Targeting the Poor in Times of Crisis:

The Indonesian Health Card

Robert Sparrow

Institute of Social Studies

The Hague, The Netherlands

October 2006

Correspondence: Robert Sparrow, Institute of Social Studies, P.O Box 29776, 2502 LT

The Hague, The Netherlands. E-mail: sparrow@iss.nl.

Key words: Access, economic crisis, Indonesia, outpatient care, poverty, price subsidy,

targeting.

Abstract

This paper looks at targeting performance of the Indonesian health card programme that

was implemented in August 1998 to protect access to health care for the poor during the

Indonesian economic crisis. By February 1999, 22 million people had received a health

card. The health card provided a user fee waiver for public health care. Targeting of the

health card was pro-poor, but with considerable leakage to the non-poor. Utilisation of

the health card for outpatient care was also pro-poor, but conditional on ownership, the

middle quintiles were more likely to use the card.

Targeting of the health card followed a decentralised design combining

geographic targeting with community based targeting. This design facilitated the rapid

implementation of the programme, but targeting performance suffered from inadequate

information on the regional impact of the crisis, while at local level not all barriers to

accessing health care services were overcome by the health card. Indirect and direct costs

of seeking health care seem to be the main deterrent to using the health card, which are

higher in more remote areas.

Micro-simulations show that geographic targeting can contribute considerably to

improving targeting performance, but most of the targeting gains are to be made at the

community level.

This study highlights the need for adequate and up to date social welfare

indicators. In addition, further research would need to focus on how local knowledge can

be utilised for signalling poverty dynamics and local barriers to access.

Acknowledgements

I thank Menno Pradhan, Jan Willem Gunning, Marleen Dekker and Sudarno Sumarto for

helpful comments. A large part of this work was done while I was at the Vrije

Universiteit and Tinbergen Institute in Amsterdam, The Netherlands. Support from the

Netherlands Foundation for the Advancement of Tropical Research (NWO/WOTRO) is

gratefully acknowledged. All errors remain my own.

1

1. Introduction

In an attempt to protect access to health care utilisation for the poor during the Indonesian

economic crisis, a nationwide health programme was introduced in August 1998, as part

of the larger Indonesian Social Safety Net – Jaring Pengaman Sosial (JPS). This health

care programme included a targeted price subsidy that operated through the so called

health card – Kartu Sehat. Households that were thought to be most vulnerable to

economic shocks were allocated health cards, which entitled all household members to

the price subsidy at public health care providers. The programme followed a partly

decentralised targeting process, involving both geographic and community based

individual targeting.

The success of such crisis interventions critically depends on the ability to

identify and reach the poor, in particular those that are most vulnerable to the effects of a

crisis. Successful targeting requires information on welfare and crisis impact for

individual households. Typically, collecting such disaggregated information centrally is

costly. The administrative capacity for providing welfare details for each household (for

example, a centralised tax administration) is often not available in developing countries

like Indonesia. Moreover, short term information regarding the crisis effects for

individual households would be hard to retrieve even under a highly developed

administrative system. For example, in case of the Indonesian crisis, Skoufias et al. (2000)

find evidence of considerable movement in and out of poverty from 1997 to 1998,

hindering accurate targeting of the poor.

The decentralised design of the JPS programmes is meant to deal with this

targeting problem. The combination of multi-level geographic and community based

2

targeting provides an alternative infrastructure for gathering and processing information

locally, and disseminating this to higher administrative levels. Several authors have

argued that a decentralised design can benefit from local knowledge and community

participation, on the premise that local communities are more capable of identifying the

poor.1 Not only do local communities have better access to information on targeting

criteria, they are also more able to prioritise amongst the set criteria or even formulate

new local criteria that better reflect the need for assistance.

However, decentralisation has its weaknesses. Recently, a number of theoretical

and empirical studies have investigated the implications and pitfalls of different aspects

of decentralisation (e.g. regional political or fiscal autonomy). A main concern is that the

benefits of using local knowledge are offset against the loss of control over the allocation

process. Decentralised programmes are prone to local elite capture and suffer from

classic principal-agent dilemmas (e.g. Bardhan and Mookherjee 2000 and 2005, Galasso

and Ravallion 2005). In a comprehensive review of the empirical literature on targeting

Coady et al. (2004) find that geographic and community based targeting perform above

average, but with a large variation between the individual projects.

This paper deals with the targeting of the JPS health programme, in light of the

decentralised design. The objective is to investigate how the programme has been

implemented and who were the beneficiaries of the health cards. Has health card

allocation been pro-poor and is there evidence of leakage or local capture of benefits by

the non-poor? Particular focus will be on the effectiveness of regional targeting policy in

contrast to local (within-district) targeting by the allocation committees. What factors

underlie the observed patterns of benefit incidence? Has the centre been able to identify

3

the regions hit hardest by the crisis? What determines targeting at local level? Are there

barriers unabridged by the health card? Finally, a micro-simulation exercise will

decompose overall benefit incidence into geographic and local targeting performance.

The next section sets the context of the programme and contains a detailed

description of the programme design. Section 3 proposes the methodology and describes

the data. The results are presented and discussed in section 4, while section 5 concludes.

2. Background

The Indonesian economic crisis

The Indonesian economic crisis was triggered by a financial crisis that hit Southeast Asia

mid 1997. In addition, eastern Indonesia, Java and Sumatra were struck by El Niño

related droughts in the second half of 1997, while Sumatra, Kalimantan and eastern

Indonesia suffered large forest fires. In 1998 real GDP decreased by roughly 14 percent

and poverty rates had increased dramatically.2 According to estimates by the Indonesian

Bureau of Statistics (BPS) the poverty headcount increased from 17.7 percent in 1996 to

23.5 percent in 1999. Alternative estimates of poverty during the crisis abound,

unambiguously showing a daunting increase in poverty.3

The crisis saw a sharp decrease in the utilisation of outpatient health care from

1997 to 1998. Outpatient utilisation rates for modern health care drop from 0.193 visits

per person per month to 0.142 visits, with utilisation of public care declining from 0.095

to 0.064.4 This declining utilisation of public services was concentrated with local public

health clinics. A mayor cause of the decline in public health care utilisation was a

shortage of drugs and supplies during the crisis and a subsequent quality deterioration of

4

public care (Waters et al. 2003, Frankenberg et al. 1999, Knowles et al. 1999). Knowles

and Marzolf (2003) report cutbacks in public health spending due to reduced government

revenues during the crisis. The lack of operational funds and shortage of drugs disrupted

services in public health care facilities in 1998. In 1999 public health care utilisation

showed slight signs of recovery, which was partly attributed to the health card

programme (Pradhan et al. forthcoming).

The JPS health card programme

At the onset of the Indonesian financial crisis an important concern was whether the

achievements made in the social sectors over the past decades could be sustained. The

Indonesian government, with help of donors, reacted swiftly by introducing a social

safety net intervention aimed at safeguarding real incomes and access to social services

for the poor.

A main component of the social safety net was the revitalisation of the health card

programme, which started in the fall of 1998.5 This card existed before the onset of the

crisis, but its use had been negligible. A health card entitled a household to free services

at public health care providers consisting of (1) outpatient and inpatient care, (2)

contraceptives for women of child bearing age, (3) pre-natal care and (4) assistance at

birth. This study is limited to outpatient health care utilisation. The public health care

providers where the health cards could be used received budgetary support. These grants

were meant to compensate for the expected demand due to the health card and maintain

quality of health care. However, there was a loose relationship between the utilisation of

the health card and the compensation that the health care providers received in return.

