Health Policy and Planning Advance Access originally published online on March 11, 2008
Health Policy and Planning 2008 23(3):188-199; doi:10.1093/heapol/czn003
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Targeting the poor in times of crisis: the Indonesian health card
Institute of Social Studies, P.O. Box 29776, 2502 LT The Hague, The Netherlands.
E-mail: sparrow{at}iss.nl
 |
Abstract |
|---|
 Top Abstract IntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
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. Utilization 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 decentralized design combining geographic targeting with community-based targeting instruments. This design facilitated the rapid implementation of the programme, but targeting performance suffered from a lack of 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, and 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 local level, with district programme management and public health care providers.
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 utilized for signalling poverty dynamics and local barriers to access.
Key Words: Access, economic crisis, Indonesia, outpatient care, poverty, price subsidy, targeting
KEY MESSAGES
|
|
Introduction |
|---|
|
TopAbstractIntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
In an attempt to protect access to health care utilization 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 decentralized targeting process, involving both geographic and community-based targeting instruments.
The success of such crisis interventions depends critically 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 centralized 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 decentralized design of the JPS programmes is meant to deal with this targeting problem. The combination of geographic and community-based targeting instruments provides an alternative infrastructure for gathering and processing information locally, and disseminating this to higher administrative levels. Several authors have argued that a decentralized design can benefit from local knowledge and community participation, on the premise that local officials and community members are more capable of identifying the poor.1 Not only do they have better access to information on targeting criteria, they are also more able to prioritize amongst the set criteria or even formulate new local criteria that better reflect the policy objectives.
However, decentralization has its weaknesses. Recently, a number of theoretical and empirical studies have investigated the implications and pitfalls of different aspects of decentralization (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. Decentralized programmes are prone to local elite capture and suffer from classic principal-agent dilemmas (e.g. Bardhan and Mookherjee 2000, 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 decentralized design. The objective is to investigate how the programme has been implemented and who were the beneficiaries of the health cards. Have health cards been allocated to the poor or 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 targeting performance? Has the centre been able to identify the regions hit hardest by the crisis? What determines targeting at local level?
Besides allocation of health cards, the analysis will focus on utilization of the cards, conditional on receipt. The potential benefits of a health card are only reaped if the card is used. Are there barriers unabridged by the health card? Would changes in targeting regime lead to corresponding distributional changes in utilization of health cards?
Finally, a micro-simulation-based decomposition method is employed to attribute overall targeting performance to geographic and local targeting instruments. This methodology allows us to (1) compare the effectiveness of different geographic targeting rules (relative to random targeting), (2) compare the effectiveness of geographic and local targeting instruments and (3) measure the scope for potential improvements for targeting performance at different stages of the targeting mechanism.
The next section sets the context of the programme and contains a detailed description of the programme design. The following section proposes the methodology and describes the data. The results are then presented and discussed, before concluding.
|
The Indonesian economic crisis and the JPS health card programme |
|---|
|
TopAbstractIntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
The Indonesian economic crisis was triggered by a financial crisis that hit Southeast Asia in mid 1997. In 1998, real GDP decreased by roughly 14% and poverty rates increased dramatically.2 According to estimates by the Indonesian Bureau of Statistics (BPS), the poverty headcount increased from 17.7% in 1996 to 23.5% in 1999.
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 crisis saw a sharp decrease in the utilization of outpatient health care from 1997 to 1998. Outpatient utilization rates for modern health care dropped from 0.193 visits per person per month to 0.142 visits, with utilization of public care declining from 0.095 to 0.064.3 This declining utilization of public services was concentrated with local public health clinics. A major cause of the decline in public health care utilization was a shortage of drugs and supplies during the crisis and a subsequent deterioration in quality of public care (Frankenberg et al. 1999; Knowles et al. 1999; Waters et al. 2003). 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 utilization showed slight signs of recovery, which was partly attributed to the JPS health card programme (Pradhan et al. 2007).
The health card programme was a main component of the social safety net, which was initiated in the autumn of 1998 with the aim of safeguarding real incomes and access to social services for the poor.4 The health 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 utilization.
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 to maintain quality of health care. However, there was a loose relationship between the utilization of the health card and the compensation that the health care providers received in return. Compensation was allocated based on the estimated number of households eligible for the health card programme rather than actual utilization 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 decentralized design, with part of the targeting and allocation process delegated to district administrations, villages officials and public health care providers. But while the motivation of the JPS health programme was to shield access to health care from the effects of the economic crisis, at the time of implementation no information was yet available on the regional variation in the intensity of the crisis. This lack of information hindered geographic targeting. In the absence of income or expenditure data (both at district and household level), other measures of welfare were used instead. The number of health cards and the amount of compensation to health care providers was therefore determined by a prosperity measure for 307 districts (urban kota and rural kabupaten), 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 five basic needs criteria: households (1) can worship according to faith, (2) eat basic food twice a day, (3) have different clothing for school/work and home/leisure activities, (4) have a floor that is made out of something other than earth, and (5) 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 criticized for being an unsuitable allocation rule for the JPS, since its components are fairly inflexible and inappropriate for measuring economic shocks or the impact of a crisis on individual incomes.5 However, at the time of programme design the BKKBN prosperity measure for December 1997 was the most up-to-date welfare measure at hand. The BKKBN measure was also used as an allocation rule for both the budgetary support to facilities and health cards to households.
