/a 0.0519
Sample Size 1392 1392
40
Notes to table 6.10: (a) maximum likelihood (asymptotic) standard errors are reported in parentheses; (b) ***, **, and *
denote statistical significance at the 0.01, 0.05 and 0.1 level of statistical significance respectively.
6.5 Some Concluding Remarks The contemporary profile of Rwandan tea farmers emerging from both the analysis
undertaken in the current section and earlier in section four suggests, on average, a more
affluent group compared to their non-tea growing counterparts. This affluence is revealed
along a number of different dimensions including household income, household assets, and
land holdings. In addition from section four, we find that tea farmers also tend to be more
diversified in terms of their cropping activities. Overall, they also report themselves as
considerably more satisfied with their current standard of living as opposed to their non-tea
farming counterparts.
The majority of tea farmers are in cooperatives and although some sizeable variations in
revenues across these cooperatives were detected, the cooperative “top-slice” per unit
price for the services provided was modest, being of the order of 20 percent, on average.
The majority of non-tea farmers are reluctant to engage in tea cultivation at any price, citing
access to land, a lack of tea production expertise, and more profitable alternative crops as
key explanations for such reluctance. In particular, climbing beans and Irish potatoes
represent the more popular alternatives and both currently enjoy relatively high market
prices. Furthermore, they both require less demanding and intensive husbandry skills than
tea, though both are more vulnerable to disease, given the lack of adequate crop rotation,
and require a higher usage of pesticides compared to tea. However, such cash crops
compete with tea and clearly represent part of the challenge in persuading farmers to
engage in tea production.
Finally, a key finding from the econometric evidence here is the lack of knowledge of the
price reforms among the overwhelming majority of non-tea farmers. Thus, it is possible that
an information campaign on the nature of the price reforms and the scale of the more
recent and future potential benefits accruing to tea farmers might better inform and
educate non-tea farmers about the relative advantages associated with engagement in tea
production.
41
7. Recommendations and Suggestions for Follow-up Survey
There are four key recommendations for the follow on survey that emerge from the
baseline analysis. These are:
(a) It is important to minimize the use of non-heads of households in providing relevant
household data in the follow-up survey. In particular, it is imperative that the follow-up
survey envisaged under this project increases the viewing rate of both land title deeds and
farmer tea factory books by interviewers in order to enhance the quality of the responses
for these particular questions. There is clearly an enumerator training issue that needs to
be addressed here for the future. An emphasis on using the head of household should also
enhance the prospects of addressing some of the current problems in regard to existing
gaps that that emerged in the 2013 survey in terms of the responses on land size.
(b) The impact evaluation will use a separate factory survey to obtain the historical and
current data maintained by factories on the quantities of tea supplied and the prices given
to the tea farmers that feature in this baseline survey. This is desirable to triangulate the
survey results on quantity and yields, as well as being the only source of quality data (one of
the aspects of the theory of change for the price reform).
(c) It would be useful to investigate the possibility of accessing information held at Rwanda’s
national land registry as an additional way of dealing with the current informational deficit
in regard to land size information in the current survey data. The national land registry
should hold copies of the land title deeds. However, the research benefits of such an
exercise would need to be weighed against the costs of such an exercise, which would not
be trivial.
(d) Given the putative prospect of financial benefits accruing to tea farmers consequent on
the price reforms, an additional module required for the follow-up impact survey should
focus on eliciting detailed information on the investment behaviour of tea farmers in non-
tea related activities in the post-price reform era.
42
References
Carletto, C, Gourlay, S., and P. Winters (2013), Land Area Measurement and Implications for
Agricultural Analysis, Policy Research Working Paper Number 6550, The World Bank,
Development Research Group.
Essama - Nassah .B, Ezemenari Kene and Korman Vijdan, Reading Tealeaves on the Potential
Impact of Privatization of Tea Estates in Rwanda, Policy Research Working Paper, 4566, The
World Bank, 2008
Haisken-DeNew, J.P. and Schmidt, C.M. (1997), Inter-industry and Inter-region Differentials:
Mechanics and Interpretation, Review of Economics and Statistics, Vol. 79, No. 3, pp. 516-
21.