5

Compensation was allocated based on the estimated number of households eligible for

the health card programme rather than actual utilisation of the health cards. The

1998/1999 budget for JPS health grants to primary health centres (Puskesmas) and village

midwives (Bidan di desa) amounted to US $ 29 million, financed by the Government of

Indonesia and the Asian Development Bank.

The JPS health programme followed a decentralised design, with part of the

targeting and allocation process delegated to district administrations, villages officials

and public health care providers. The number of health cards and the amount of

compensation to health care providers was determined by a prosperity measure for 307

district (urban kota and rural kabupaten), which was provided by the National Family

Planning Coordinating Agency – Badan Koordinasi Keluarga Berencana Nasional

(BKKBN). This prosperity measure is a poverty headcount ratio, defining a household as

poor if it fails one or more of the following 5 basic needs criteria: (i) households can

worship according to faith, (ii) eat basic food twice a day, (iii) have different clothing for

school/work and home/leisure activities, (iv) have a floor that is made out of something

other than earth, and (v) have access to modern medical care for children or access to

modern contraceptive methods. The BKKBN collects this information nationwide on a

census basis. The BKKBN prosperity measure has been criticised to be an unsuitable

allocation criterion for the JPS, since its components are fairly inflexible and

inappropriate for measuring economic shocks or the impact of a crisis. However, at the

time of implementation it was the only up to date welfare measure at hand. The BKKBN

prosperity measure was also used as allocation rule for both the budgetary support to

facilities and health cards to households.

6

Targeting within districts followed a two stage process. Special district

committees allocated compensation funds to public health care providers according to the

estimated number of health card eligible households living in the area served by the

provider. The transfer size was not influenced by the actual services provided to health

card owners, but based on BKKBN headcounts in the village or sub-district. Health cards

were allocated to villages, again based on the BKKBN measure, and subsequently

distributed to eligible households through local health centres and village midwives.

Eligible households were those that were considered poor following the BKKBN

classification. In addition, targeting of households relied heavily on local knowledge by

granting local leaders to define additional criteria according to their own insights

regarding the effects of the crisis.

3. Methodology and data

Benefit incidence analysis and targeting performance

The analysis starts with a benefit incidence analysis of health card allocation and

utilisation, and then investigates the factors driving observed benefit incidence patterns

by focusing on the targeting instruments and barriers at three layers of the targeting

process: the geographic unit (inter-district targeting), the community (intra-district

targeting) and the household (utilisation). Finally, geographic targeting performance is

linked to individual heath card allocation and utilisation by simulating targeting regimes.

The benefit incidence analysis describes the coverage and concentration of receipt

and allocation of health cards, and how this varies by economic status. Coverage reflects

the percentage of the population participating in the programme, while the concentration

7

measure indicates how the benefits are distributed across the population. Monthly

household expenditure per capita is used as indictor for economic welfare, by defining

quintiles. Following BPS poverty estimates, the poorest quintile roughly corresponds to

the share of the population living below the poverty line.

Table 1 shows the different targeting phases, with the main actors, decisions and

actions, targeting instruments and main barriers. In the first targeting phase the district

BKKBN measure was the main targeting instrument of the central management unit in

Jakarta. The main question here is whether the BKKBN prosperity measure adequately

reflects the poverty profile and the geographic heterogeneity of the crisis impact.

For the second targeting phase, the determinants of within-district targeting are

scrutinised by means of multivariate analysis, at household level. A probit is estimated

for the probability of receiving a health card (hcjk) conditional on geographic targeting

( ) ( ) ( )0Pr0Pr1Pr 321

* >+++=>== jkkkjkjkjk BKKBNBPSXhchc εααα (1)

where Xjk reflects the characteristics and village conditions for a household j (in district k)

that may be considered as local targeting criteria. BKKBNk contains targeting criteria for

sub-districts and districts. Although it was no an official targeting criterion, the model

also control for the BPS district poverty headcount, as irregularities in allocation were

correlated with the average level of poverty. The error term εjk reflects unobserved factors

affecting targeting. The marginal effects can be interpreted as reflecting intra-district

targeting since the district targeting rules have been controlled for.6

8

Even if households receive a health card, there may still be barriers to using it,

such as lack of information, regional shortage of providers, or opportunity costs

unabridged by the health card. Such barriers are likely to vary by households and are

likely to be higher for the poor. To this end, the third stage of the targeting process is

considered through probit analysis of the determinants of health card utilisation (hcuijk)

conditional on health card ownership

( ) ( ) ( ) 1if0Pr0Pr1Pr 321

* =>+++=>== jkijkijkkijkijkijk hcvBPSXhcuhcu λβββ (2)

The unit of analysis is the individual. Xijk contains the same variables as in equation (1),

in addition to gender and age of individual i. Since the model is estimated for health card

recipients, and health card allocation was not random, consistency of the β estimates is

typically compromised by selection effects. Sample selection is controlled for by the

selection term λ = E(ε | hc = 1). Under the normality assumptions of a probit analysis, λ is

the inverse Mills ratio computed from the health card allocation probit estimates.7 To

facilitate identification, the BKKBN district and sub-district targeting variables are

excluded in equation (2). These regional BKKBN indicators were used for geographic

targeting but should not play a role in the individuals’ decision to use the health card.8

Micro-simulation

To separate geographic from local targeting performance, I follow Alatas and Pradhan

(2003) who propose a micro-simulation based approach to attribute changes in targeting

performance to specific targeting stages. The estimation results from (1) and (2) are used

9

to simulate targeting outcomes under different geographic targeting regimes. The actual

targeting performance will be compared to that under geographic targeting following the

BPS indicators and a uniform regime where allocation to district simply depends on

population size. This will show how local targeting performs on its own, and whether

there is scope for geographical targeting to improve pro-poor allocation of health cards.

The simulation exercise comes down to using the probit estimates for ranking

households within districts in terms of eligibility. Households are selected into the

programme according to their eligibility rank until the district quota is reached.

Geographic targeting regimes are then imposed by changing the number of households

selected for the programme in each district, keeping the overall size of the programme

constant. That is, the districts’ share in the programme is altered, not the total amount of

health cards.

In order to rank households within districts, the probit estimates are not sufficient;

the unobserved errors are also needed. To see this, consider equation (1): hc* can be

interpreted as latent programme eligibility, where the eligibility threshold of selection is

normalised to zero. This latent eligibility can be constructed from the probit’s linear

prediction and the unobserved error term. Therefore the errors are drawn from a truncated

standard normal distribution, such that 0~ˆˆˆ 321 >+++ jkkkjk BKKBNBPSX εααα if

household j received a health card, and the opposite if otherwise.9 A similar approach is

taken for utilisation: conditional on health card allocation to households under a specific

geographic targeting regime, an individual’s utilisation behaviour is evaluated according

to the linear prediction of the utilisation probit (2) and error ijkv~ . In this way it is possible

10

to track the effects of regional redistribution of health cards to household level allocation

of health cards, and then to individual utilisation of the health card for outpatient care.

Data sources

The key source of data is Indonesia's main socioeconomic survey (Susenas). The Susenas

is conducted annually on a national scale, collecting information on health care utilisation,

socioeconomic background of individuals and households, and detailed information on

household expenditures. In 1999 a special JPS module was included, which covered

202,089 households and 822,607 individuals. This module provides information on

household and individual participation in each of the JPS programmes. The Susenas

survey is fielded in February, so the JPS module only reflects programme coverage

during the initial 6 months of implementation. The first health cards were distributed in

the fourth quarter of 1998. By February 1999, the health card programme covered about

11 percent of the population.