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. 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 allowing local health officials and community members to define additional criteria according to their own insights regarding the effects of the crisis.
The decentralized targeting design shows clear elements of community-based targeting as the programme is implemented by a mix of local officials and community members, who determine eligibility for the health card programme, engage in the distribution of health cards or take part in the monitoring process.6 In this regard the analysis fits within the community-based targeting literature. On the other hand, the notion of a community may not exactly match the context of the JPS programme. Districts in Indonesia are large geographic units, within which the targeting mechanism encompasses a multi-level design and network of intermediate agents that may go beyond the notion of community-based targeting. In the remainder of the paper I will therefore refer to local and with-district targeting, instead of community-based targeting.
|
Methodology and data |
|---|
|
TopAbstractIntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
Benefit incidence analysis and targeting performance
The analysis starts with a static benefit incidence analysis of health card allocation and utilization, 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 central management unit (inter-district targeting), the geographic unit (intra-district targeting) and the household (utilization through self selection). Finally, geographic targeting performance is linked to individual heath card allocation and utilization 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 measure indicates how the benefits are distributed across the population. Targeting performance is defined as the ability of the programme to reach the poor, that is, to what extent the coverage and concentration indices show a pro-poor relative to a distribution neutral pattern.
As the economic crisis was the main motivation for the JPS interventions, an indicator of economic welfare is used as a yardstick for targeting performance. In this case, monthly household expenditure per capita is a more suitable indictor for economic welfare than income, as it can be seen as an outcome of a consumption smoothing process reflecting income, asset base and economic shocks, as well as the ability of households to deal with these shocks. Following an expenditure-based BPS poverty headcount of 23.4%, the poorest population quintile roughly corresponds to the share of the population living below the poverty line. The volatility in expenditure-based poverty found by Skoufias et al. (2000) and Suryahadi et al. (2003) suggests that the bulk of the second poorest quintile lives close to 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 reflects the expenditure-based poverty profile and the geographic heterogeneity of the crisis impact. It is important to note here that it is not the aim of this paper to explore which welfare measure is a better reflection of poverty, as they clearly follow different dimensions and definitions of deprivation. Rather, this analysis aims to illustrate the difficulty of geographic targeting given the limited information available at the time, and, in particular, how the choice of geographic targeting criteria affects distributional outcomes. This will then also shed light on the scope for targeting improvements through geographic poverty indicators.
|
For the second targeting phase, the determinants of within-district targeting are scrutinized 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:
|
| (1) |
jk 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.7
The benefits of the health cards are acquired through the associated price subsidies at public health care providers. But even if households receive a health card there may still be barriers to using it, and thus to enjoying the benefits. The incentive to use a health card for obtaining health care could be reduced by a lack of information, regional shortage of providers, stigmatization, 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 utilization (hcuijk) conditional on ownership:
|
| (2) |
= E( | hc = 1). Under the normality assumptions of a probit analysis, is the inverse Mills ratio computed from the health card allocation probit estimates. However, there is no evidence of selection effects, as β3 is small and not statistically significant.8 To facilitate identification of selection effects, 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.9 Micro-simulation
To separate geographic from local targeting performance, I follow Alatas and Pradhan (2003) who propose an approach based on micro-simulation to attribute changes in targeting performance to specific targeting stages. Targeting performance is measured as deviation from a distribution neutral outcome. Through the simulation method this deviation can be decomposed into a district and local targeting effect, and different geographic targeting regimes can be compared. This will show how local targeting performs on its own, and whether there is scope for geographical targeting in reallocation of health cards to the poor.
Denote the outcome of a two-stage targeting process by the distribution D(G, L), where G and L reflect the geographic and local targeting regime, respectively. The performance of these regimes will be assessed by comparing the outcome D(G, L) with a hypothetical non-targeted outcome, D(G0, L0). This neutral targeting outcome is driven by a uniform geographic targeting regime where allocation to districts simply depends on population size (G0) and random allocation of health cards within districts (L0). Under this regime the probability of receiving a health card is the same for everyone. Overall targeting performance can then be decomposed such that the difference between a specific targeting regime and random allocation can be attributed to the geographic and local targeting instruments separately:
|
|
The estimation results from (1) and (2) are used to simulate targeting outcomes D(G, L) and D(G0, L). 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. Several 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. The analysis will consider two counterfactual within-district eligibility ranks: the rank based on the probit estimates (L) and a random allocation (L0).