Kym Andersen, ed. (2009), Distortions to Agricultural Incentives: A Global Perspective, 1955-2007, The World Bank, 2009.
National Institute of Statistics of Rwanda (2012), EICV3 Thematic Report: Income, Kigali,
Rwanda.
Ministry of Agriculture and Animal Resources (2008), A Revised Tea Strategy for Rwanda –
Transforming Rwanda’s Tea Industry.
MINAGRI. 2012. Cabinet Briefing Paper: New tea green leaf pricing model to tea farmers in
Rwanda. Kigali, Rwanda.
Krueger, A.B. and Summers, L.H. (1988), Efficiency Wages and the Inter-industry Wage
Structure, Econometrica, Vol. 56, No. 2, pp. 259-93.
43
Annex 1: Tea Growing Districts
44
Annex 2: Tea Growing Sectors per District Covered
District # Tea Growing Sectors
Covered
Rulindo 6
Nyabihu 6 Burera 1
Nyaruguru 6
Nyamagabe 4
Gicumbi 11
Rusizi 6
Nyamasheke 5
Rubavu 3
Ngororero 5
Rutsiro 3
Karongi 7
45
Annex 3: Mean Differences between Co-op and non-Co-op Thé Villageois Farmers
Characteristics Co-op Member
Non-Co-op Member
Test of Differences
Age of Head of Household (HoH) (years) 53.1 (0.81)
51.6 (1.92)
pv = 0463
HoH Female (=1) 0.249 0.232 pv = 0.785
HoH Physical Health Disability (=1) 0.195
0.304 pv = 0.067
Educational Background of HoH:
No Education (=1) 0.31 0.30 pv=0.963
Household Demographics:
Number of Household Members 1.39 (0.07)
1.11 (0.16)
pv=0.163
Household Income & Wealth Measures:
Log of Annual Household Income 12.16 (0.06)
11.79 (0.17)
pv=0.019
Wealth Index based on Principal Components Analysis (PCA) 0.333 (0.08)
0.262 (0.20)
pv=0.749
Land Holdings
Number of Tea Plots 1.59 (0.08)
1.14 (0.05)
pv=0.029
Sample Size 300 54 Notes to table: (a) Standard deviations are reported in parentheses for continuous variables only; (b) pv denotes the prob-
value for the test of differences in means; (c) sample sizes may differ slightly from those reported in the final row across
some characteristics given missing values.
46
Annex 4: Household Income Determinants for Female Farmers
Characteristics Sample Mean
OLS Estimates
Constant
1.0000 10.5046*** (0.7392)
Age of Head of Household (years) 54.22 (14.47)
0.04252* (0.0126)
Age of Head of Household Squared (years) 3148.12 (1617.4)
-0.0004** (0.0002)
HoH Physical Health Disability (=1) 0.2046 -0.2758* (0.1574)
Marital Status of HoH:
Married – monogamous (=1) 0.1643 0.1388 (0.2055)
Married – polygamous (=1) 0.0057 -0.6705** (0.3031)
Divorced (=1) 0.0720 -0.2050 (0.1995)
Single (=1) 0.0489 -0.5409* (0.3194)
Widowed (=1) 0.7091 ‡
Educational Background of HoH:
No Education (=1) 0.5591 -1.1238*** (0.2512)
Some Primary Education (=1) 0.2478 -1.0677*** (0.2720)
Complete Primary Education (=1) 0.1354 -0.4639*** (0.1322)
All Other Education Greater than Primary (=1) 0.0577 ‡
Household Demographics:
Number of Adult Household Members 2.87 (1.45)
0.0732*** (0.0461)
Number of Children Household Members 1.74 (1.51)
-0.0012 (0.0561)
Number of Household Members Engaged in Paid Work (Last Month)
0.71 (0.86)
0.1395*** (0.0758)
Land Holdings & Agricultural Activities:
Log of Total Land Size (Hectares) -1.0282 (1.1863)
0.2200*** (0.0474)
Number of Crops Grown in Season A 4.05 (2.86)
0.0281 (0.0277)
Number of Crops Grown in Season B 3.50 (2.17)
0.0121 (0.0490)
Number of Crops Grown in Season C 1.19 (1.79)
0.0273 (0.0429)
Non-tea Farmer (=1) 0.7003 ‡
47
Tea Farmer Primary Activity (=1) 0.2997 0.6311*** (0.1310)
Community Attributes & District Controls:
Improved Roads to Markets in the Last 5 Years (=1) 0.7176
0.1225 (0.1329)
Twelve District Controls Included
not reported
Yes
Adjusted R-Squared
n/a 0.3290
Sample Size 347 347 Notes to Annex 4: (a) standard deviations are reported in parentheses for the sample means; (b) robust standard errors are
reported in parentheses for the OLS estimates; (c) ***, **, and * denote statistical significance at the 0.01,0.05 and 0.1
level of statistical significance respectively; (d) ‡ denote base group in OLS estimation; (e) n/a denotes not applicable.