Household characteristics include demographic information such as household

composition and the gender of the head of household. Besides per capita expenditure

quintile and the five factors that determine BKKBN prosperity status, alternative

reflections of socioeconomic status include the main source of income for the household,

education of the head of household and living conditions. Household specific shocks are

partly reflected by employment status of the head of household and a variable indicating

whether daily activities of a household member was disrupted due to illness in the past

month. The Susenas also provides some information on the village characteristics where

the household resides, such as rural/urban and the IDT village classification.10

11

A 1996 village census (Podes), provides pre-intervention data on the availability

of health care facilities in each village (desa) and urban precinct (kelurahan) in Indonesia.

The Podes includes 66,486 of these communities and can be merged with the Susenas

data. The village variables used in the analysis reflect access to health care: the number of

public health facilities located in the village, the number of doctors and midwives that

live in the village (per 1,000 inhabitants), and a variable indicating the village leader's

opinion on accessibility of health facilities in the village.11

Finally, I use data on the geographic targeting criteria and consumption based

poverty. This includes the percentage of BKKBN poor households in districts (December

1997) and in sub-districts (January 1999). The consumption based poverty headcount

ratios are computed by BPS based on the 1996 and 1999 Susenas.12

4. Results and Discussion

Benefit incidence

The health card programme was already of a substantial magnitude in February 1999

with 10.6 percent of Indonesians living in a household with a health card. Health card

recipients show a higher utilisation of outpatient care than non-recipients. The difference

is largest with utilisation of public care. Amongst health card recipients, 15.1 percent

visited an outpatient provider during the 3 months prior to the survey. For non-health

card recipients this is lower, at 12.9 percent. Although health card owners tend to choose

public providers more often, they do not always use their health card. 3.7 out of 10.4

percent of the health card owners report not to use the health card when seeking care at a

public provides.

12

What could explain the weak link between ownership and utilisation? Providers

were not reimbursed based on actual services provided, but on the predicted demand.

Possibly, the providers themselves selected who they deemed in need for subsidised

services and did not always honour the rule that those who could present a health card

should be provided free services. Alternatively, rich household may decide to forgo the

option of free health care, preferring the higher quality private facilities instead of the

public health care centre.

Strauss et al. (2004) show that at some public health clinics not all services were

covered by the health card, but that this can not fully explain the under-usage of health

cards. Qualitative research by Soelaksono et al. (1999) find that at some public facilities,

the time allocated to patients with a health card was limited, and that in remote areas the

lack of access to the nearest public facility was a possible deterrent to use the health card.

They also found indications that patients perceived the care received using a health card

to be of lower quality than services and medicines obtained when not using the health

card. In addition, the public perception was that treatment at the public clinic was less

effective than at private sector.

Health cards are distributed pro-poor, as shown in Table 2. Amongst the poorest

20 percent of the population 18.5 percent had a health card. Incidence of health card

receipt drop as per capita expenditure increases, from 13.7 percent in the second quintile

to 3.7 in the richest quintile. The allocation shares for ownership and utilisation are

presented in columns 3 and 5 of Table 2. The poorest 20 percent of the population own

33.7 percent of the health cards. Still, there is considerable leakage to the more wealthy

households. Considering that about 11 percent of the population received a health card,

13

perfect targeting would imply that all health cards are concentrated with the poorest

quintile. However, the wealthiest 60 percent of the population own about 40 percent of

the health cards.

Utilisation of health cards for outpatient care shows a less pro-poor distribution

than allocation. Conditional on having a health card, the middle quintiles are more likely

to use it. That means that those who received benefits were on average wealthier than

those who received the card. Barriers of access to health care seem higher for the poor.

Even though the non-poor are more likely to use their health card when they have one,

most of the benefits still accrue to the poor. This is because the initial distribution of the

health card is distributed pro-poor. Almost three quarters of the health cards is distributed

in rural areas. But relative to this distribution, the use of the cards is similar between rural

and urban areas.

Geographical targeting and crisis impact

How well did the district targeting criteria reflect regional differences in poverty and

impact of the crisis? Several studies have raised concern about the lack of reliable data

available for geographic targeting (e.g. Ananta and Siregar 1999, Daly and Fane 2002,

Dhanani and Islam 2002, Pritchett et al. 2002). Given the heterogeneous nature of the

crisis, it is likely that criteria for regional targeting misjudged the degree of poverty in the

districts, since only pre-crisis information on regional poverty was available. There are

two reasons for this. First, the crisis has given rise to large relative price changes,

between products (especially food) and across regions (Cameron 1999, Frankenberg et al.

2003, Friedman and Levinsohn 2002). This variation is completely ignored in the

14

targeting process when pre-crisis poverty estimates are applied as allocation rule in 1999.

Second, the effects of the crisis varied strongly between regions and were only weakly

correlated with the initial level of poverty (Sumarto et al. 1998). This heterogeneity of the

crisis impact is shown in Figure 1, which plots 1996 consumption based poverty against

the change from 1996 to 1999. The difference between 1996 and 1999 estimates reflects

the impact of the crisis. It indicates the absolute change in the fraction of people that

moved into or out of poverty during the crisis. In line with Sumarto et al. (1998), there

appears to be no correlation between the initial level of poverty and the impact of the

crisis.

The BKKBN data are collected at more frequent intervals than the consumption

surveys. They can provide fairly up to date information, as far down as the household

level. The problem, however, is that the components of the BKKBN classification are

inflexible measures and inappropriate for capturing the degree of poverty when faced

with severe economic shocks of a crisis.13

Table 3 is illustrative for the difficulty of capturing the effect of the crisis using

pre-crisis data. The table shows the ranking of provinces (from low to high) according to

the 1999 (BPS99) and 1996 (BPS96) poverty headcount of BPS, the BKKBN measure.

The different welfare measures show different levels of poverty. As expected, the

consumption based poverty headcount estimates for 1996 are lower than for 1999.

Evaluating welfare by the basic needs criteria of the BKKBN yields a higher count of

deprived households. In itself this is not surprising. What is important is that the ranking

is different. The ranking following the BKKBN measure differs from both the levels and

changes of poverty, as measured by BPS.

15

The differences between the welfare measures are further illustrated by a

graphical exposition. Figure 2 and Figure 3 plot the BKKBN targeting rule against the

poverty BPS estimates. The BKKBN prosperity score is strongly positively correlated

with the 1999 poverty headcount, but with a lot of variation around the trend. This is

expected since the BKKBN criteria are not solely based on household consumption.

There is a weakly positive correlation between the change in poverty and the BKKBN

measure for the main body of districts and a greater variation amount the mean. The trend

line is pulled up by a small number of districts that experienced a large increase in

poverty.

Actual allocation of health cards is less correlated with expenditure based poverty

than the BKKBN criteria are. There seems to be little correlation between BPS poverty

and the allocation of health cards reported in the Susenas data (Figure 4). This indicates a

fuzzy relationship between the BKKBN targeting criteria and actual allocation. This

could be due to delays in the implementation or regional difference in efficiency of the

programme. But if it is, then these irregularities are greater in relatively poor districts.

Community based targeting and health card utilisation

What are the key factors that determine targeting at the community level? Table 4 reports

probit estimates of the determinants for health card allocation as well as utilisation

conditional on ownership, at household level. The table reports marginal effects. The

standard errors have been adjusted for the stratified sampling design of the Susenas

survey.

16

There are significant negative effects of per capita consumption on the probability

of receiving a health card, confirming the pro-poor targeting found in the benefit

incidence analysis. For the use of the health card conditional upon owning one, the results

do not reflect the slightly non-poor pattern found in the benefit incidence. It could be that

observed utilisation differences between quintiles are too small to yield statistically

significant marginal effects in the multivariate analysis. Alternatively, it could be that the

non-poor bias in utilisation is not a direct wealth effect but follows from other

background characteristics, which are correlated with wealth.