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 normalized 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 
if household j received a health card, and the opposite if otherwise.10 A similar approach is taken for utilization: conditional on health card allocation to households under a specific geographic targeting regime, an individual's utilization behaviour is evaluated according to the linear prediction of the utilization probit (2) and error 
. In this way it is possible to track the effects of regional redistribution of health cards to household level allocation of health cards, and then to individual utilization of the health card for outpatient care.
Data sources
The key source of data is Indonesia's main socio-economic survey (Susenas). The Susenas is conducted annually on a national scale, collecting information on health care utilization, socio-economic background of individuals and households, and detailed information on household expenditures.11 In 1999, a special JPS module was included, which covered 202 089 households and 822 607 individuals.12 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.
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, the survey covers socio-economic information such as 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 were 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.13
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 1000 inhabitants), and a variable indicating the village leader's opinion on accessibility of health facilities in the village.14
Finally, I use data on the geographic targeting criteria and household expenditure-based poverty. This includes the percentage of BKKBN poor households in districts (December 1997) and in sub-districts (January 1999). The expenditure-based poverty headcount ratios are computed by BPS based on the 1996 and 1999 Susenas.15
|
Results and discussion |
|---|
|
TopAbstractIntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
Benefit incidence
The health card programme was already of a substantial magnitude in February 1999, with 10.6% of Indonesians living in a household with a health card. Health card recipients show a higher utilization of outpatient care than non-recipients. The difference is largest with utilization of public care. Amongst health card recipients, 15.1% visited an outpatient provider during the 3 months prior to the survey. For non-recipients this is lower, at 12.9%. Although health card owners tend to choose public providers more often, they do not always use their health card. About a third of the health card owners who sought care at a public provider reported not to have used the health card.
What could explain the weak link between ownership and utilization? 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 subsidized services and did not always honour the rule that those who could present a health card should be provided free services. Alternatively, rich households 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 cannot fully explain the under-usage of health cards. Qualitative research by Soelaksono et al. (1999) found 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 providers.
Health cards distribution is pro-poor, as shown in Table 2. Amongst the poorest 20% of the population, 18.5% had a health card. Incidence of health card receipt drops as per capita expenditure increases, from 13.7% in the second quintile to 3.7% in the richest quintile. The allocation shares for ownership and utilization are presented in columns 3 and 5 of Table 2. The poorest 20% of the population own 33.7% of the health cards. Still, there is considerable leakage to the more wealthy households. Considering that about 10.6% of the population received a health card, perfect targeting would imply that all health cards are concentrated with the poorest quintile. However, the wealthiest 60% of the population own 40.6% of the health cards.
|
Utilization of health cards for outpatient care shows a less pro-poor distribution than allocation. The inverted U-shaped pattern of utilization of the health cards, with small differences between quintiles, corresponds closely with the pattern of overall utilization of public outpatient services. Note that utilization of modern outpatient care is strongly correlated with income, but this is mainly driven by the differences in private care, not public (Lanjouw et al. 2002
). Conditional on having a health card, the middle quintiles are more likely to use it. This 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. Nevertheless, 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 because the initial distribution of the health card is skewed towards the poor. Almost three-quarters of the health cards are 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
To what extent did the district targeting criteria reflect regional differences in expenditure-based poverty and impact of the crisis? Several studies have raised concerns about the lack of up-to-date information 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 pre-crisis information would miscalculate the degree of poverty in the districts during the crisis. There are two reasons for this. First, the crisis gave rise to large relative price changes, between products (especially food) and across regions (Cameron 1999; Friedman and Levinsohn 2002; Frankenberg et al. 2003). This variation would be ignored in the targeting process when pre-crisis poverty estimates are applied as allocation rules. Secondly, 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 expenditure-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 household expenditure surveys. Although criticized for suitability, they can provide fairly up-to-date information, as far down as the household level. Nevertheless, the December 1997 BKKBN data would still miss part of the regional variation in the crisis impact.
Table 3 illustrates 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 and the BKKBN measure. The different welfare measures show different levels of poverty.16 As expected, the expenditure-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. This is also reflected by rank correlation coefficients between the rankings under different measures. The Spearman's rank correlation coefficient between the BKKBN and the BPS district indices is positive and significant, but well below 1 (0.67 and 0.64 for BPS 1996 and 1999, respectively); while there is no correlation between the districts ranking of the BKKBN and the change in the BPS headcount (0.09).