48
Annex 5: Household Income Determinants for Farmers with Health Disability
Characteristics Sample Mean
OLS Estimates
Constant
1.0000 11.1374*** (1.1481)
Age of Head of Household (years) 56.26 (14.96)
0.0073 (0.0364)
Age of Head of Household Squared (years) 3388.5 (1728.0)
-0.0002 (0.0003)
HoH Female (=1) 0.3087 -0.0998 (0.3147)
Marital Status of HoH:
Married – monogamous (=1) 0.6696 0.2895 (0.3504)
Married – polygamous (=1) 0.0391 0.6408 (0.4155)
Divorced (=1) 0.0217 -0.6047** (0.2554)
Single (=1) 0.0261 -0.6051 (0.4315)
Widowed (=1) 0.2435 ‡
Educational Background of HoH:
No Education (=1) 0.4956 -0.4473 (0.4095)
Some Primary Education (=1) 0.2696 -0.2397 (0.4353)
Complete Primary Education (=1) 0.1783 -0.0948 (0.4545)
All Other Education Greater than Primary (=1) 0.0565 ‡
Household Demographics:
Number of Adult Household Members 3.47 (1.67)
0.0622 (0.0489)
Number of Children Household Members 2.02 (1.76)
-0.0253 (0.0493)
Number of Household Members Engaged in Paid Work (Last Month)
0.71 (0.86)
0.0522 (0.1092)
Land Holdings & Agricultural Activities:
Log of Total Land Size (Hectares) -0.7634 (1.2003)
0.2053*** (0.0695)
Number of Crops Grown in Season A 4.07 (2.77)
0.0888* (0.0450)
Number of Crops Grown in Season B 3.49 (2.49)
-0.0322 (0.0515)
Number of Crops Grown in Season C 1.17 (1.86)
0.0106 (0.0496)
Non-tea Farmer (=1) 0.6435 ‡
49
Tea Farmer Primary Activity (=1) 0.3565 0.7521*** (0.1542)
Community Attributes & District Controls:
Improved Roads to Markets in the Last 5 Years (=1) 0.6913
0.1478 (0.1614)
Twelve District Controls Included
not reported
Yes
Adjusted R-Squared
n/a 0.4109
Sample Size 230 230 Notes to Annex 5: (a) standard deviations are reported in parentheses for the sample means; (b) robust standard errors are
reported in parentheses for the OLS estimates; (c) ***, **, and * denote statistical significance at the 0.01,0.05 and 0.1
level of statistical significance respectively; (d) ‡ denote base group in OLS estimation; (e) n/a denotes not applicable.