Sector of employment affects both allocation and utilisation. The probability of

receiving a health card is lower for households for whom agriculture is the main source

of income, while they are also less likely to use it for outpatient care. This may indicate

that the opportunity costs of time spent at the health clinic or travelling are relatively

higher for farm households.

Supply and access of health care at village level plays an important role in the

targeting process. The number of auxiliary public clinics negatively affects the

probability of receiving a health card. But conditional on ownership the presence of

primary and auxiliary public clinics in the village strongly increases the use of health

cards for outpatient care. Further, utilisation of health cards is higher in villages where

the village leader views health care facilities to be easy or very easy to reach. Overall, the

results suggest that, while remote and less wealthy areas with little access to health care

receive priority in the targeting process, the direct and indirect costs of using the cards are

relatively high. On the other hand, the probability of selection increases as the number of

midwives living in the village increases. In addition, there is also a positive correlation

17

with utilisation. Since it is the medical staff of local clinics that actually distribute the

health cards to households, this might reflect the importance of informal contacts within

the village for awareness of, and participation in, social programmes.

The probit results confirm that health cards have been awarded to households

based on health status. The official allocation rules require health cards to be distributed

to the poor, irrespective of their health status. But the clearly positive effect on the

variable measuring whether any household member has its daily activities disrupted

through illness indicates that often health cards were given based on acute need. For

those who fall ill and do not own a health card, it is still possible to get a health card after

seeking medical care (Soelaksono et al. 1999).

Turning to household composition, the results show that (conditional on per capita

consumption) households with a relatively large share of children and elderly have a

higher probability of receiving and using a health card. Conditional on household

composition, larger households are less likely of receiving and using a health card.

Household headed by females have a significantly higher chance of receiving a health

card. There is a negative correlation between the education of the head of household and

the probability of receiving a health card. But controlling for other household variables,

the characteristics of the head of household seem less important for utilisation.

Women tend to use the cards more for outpatient care than men do. The outpatient

utilisation variable does not reflect the use of health cards for contraception and family

planning services. Nevertheless, it could be that the availability of these services under

the health card has raised awareness of its usefulness amongst women. Utilisation shows

a U-shaped age pattern as the young and the old use the card relatively more.

18

The results confirm that the BKKBN prosperity status variables have influenced

health card allocation. An increase in the district and sub-district BKKBN basic needs

measure increases the probability of receiving a health card. Amongst the individual

BKKBN criteria floor material is a strong predictor for health card ownership. Those with

an earth floor have a higher chance of receiving a health card, but a lower probability of

actually using it for outpatient care. Being able to worship according to faith decreases

the probability of receiving a health card. Interestingly, the individual’s perception of

freedom in practicing personal faith is positively correlation with utilisation of the health

card. Owning different sets of clothing for work and leisure decreases the probability of

using a health card. In contrast to the other BKKBN indicators, having access to modern

care does not yield the expected effect, as it increases both the chance of receiving a

health card and, conditional on ownership, the probability of using it. This suggests the

presence of two countervailing effects: accessibility to public providers increases

exposure to the programme, which outweighs the official targeting criterion.

Simulation results

The simulations are presented in Table 5. The table shows the actual outcome for the

sample used in the multivariate analysis and three counterfactual outcomes: geographic

targeting following BPS and BKKBN indicators, and uniform targeting. The table shows

concentration of health cards and utilisation, to make the performance of different

targeting regimes comparable.14

Not surprising, the actual outcome for the sample corresponds quite closely to a

geographical targeting regime that would follow the BKKBN rule strictly. However, the

19

BPS indicator would increase pro-poor targeting compared to the actual regime, as this

would redistribute 2.7 percent of the health cards to the poorest quintile. This may seem a

small change, but in terms of population it reflects roughly 600,000 people in the poorest

quintile that would move into the programme due to improved geographic targeting.

The uniform targeting regime shows the counterfactual outcome if a districts

share of health cards would be proportional to the relative population size. That is, if

purposive targeting would only occur within the districts. Compared to uniform

geographic targeting, the actual targeting regime increases the share of the poorest

quintile by 5.4 percentage points, from 28.9 to 34.1 percent. Considering that complete

random targeting at all levels would yield coverage of 20 percent per quintile, local

targeting alone increases health card concentration with the poorest by 8.8 percentage

points. Hence, for the poorest quintile the overall gain from decentralised targeting is an

additional 14.2 percent of the health card programme, 38 percent (i.e. 5.4 percentage

point) of which is due to geographic targeting and 62 percent (i.e. 8.8 percentage point) to

local targeting. The BPS targeting rule would increase this even further to a gain 16.9

percent, of which 48 percent (i.e. 8.1 percentage point) would be due to geographic

targeting.

The simulations for health card utilisation show very similar patterns, indicating

that improved targeting of health cards leads to a proportional improvement in targeting

of the benefits from the health card.

The results from the simulation exercise suggest that there is indeed scope for

geographic targeting. Under the BPS targeting rule almost half of the gain from targeting

would be due to geographic targeting. However, the results also show the limits to

20

geographic targeting. Coverage amongst the poorest quintile under the best performing

scenario is still only 37 percent, while perfect targeting would imply 53 percent coverage

(i.e. 10.6 percent / 20 percent). For this difference to be bridged community based

targeting would have to be improved. Moreover, barriers to access seem to be locally

determined and not much affected by geographic targeting.

5. Conclusion

There is clear evidence that the JPS health card programme was pro-poor in the sense that

the poor had a higher probability of receiving a health card and using it to obtain free

health services. However, despite pro-poor targeting, a considerable number of health

cards went to households in the richer quintiles.

The programme was implemented at the remarkable speed: by February 1999

approximately 22 million people (about 11 percent of Indonesians) lived in households

that received a health card. The decentralised programme design may well have

facilitated this swift reaction, by relying on existing administrative and operational

infrastructure within the districts. However, at such short notice there was no reliable data

on the impact of the crisis across districts. Geographic targeting criteria were therefore

based poverty estimates that reflect the actual level of poverty to some extent but do not

capture the impact of the crisis. There appears to be no correlation between the initial

level of poverty and the impact of the crisis.

A notable finding is that some health card owners did not use their health card

when obtaining care from public service providers. The particular design resulted in a

discrepancy between health card ownership and utilisation. Moreover, utilisation of

21

subsidised services is less pro-poor than ownership. Conditional on ownership, the

middle quintiles have a higher propensity to use their health card, suggesting that access

barriers to health care are not fully overcome by a user fee waiver. The main deterrent

seems to be the opportunity costs of seeking health care. The direct and indirect costs of

using the health card are relatively higher in the more remote, and rural villages with little

access to public health care providers. While the more remote areas were targeted

because of the lack of access to health care facilities, it is for the same reason that usage

rates are low.

Micro-simulations show that geographic targeting can contribute considerably to

improving targeting performance. Nevertheless, most of the targeting gains are to be

made at the community level.

In terms of policy implications for future safety nets and crisis responses, the need

for adequate information systems is apparent from this study. While the decentralised

design of the programme seems flexible in a crisis situation, up to date information and

early signalling of crisis effects are crucial. In addition, further research would need to

investigate how local information can best be exploited under decentralised and

community based targeting in signalling crisis related poverty dynamics and local

barriers to access.

22

Endnotes

1 See, for example, Alderman (2001 and 2002). Conning and Kevane (2002)

provide an extensive review of community based targeting.

2 For an account of the economic crisis in Indonesia see, amongst others, Cameron

(1999), Smith et al. (2002) and Frankenberg et al. (2003).

3 Using different data sources, Suryahadi et al. (2003) trace poverty movement

during the crisis and find that the poverty headcount has more than doubled from15.3

percent in February 1997 to 33.2 percent at the height of the crisis, late 1998.

4 Modern health care is here defined as public health care providers – hospitals,

health clinics (Puskesmas), village maternity posts (Polindes) and integrated health posts

(Posyandu) – and private providers – hospitals, doctors, clinics and paramedical services.

Traditional health care is not included.

5 The JPS further included an education, food security, labour creation and micro

credit programmes. Ananta and Siregar (1999) and Daly and Fane (2002) provide a good

overview of all the JPS programmes.

6 An alternative specification would be to include the observed health card coverage

in districts. However, this variable would be endogenous to the outcome variable.

7 Since the utilisation probit (2) is estimated for individuals, so is the first stage

allocation probit that is used for computing the selection term. The variables are the same

as those included in the household level analysis in (1) and the estimation results are very

similar. The results are available upon request.

23

8 Nevertheless, there may be reasons why the BKKBN indicators could still be

correlated with utilisation. The BKKBN indicators determined the amount of JPS

financial compensation to health facilities, and thus health care quality and supply.

Moreover, they might be correlated with local level of welfare (or deprivation), and thus

with health care demand. I therefore estimated equation (2) with the BKKBN variables

(without λ). Their marginal effects were small and statistically non significant. Although

this does not constitute a formal test, it is a strong indication that the exclusion restriction

is justified. The results are available upon request.

9 The errors are computed as ( )[ ])()()(~ 1 LLHu Φ+Φ−Φ⋅Φ= −ε where H and L are

the upper and lower truncation points, respectively. Ф reflects the standard normal cdf

and u is a drawn from a uniform distribution with a value between 0 and 1.

10 IDT refers to the Inpres Desa Tertinggal program, an anti-poverty programme for

economically less developed villages. For this program, each village or urban precinct in

Indonesia has been classified as either developed or less developed.

11 The Podes survey asks village leaders whether the closest public health clinics are

(i) very easy, (ii) easy, (iii) difficult or (iv) very difficult to reach by the majority of the

village population.

12 See BPS (2000) for details.

13 The main criticism in this respect is that the BKKBN measure is based on fixed

assets (type of floor and owning clothes) and non-economic questions regarding religious

practices. Sumarto et al. (2003) place further questions with interregional consistency of

the BKKBN measure as the village staff who collect the BKKBN data receive relatively

little training, and the figures are vulnerable to manipulation by local government

24

officials. Using data from a longitudinal survey in 100 villages, Suryahadi et al. (1999)

show that there is a high degree of mismatch between the BKKBN classification and

expenditure based poverty measures. For example, the BKKBN data classify 49 percent

of the households in the sample as poor. But according to per capita consumption, only

57 percent of these households rank with the poorest 49 percent of the population.

14 Some observations were lost due to a few missing values in the covariates, and

merging the Susenas and Podes data. The simulated programme sizes therefore differ

slightly from the actual observed size. Overall incidence varies between 10.7 and 10.9

percent, slightly above the observed 10.6 percent reported in Table 2. Comparing

coverage results could then be misleading if changes for quintiles are due to change in

programme size instead of redistribution. With the concentration measures the overall

size of the programme is normalised to 100 percent for all counterfactual regimes. The

coverage results are not reported here but are available upon request.

25

References

Alatas V and Pradhan MP. 2003. Geographic Targeting in Indonesia. World Bank,

Jakarta. Unpublished document.

Alderman H. 2001. Multi-Tier Targeting of Social Assistance: The Role of

Intergovernmental Transfers. The World Bank Economic Review 15: 33-53.

Alderman H. 2002. Do Local Officials Know Something We Don't? Decentralization of

Targeted Transfers in Albania. Journal of Public Economics 83: 375-404.

Ananta A and Siregar R. 1999. Social Safety Net Policies in Indonesia: Objectives and

Shortcomings. ASEAN Economic Bulletin 16: 344-359.

Bardhan P and Mookherjee D. 2000. Capture and Governance at Local and National

Levels. American Economic Review 90: 135-139.

Bardhan P and Mookherjee D. 2005. Decentralizing Antipoverty Program Delivery in

Developing Countries. Journal of Public Economics 89: 675-704.

BPS. 2000. Tingkat Kemiskinan Kabupaten. Jakarta. Unpublished document.

Cameron LA. 1999. Survey of Recent Developments. Bulletin of Indonesian Economic

Studies 35: 2-40.

Coady D, Grosh ME, and Hoddinott J. 2004. The Targeting of Transfers in Developing

Countries: Review of Experiences and Lessons. Washington, D.C: World Bank.

Conning J and Kevane M. 2002. Community-Based Targeting Mechanisms for Social

Safety Nets: A Critical Review. World Development 30: 375-394.

Daly A and Fane G. 2002. Anti-Poverty Programs in Indonesia. Bulletin of Indonesian

Economic Studies 38: 309-329.

26

Dhanani S and Islam I. 2002. Poverty, Vulnerability and Social Protection in a Period of

Crisis: The Case of Indonesia. World Development 30: 1211-1231.

Frankenberg E, Thomas D, and Beegle K. 1999. The Real Costs of Indonesia's Economic

Crisis: Preliminary Findings from the Indonesian Family Life Survey. Labor and

Population Program Working Paper 99-04, RAND, Santa Monica.

Frankenberg E, Smith JP, and Thomas D. 2003. Economic Shocks, Wealth, and Welfare.

Journal of Human Resources 38: 280-321.

Friedman J and Levinsohn J. 2002. The Distributional Impact of Indonesia's Financial

Crisis on Household Welfare: A "Rapid Response" Methodology. The World Bank

Economic Review 16: 397-423.

Galasso E and Ravallion M. 2005. Decentralized Targeting of an Antipoverty Program.

Journal of Public Economics 89: 705-727.

Knowles JC, Pernia EM, and Racelis M. 1999. Social Consequences of the Financial

Crisis in Asia. Economic Staff Paper Number 60, Asian Development Bank, Manila.

Knowles JC and Marzolf JR. 2003. Health Financing for the Poor in Indonesia. World

Bank, Washington D.C. Unpublished document.

Lanjouw P, Pradhan MP, Saadah F, Sayed H, and Sparrow R. 2002. Poverty, Education

and Health in Indonesia: Who Benefits from Public Spending? In: Morrisson C (ed).

Education and Health Expenditures, and Development: The cases of Indonesia and

Peru. Paris: Development Centre Studies, OECD Development Centre, pp. 17-78.

Pradhan MP, Saadah F, and Sparrow R. Forthcoming. Did the Health Card Program

Ensure Access to medical Care for The Poor During Indonesia's Economic Crisis?

The World Bank Economic Review.

27

Pritchett L, Sumarto S, and Suryahadi A. 2002. Targeted Programs in an Economic Crisis:

Emperical Findings from the Experience of Indonesia. SMERU working paper,

Jakarta.

Skoufias E, Suryahadi A, and Sumarto S. 2000. Changes in Household Welfare, Poverty

and Inequality during the Crisis. Bulletin of Indonesian Economic Studies 36: 97-

114.

Smith JP, Thomas D, Frankenberg E, Beegle K, and Teruel G. 2002. Wages, employment

and economic shocks: Evidence from Indonesia. Journal of Population Economics

15: 161-193.

Soelaksono B, Budiyati S, Hastuti et al. 1999. The Impact of the Crisis on the People's

Behavior and Perceptions of the Use and Effectiveness of Puskesmas, Posyandu,

and the Role of Midwifes. SMERU, Jakarta. Unpublished document.

Strauss J, Beegle K, Dwiyanto A et al. 2004. Indonesian Living Standards: Before and

After the Financial Crisis. Santa Monica: RAND.