50
Annex 6: Probit Models Explaining “Yes” Outcomes in Table 6.18
Characteristics (1) (2) (3)
Age of Head of Household (years) -0.0050*** (0.0014)
0.0007 (0.0010)
-0.0011 (0.0009)
HoH Female (=1) -0.0722 (0.0649)
0.1004** (0.0545)
-0.0052 (0.0449)
HoH Physical Health Disability (=1) 0.0033 (0.0465)
0.0102 (0.0356)
-0.1019*** (0.0224)
Marital Status of HoH:
Married (=1) 0.0013 (0.0696)
0.0429 (0.0454)
0.0018 (0.0514)
Divorced (=1) 0.0137 (0.1230)
-0.1182 (0.0442)
-0.0722 (0.0569)
Single (=1) -0.0660 (0.1282)
0.0066 (0.0839)
-0.0078 (0.0835)
Widowed (=1) ‡ ‡ ‡
Educational Background of HoH:
No Education (=1) 0.0782* (0.0433)
-0.0639** (0.0287)
0.0117 (0.0301)
Some Primary Education (=1) -0.0277 (0.0412)
-0.0706** (0.0276)
0.0257 (0.0277)
All Other Education Education Categories (=1) ‡ ‡ ‡
Household Demographics:
Number of Adult Household Members -0.0020 (0.0126)
0.0051 (0.0094)
-0.0030 (0.0088)
Number of Children Household Members -0.0078 (0.0106)
-0.0064 (0.0078)
-0.0017 (0.0069)
Household Income & Wealth Measures:
Log of Annual Household Income -0.0141 (0.0147)
-0.0168 (0.0106)
0.0217*** (0.0098)
Wealth Index based on Principal Components Analysis (PCA) -0.0455*** (0.0140)
0.0031 (0.0103)
0.0004 (0.0086)
Land Holdings & Agricultural Activities:
Log of Total Land Size (Hectares) -0.0246* (0.0141)
0.0041 (0.0100)
0.0017 (0.0092)
Number of Crops Grown in Season A 0.0000 (0.0101)
-0.0007 (0.0076)
0.0203*** (0.0064)
Number of Crops Grown in Season B -0.0004 (0.0111)
0.0039 (0.0084)
-0.0015 (0.0082)
Number of Crops Grown in Season C -0.0071 (0.0105)
0.0184** (0.0077)
-0.0134 (0.0089)
Former Tea Farmer (=1) -0.1923** (0.0846)
-0.1392*** (0.0267)
-0.0335 (0.0454)
Knowledge of Recent Tea Sector Policy Reforms
Aware of Tea Factory Privatizations (=1) 0.0843* (0.0448)
-0.0679** (0.0321)
-0.0049 (0.0454)
Aware of Green Leaf Price Reform (=1) 0.0502 (0.0507)
-0.0184 (0.0399)
0.0628* (0.0391)
Community Attributes & District Controls:
Improved Roads to Markets in the Last 5 Years (=1) 0.0011 (0.0375)
-0.0247 (0.0281)
-0.0077 (0.0238)
Cultivates Irish Potato in Season A n/a n/a 0.0066 (0.0265)
51
Characteristics (1) (2) (3)
Cultivates Irish Potato in Season B n/a n/a 0.0434 (0.0292)
Cultivates Irish Potato in Season C n/a n/a 0.0993** (0.053)
Cultivates Haricot Beans in Season A n/a n/a -0.0246 (0.0292)
Cultivates Haricot Beans in Season B n/a n/a 0.0106 (0.0242)
Cultivates Haricot Beans in Season C n/a n/a 0.0143*** (0.0242)
District Controls Included
Yes Yes No
McFadden R-Squared 0.0870 0.0994 0.0908
Sample Size 992 992 992 Notes to table: (a) column (1) contains probit marginal/impact effects for the model where the dependent variable is
coded 1 if the responses was ‘ yes’ to the question on inadequate land; column (2) contains probit marginal/impact effects
for the model where the dependent variable is coded 1 if the responses was ‘ yes’ to the question of no tea expertise;
column (3) contains probit marginal/impact effects for the model where the dependent variable is coded 1 if the responses
was ‘ yes’ to the question that it was more profitable to cultivate other crops; (b) maximum likelihood (asymptotic)
standard errors are reported in parentheses; (c) ***, **, and * denote statistical significance at the 0.01, 0.05 and 0.1 level
of statistical significance respectively. (d) ‡ denotes base group in estimation; (e) denotes n/a in estimation; (f) the sample
size are different relative to Table 6.11 given missing values on the included explanatory variables; (g) district level
dummies were not included in column (3) specification because of a high correlation with the included crops variables.