Sumarto S, Wetterberg A, and Pritchett L. 1998. The Social Impact of the Crisis in

Indonesia: Results from a Nationwide Kecamatan Survey. SMERU working paper,

Jakarta.

Sumarto S, Suryahadi A, and Pritchett L. 2003. Safety Net or Safety Ropes? Dynamic

Benefit Incidence of Two Crisis Programs in Indonesia. World Development 31:

1257-1277.

Suryahadi A, Sumarto S and Pritchett L. 2003. The Evolution of Poverty During the

Crisis in Indonesia. Asian Economic Journal 17:221-241.

28

Suryahadi A, Suharso Y, and Sumarto S. 1999. Coverage and Targeting in the Indonesian

Social Safety Net Programs: Evidence from 100 Village Survey. SMERU working

paper, Jakarta.

Waters H, Saadah F, and Pradhan MP. 2003. The Impact of the 1997-98 East Asian

Economic Crisis on Health and Health Care in Indonesia. Health Policy and

Planning 18: 172-181.

29

Tables

Table 1 Targeting process of the JPS health card programme

Targeting phase

Actors

Objectives/decisions

Targeting instrument

Barriers

Phase 1:

CMU, Jakarta

Target districts

- BKKBN 1997

- No crisis information

- Welfare indicator quality

Phase 2:

District committees

Target public health care

providers and households

- Local BKKBN data

- Local knowledge

- Lack of crisis information of in

sub-districts, villages and

individual households

- Elite capture

Phase 3:

Households

Self targeting: utilisation

of health cards

- Self selection

- Needs

- Lack of access

- Perceptions

- Crowding out

30

Table 2 Distribution of health card allocation and utilisation for outpatient care during past three months, 1999

Health card allocation

(amongst Indonesian population)

Health card utilisation

(am

ongst health card recipients)

Incidence (%)

Share (%

) Incidence (%)

Share (%

)

Quintile 1 (poorest)

18.45

33.74

6.51

33.28

Quintile 2

13.71

25.68

6.54

25.43

Quintile 3

10.61

20.06

6.84

20.79

Quintile 4

7.09

13.41

6.78

13.77

Quintile 5 (richest)

3.71

7.10

6.25

6.72

Urban

7.23

26.79

6.65

26.89

Rural

12.82

73.21

6.62

73.11

All

10.62

100.00

6.63

100.00

Number of observations: 822,607.

31

Table 3 provinces ranked by poverty measures according to BPS and BKKBN headcount

Province BKKBN 1997 BPS 1996 BPS 1999 ∆BPS 99-96

% (rank) % (rank) % (rank) % (rank)

Aceh 51.65 (18) 12.72 (8) 14.75 (6) 2.03 (11)

North-Sumatra 27.56 (3) 13.23 (9) 16.74 (8) 3.51 (14)

West-Sumatra 33.89 (7) 9.84 (5) 13.24 (3) 3.40 (13)

Riau 35.97 (10) 12.62 (7) 14.00 (4) 1.38 (5)

Jambi 29.67 (4) 14.84 (11) 26.64 (17) 11.80 (25)

South-Sumatra 41.27 (12) 15.89 (12) 23.53 (14) 7.64 (20)

Bengkulu 43.22 (16) 16.71 (14) 19.79 (12) 3.08 (12)

Lampung 60.72 (22) 25.59 (21) 29.11 (20) 3.52 (15)

Jakarta 17.32 (2) 2.35 (1) 3.99 (1) 1.64 (9)

West-Java 33.25 (6) 11.06 (6) 19.78 (11) 8.72 (23)

Central-Java 54.30 (19) 21.61 (17) 28.46 (18) 6.85 (18)

Yogyakarta 34.57 (9) 18.43 (16) 26.11 (15) 7.68 (21)

East-Java 42.24 (14) 22.13 (18) 29.48 (21) 7.35 (19)

Bali 0.00 (1) 7.81 (2) 8.53 (2) 0.72 (3)

NTB 62.49 (23) 31.97 (23) 32.95 (23) 0.98 (4)

NTT 82.68 (26) 38.89 (24) 46.73 (25) 7.84 (22)

West-Kalimantan 46.82 (17) 24.21 (20) 26.18 (16) 1.97 (10)

Central-Kalimantan 42.71 (15) 13.50 (10) 15.05 (7) 1.55 (6)

South-Kalimantan 34.03 (8) 8.53 (3) 14.37 (5) 5.84 (16)

East-Kalimantan 32.14 (5) 9.73 (4) 20.16 (13) 10.43 (24)

North-Sulawesi 39.90 (11) 17.94 (15) 18.19 (9) 0.25 (1)

Central-Sulawesi 56.84 (20) 22.30 (19) 28.68 (19) 6.38 (17)

South-Sulawesi 41.53 (13) 16.71 (13) 18.32 (10) 1.61 (8)

South-East-Sulawesi 64.09 (24) 29.23 (22) 29.51 (22) 0.28 (2)

Maluku 60.21 (21) 44.56 (26) 46.14 (24) 1.58 (7)

Irian-Jaya 70.40 (25) 42.28 (25) 54.75 (26) 12.47 (26)

Indonesia 41.98 17.70 23.43 5.73

32

Table 4 Within-district targeting of JPS health cards to households, and determinants of

utilisation for outpatient treatment (probit marginal effects)

(1) (2)

Health card

allocation

Health card

utilisation

Unit of analysis Household Individual

Female 0.0172

[0.0017]**

Age -0.0021

[0.0002]**

Age squared 0.00004

[0.0000]**

Female head of household 0.0197 -0.0020

[0.0029]** [0.0045]

Education head of household

(reference = none / primary not completed)

Primary -0.0075 0.0078

[0.0019]** [0.0034]*

Junior secondary -0.0241 0.0002

[0.0023]** [0.0056]

Senior secondary -0.0407 0.0077

[0.0023]** [0.0079]

Higher -0.0474 -0.0031

[0.0035]** [0.0148]

Head of household unemployed 0.0152 -0.0055

[0.0086]+ [0.0112]

Member of household ill last month 0.0119 0.0821

[0.0019]** [0.0040]**

Log household size -0.0082 -0.0424

[0.0022]** [0.0047]**

33

(Table 4 continued)

(1) (2)

Household composition

(reference = share of males age 18-60)

Share of males age < 6 0.0313 0.0397

[0.0074]** [0.0141]**

Share of females age < 6 0.0269 0.0282

[0.0078]** [0.0145]+

Share of males age 6-12 0.0014 0.0401

[0.0063] [0.0122]**

Share of females age 6-12 0.0116 0.0213

[0.0065]+ [0.0124]+

Share of males age 13-17 0.0119 0.0020

[0.0069]+ [0.0143]

Share of females age 13-17 -0.0062 -0.0115

[0.0071] [0.0156]

Share of females age 18-60 0.0018 0.0044

[0.0057] [0.0130]

Share of males age > 60 0.0150 -0.0120

[0.0059]* [0.0125]

Share of females age > 60 0.0102 -0.0362

[0.0059]+ [0.0139]**

Agriculture main source of income -0.0090 -0.0056

[0.0021]** [0.0034]+

Per capita expenditure quintile

(reference = quintile 1, poorest)

Quintile 2 -0.0150 0.0009

[0.0024]** [0.0043]

Quintile 3 -0.0255 -0.0028

[0.0026]** [0.0050]

34

(Table 4 continued)

(1) (2)

Quintile 4 -0.0418 -0.0058

[0.0027]** [0.0066]

Quintile 5, richest -0.0657 -0.0112

[0.0029]** [0.0092]

BKKBN criteria

Can worship according to faith -0.0191 0.0140

[0.0041]** [0.0045]**

Eat basic food twice a day -0.0059 0.0108

[0.0079] [0.0086]

Different clothing school/work/leisure -0.0023 -0.0215

[0.0052] [0.0083]**

Floor made of material other than earth -0.0548 0.0123

[0.0033]** [0.0062]*

Access to modern medical care 0.0291 0.0337

[0.0025]** [0.0039]**

Living conditions

House made out of bamboo 0.0419 0.0031

[0.0036]** [0.0052]

Access to clean drinking water -0.0077 0.0106

[0.0029]** [0.0053]*

Closed sewer -0.0229 -0.0024

[0.0025]** [0.0047]

Access to electricity 0.0183 -0.0149

[0.0029]** [0.0056]**

Village characteristics

Rural area -0.0154 -0.0084

[0.0039]** [0.0061]

IDT village -0.0061 0.0083

[0.0035]+ [0.0046]+

35

(Table 4 continued)

(1) (2)

Nr. of primary health clinics -0.0058 0.0111

[0.0031]+ [0.0049]*

Nr. of auxiliary health clinics -0.0093 0.0159

[0.0026]** [0.0039]**

Nr. of doctors per 1,000 inhabitants 0.0067 0.0060

[0.0045] [0.0054]

Nr. of midwifes per 1,000 inhabitants 0.0066 0.0225

[0.0037]+ [0.0071]**

Health facilities easy to reach 0.0080 0.0139

[0.0062] [0.0070]*

Majority of inter village traffic by land 0.0212 0.0053

[0.0070]** [0.0116]

BKKBN poverty rate in sub-district 0.0982

[0.0075]**

BKKBN poverty rate in district 0.0178

[0.0098]+

BPS poverty rate 1999 -0.0008 0.0007

[0.0001]** [0.0002]**

Selection term -0.0070

[0.0166]

Observations 185,608 76,684

Pseudo R-squared 0.10 0.09

Robust standard errors in brackets

+ significant at 10%; * significant at 5%; ** significant at 1%

36

Table 5 Health card allocation and outpatient utilisation simulations under different geographic targeting regimes (percentage shares)

Health card allocation

(% share of total programme)

Health card outpatient utilisation

(% share of total utilisation)

Targeting regime

Current

BPS

BKKBN

Uniform

Current

BPS

BKKBN

Uniform

Quintile 1 (poorest)

34.21

36.90

33.43

28.84

33.43

37.11

33.94

28.11

Quintile 2

25.71

24.95

24.98

23.87

25.81

25.07

25.14

23.45

Quintile 3

19.76

18.91

19.50

20.51

19.94

17.97

18.45

20.61

Quintile 4

13.38

13.07

14.39

16.38

13.96

13.23

14.80

16.88

Quintile 5 (richest)

6.93

6.17

7.70

10.4

6.85

6.63

7.67

10.94

Urban

27.06

25.41

26.01

36.41

25.74

23.33

23.26

35.56

Rural

72.94

74.59

73.99

63.59

74.26

76.67

76.74

64.44

All

100.00

100.00

100.00

100.00

100.00

100.00

100.00

100.00

37

Figures

Lowess smoother, bandwidth = .8

Change BPS 1996-1999

BPS poverty rate 19960 .7089

-.0494

.2759

Figure 1 Correlation between initial poverty in 1996 and crisis impact on BPS poverty

headcount (locally weighted regression with 0.8 bandwidth)

38

Lowess smoother, bandwidth = .8BKKBN headcount

BPS poverty rate 1999.0109 .8009

0

.9655

Figure 2 Correlation between BKKBN headcount December 1997 and BPS 1999 poverty

rate (locally weighted regression with 0.8 bandwidth)

39

Lowess smoother, bandwidth = .8BKKBN headcount

Change BPS 1996-1999-.0494 .2759

0

.9655

Figure 3 Correlation between BKKBN headcount December 1997 and crisis impact on

BPS poverty rate (locally weighted regression with 0.8 bandwidth)

40

Lowess smoother, bandwidth = .8Health card coverage

BPS poverty rate 1999.0109 .8009

0

.517606

Figure 4 Correlation between health card coverage and BPS 1999 poverty rate (locally

weighted regression with 0.8 bandwidth)

i

Supplemental appendix

Table A.1. Within-district targeting of JPS health cards to households, and determinants of utilisation for outpatient treatment (probit

marginal effects)

(1)

(2)

(3)

(4)

(5)

Health card

Health card

(selection probit)

Utilisation

(selection term)

Utilisation

(no selection term)

Utilisation

(no exclusions)

Unit of analysis

Household

Individual

Individual

Individual

Individual

Fem

ale

0.0172

0.0172

0.0172

[0.0017]**

[0.0017]**

[0.0017]**

Age

-0.0021

-0.0021

-0.0021

[0.0002]**

[0.0002]**

[0.0002]**

Age squared

0.0000

0.0000

0.0000

[0.0000]**

[0.0000]**

[0.0000]**

Fem

ale head of household

0.0197

0.0185

-0.0020

-0.0015

-0.0015

[0.0029]**

[0.0031]**

[0.0045]

[0.0043]

[0.0043]

Education head of household

(reference = none / primary not completed)

Primary

-0.0075

-0.0052

0.0078

0.0077

0.0076

[0.0019]**

[0.0021]*

[0.0034]*

[0.0033]*

[0.0033]*

Junior secondary

-0.0241

-0.0219

0.0002

-0.0007

-0.0006

[0.0023]**

[0.0026]**

[0.0056]

[0.0050]

[0.0050]

Senior secondary

-0.0407

-0.0400

0.0077

0.0057

0.0056

[0.0023]**

[0.0024]**

[0.0079]

[0.0060]

[0.0060]

Higher

-0.0474

-0.0503

-0.0031

-0.0058

-0.0059

[0.0035]**

[0.0033]**

[0.0148]

[0.0125]

[0.0124]

Head of household unem

ployed

0.0152

0.0152

-0.0055

-0.0051

-0.0048

[0.0086]+

[0.0094]

[0.0112]

[0.0112]

[0.0112]

Mem

ber of household ill last month

0.0119

0.0125

0.0821

0.0826

0.0825

[0.0019]**

[0.0021]**

[0.0040]**

[0.0039]**

[0.0039]**

Log household size

-0.0082

-0.0129

-0.0424

-0.0429

-0.0429

[0.0022]**

[0.0026]**

[0.0047]**

[0.0044]**

[0.0044]**

ii

(Table A.1. continued)

(1)

(2)

(3)

(4)

(5)

Household composition

(reference = share of males age 18-60)

Share of males age < 6

0.0313

0.0298

0.0397

0.0408

0.0407

[0.0074]**

[0.0082]**

[0.0141]**

[0.0140]**

[0.0140]**

Share of females age < 6

0.0269

0.0234

0.0282

0.0291

0.0290

[0.0078]**

[0.0085]**

[0.0145]+

[0.0145]*

[0.0145]*

Share of males age 6-12

0.0014

-0.0008

0.0401

0.0400

0.0397

[0.0063]

[0.0070]

[0.0122]**

[0.0122]**

[0.0122]**

Share of females age 6-12

0.0116

0.0100

0.0213

0.0218

0.0215

[0.0065]+

[0.0071]

[0.0124]+

[0.0124]+

[0.0124]+

Share of males age 13-17

0.0119

0.0194

0.0020

0.0028

0.0026

[0.0069]+

[0.0077]*

[0.0143]

[0.0142]

[0.0142]

Share of females age 13-17

-0.0062

-0.0015

-0.0115

-0.0117

-0.0117

[0.0071]

[0.0080]

[0.0156]

[0.0156]

[0.0156]

Share of females age 18-60

0.0018

-0.0057

0.0044

0.0043

0.0044

[0.0057]

[0.0066]

[0.0130]

[0.0129]

[0.0129]

Share of males age > 60

0.0150

0.0056

-0.0120

-0.0119

-0.0117

[0.0059]*

[0.0071]

[0.0125]

[0.0125]

[0.0125]

Share of females age > 60

0.0102

0.0129

-0.0362

-0.0357

-0.0354

[0.0059]+

[0.0070]+

[0.0139]**

[0.0140]*

[0.0140]*

Agriculture main source of income

-0.0090

-0.0094

-0.0056

-0.0059

-0.0060

[0.0021]**

[0.0022]**

[0.0034]+

[0.0033]+

[0.0033]+

Per capita expenditure quintile

(reference = quintile 1, poorest)

Quintile 2

-0.0150

-0.0162

0.0009

0.0002

0.0001

[0.0024]**

[0.0025]**

[0.0043]

[0.0039]

[0.0039]

Quintile 3

-0.0255

-0.0284

-0.0028

-0.0039

-0.0041

[0.0026]**

[0.0027]**

[0.0050]

[0.0042]

[0.0042]

Quintile 4

-0.0418

-0.0454

-0.0058

-0.0077

-0.0079

[0.0027]**

[0.0028]**

[0.0066]

[0.0047]

[0.0047]+

Quintile 5, richest

-0.0657

-0.0663

-0.0112

-0.0141

-0.0145

[0.0029]**

[0.0028]**

[0.0092]

[0.0058]*

[0.0058]*

iii

(Table A.1. continued)

(1)

(2)

(3)

(4)

(5)

BKKBN criteria

Can worship according to faith

-0.0191

-0.0192

0.0140

0.0135

0.0134

[0.0041]**

[0.0044]**

[0.0045]**

[0.0043]**

[0.0043]**

Eat basic food twice a day

-0.0059

-0.0123

0.0108

0.0103

0.0102

[0.0079]

[0.0087]

[0.0086]

[0.0085]

[0.0086]

Different clothing school/work/leisure

-0.0023

-0.0019

-0.0215

-0.0216

-0.0217

[0.0052]

[0.0057]

[0.0083]**

[0.0083]**

[0.0083]**

Floor made of material other than earth

-0.0548

-0.0556

0.0123

0.0100

0.0108

[0.0033]**

[0.0036]**

[0.0062]*

[0.0037]**

[0.0038]**

Access to modern medical care

0.0291

0.0330

0.0337

0.0346

0.0346

[0.0025]**

[0.0028]**

[0.0039]**

[0.0034]**

[0.0034]**

Living conditions

House made out of bam

boo

0.0419

0.0415

0.0031

0.0045

0.0047

[0.0036]**

[0.0039]**

[0.0052]

[0.0043]

[0.0043]

Access to clean drinking water

-0.0077

-0.0071

0.0106

0.0104

0.0103

[0.0029]**

[0.0030]*

[0.0053]*

[0.0052]*

[0.0051]*

Closed sew

er

-0.0229

-0.0242

-0.0024

-0.0034

-0.0031

[0.0025]**

[0.0026]**

[0.0047]

[0.0042]

[0.0042]

Access to electricity

0.0183

0.0176

-0.0149

-0.0140

-0.0139

[0.0029]**

[0.0031]**

[0.0056]**

[0.0049]**

[0.0050]**

Village characteristics

Rural area

-0.0154

-0.0181

-0.0084

-0.0088

-0.0096

[0.0039]**

[0.0041]**

[0.0061]

[0.0060]

[0.0061]

IDT village

-0.0061

-0.0076

0.0083

0.0081

0.0077

[0.0035]+

[0.0037]*

[0.0046]+

[0.0045]+

[0.0045]+

Nr. of primary health clinics

-0.0058

-0.0055

0.0111

0.0108

0.0109

[0.0031]+

[0.0033]+

[0.0049]*

[0.0048]*

[0.0048]*

Nr. of auxiliary health clinics

-0.0093

-0.0081

0.0159

0.0156

0.0158

[0.0026]**

[0.0027]**

[0.0039]**

[0.0038]**

[0.0038]**

Nr. of doctors per 1,000 inhabitants

0.0067

0.0085

0.0060

0.0064

0.0065

[0.0045]

[0.0047]+

[0.0054]

[0.0053]

[0.0053]

iv

(Table A.1. continued)

(1)

(2)

(3)

(4)

(5)

Nr. of midwifes per 1,000 inhabitants

0.0066

0.0080

0.0225

0.0229

0.0230

[0.0037]+

[0.0040]*

[0.0071]**

[0.0074]**

[0.0073]**

Health facilities easy to reach

0.0080

0.0062

0.0139

0.0140

0.0141

[0.0062]

[0.0066]

[0.0070]*

[0.0070]*

[0.0070]*

Majority of inter village traffic by land

0.0212

0.0232

0.0053

0.0066

0.0063

[0.0070]**

[0.0073]**

[0.0116]

[0.0110]

[0.0111]

BKKBN poverty rate in sub-district

0.0982

0.0984

-0.0000

[0.0075]**

[0.0079]**

[0.0101]

BKKBN poverty rate in district

0.0178

0.0071

0.0122

[0.0098]+

[0.0101]

[0.0127]

BPS poverty rate 1999

-0.0008

-0.0007

0.0007

0.0007

0.0006

[0.0001]**

[0.0002]**

[0.0002]**

[0.0002]**

[0.0002]*

Selection term

-0.0070

[0.0166]

Observations

185,608

777,289

76,684

76,684

76,684

Pseudo R-squared

0.10

0.10

0.09

0.09

0.09

Robust standard errors in brackets; + significant at 10%; * significant at 5%; ** significant at 1%.

v

Table A.2. Health card allocation simulations under different geographic targeting regimes

Incidence

(% coverage am

ongst Indonesian population)

Share

(% share of total programme)

Targeting regime

Current

BPS

BKKBN

Uniform

Current

BPS

BKKBN

Uniform

Quintile 1 (poorest)

18.67

20.23

18.57

16.04

34.21

36.90

33.43

28.84

Quintile 2

13.69

13.34

13.53

12.94

25.71

24.95

24.98

23.87

Quintile 3

10.49

10.08

10.53

11.09

19.76

18.91

19.50

20.51

Quintile 4

7.14

7.00

7.81

8.90

13.38

13.07

14.39

16.38

Quintile 5 (richest)

3.69

3.30

4.17

5.65

6.93

6.17

7.70

10.4

Urban

7.29

6.87

7.12

9.98

27.06

25.41

26.01

36.41

Rural

12.95

13.30

13.37

11.50

72.94

74.59

73.99

63.59

All

10.70

10.75

10.89

10.90

100.00

100.00

100.00

100.00

Table A.3. Health card outpatient utilisation simulations under different geographic targeting regimes

Incidence

(% utilisation amongst health card recipients)

Share

(% share of total utilisation)

Targeting regime

Current

BPS

BKKBN

Uniform

Current

BPS

BKKBN

Uniform

Quintile 1 (poorest)

6.43

7.48

7.27

6.94

33.43

37.11

33.94

28.11

Quintile 2

6.57

7.46

7.19

6.92

25.81

25.07

25.14

23.45

Quintile 3

6.58

7.02

6.72

7.04

19.94

17.97

18.45

20.61

Quintile 4

6.80

7.47

7.28

7.21

13.96

13.23

14.80

16.88

Quintile 5 (richest)

6.27

7.83

6.97

7.26

6.85

6.63

7.67

10.94

Urban

6.35

7.11

6.57

6.98

25.74

23.33

23.26

35.56

Rural

6.60

7.50

7.31

7.07

74.26

76.67

76.74

64.44

All

6.53

7.41

7.12

7.04

100.00

100.00

100.00

100.00