|
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, at the district level. The BKKBN prosperity score is strongly positively correlated with the 1999 poverty headcount, but with a lot of variation around the trend. There is a weak positive correlation between the change in poverty and the BKKBN measure for the main body of districts and a greater variation around 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). Although the correlation is positive and statistically significant, the slope is flat. This is in part due to under-coverage, as for the majority of districts the coverage rate lies well below the BPS poverty rate. But there is also a large variation around the trend line. This indicates a fuzzy relationship between the BKKBN targeting criteria and expenditure-based poverty measured by BPS in 1999. Another reason is that at the time of the Susenas survey the programme was still expanding in some districts and speed of implementation could differ between districts. But if it does, then these irregularities are greater in relatively poor districts.
|
Local targeting and health card utilization
What are the key factors that determine local targeting within districts? Table 4 reports probit marginal effects of the determinants for health card allocation as well as utilization conditional on ownership, at household level. The standard errors have been adjusted for the stratified sampling design of the Susenas survey.
|
There are significant negative effects of per capita expenditure 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 utilization differences between quintiles are too small to yield statistically significant marginal effects in the multivariate analysis. Alternatively, it could be that the non-poor trend in utilization is not a direct wealth effect but follows from other background characteristics, which are correlated with wealth.
Sector of employment affects both allocation and utilization. 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 play 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 a village strongly increases the use of health cards for outpatient care. Further, utilization 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 with utilization. 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 had 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).
Women tend to use the cards more for outpatient care than men do. The outpatient utilization 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.
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. But in contrast to the other individual BKKBN criteria, 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, outweighing the official selection rule that areas with little access should be targeted.
Simulation results
The counterfactual outcomes needed for the decompositions are presented in Table 5. The first three outcomes refer to specific geographic targeting regimes, D(G, L) in equation (3): the actual outcome for the sample used in the multivariate analysis and two simulated outcomes for geographic targeting following BPS and BKKBN indicators. The fourth outcome refers to D(G0, L): uniform geographic targeting combined with observed local targeting. That is, purposive targeting occurs only within districts. The counterfactual D(G0, L0) is not shown as it is simply an equal share of health cards for each quintile. The table shows concentration of health cards and utilization, to make the performance of different targeting regimes comparable.17
|
Not surprisingly, the actual outcome for the sample corresponds quite closely to a geographical targeting regime that would follow the BKKBN rule strictly. However, the BPS indicator would increase allocation to the poor compared with the actual regime, as this would redistribute 2.7% 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.
Relative to uniform geographic targeting, the effect of the actual geographic targeting regime [D(G, L)- D(G0, L)] is an increase in the share of the poorest quintile by 5.4 percentage points, from 28.8 to 34.2%. Considering that complete random targeting at all levels would yield a concentration of 20% per quintile, local targeting alone [D(G0, L)- D(G0, L0)] increases health card concentration with the poorest by 8.8 percentage points. Hence, for the poorest quintile the overall gain from decentralized targeting [D(G, L)- D(G0, L0)] is an additional 14.2% share of the health card programme, 37.8% (5.4/14.2) of which is due to geographic targeting and 62.2% (8.8/14.2) to local targeting. The BPS targeting rule would increase this even further to a 16.9% share, of which 47.7% (8.1/16.9) would be due to geographic targeting.
The simulations for health card utilization 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 geographic targeting. Health card concentration with the poorest quintile under the best performing scenario is still only 36.9%, while perfect targeting to the poorest quintile would imply 100% concentration. The results do not change much if we take a broader target group that includes the second quintile. The poorest 40% of the population would receive a 61.9% share with the BPS geographic targeting regime accounting for 41.8% (9.1 percentage points) of the deviation from random targeting. Thus, the difference with perfect targeting would need to be bridged by local targeting, as most barriers to access seem to be locally determined and little affected by geographic targeting.
|
Conclusion |
|---|
|
TopAbstractIntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
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 remarkable speed: by February 1999 approximately 22 million people (10.6% of Indonesians) lived in households that received a health card. The decentralized 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 were no reliable data on the impact of the crisis across districts. Geographic targeting criteria were therefore based on poverty estimates that reflect the actual level of poverty to some extent but do not fully capture the income shocks 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 utilization. Moreover, utilization of subsidized 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 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, potential gains in targeting performance through higher accuracy of geographic poverty indicators are modest. Most of the targeting gains are to be made within the districts, with the district allocation committees and public health care providers. In addition, the simulation exercise shows that improved targeting of price subsidies does not automatically imply improved access to public services for the poor, as the gap between ownership and utilization remains largely unaffected by health card targeting mechanisms. The results of the multivariate analysis would suggest that price subsidies need to be complemented by other interventions that reduce indirect costs and other barriers to utilization of public health care, such as reduction in transport costs or supply side impulses.
In terms of policy implications for targeting of future safety nets and crisis responses, the need for adequate information systems is apparent from this study. While the decentralized 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 decentralized targeting in signalling crisis-related poverty dynamics and local barriers to access.
|
Acknowledgements |
|---|
I thank Menno Pradhan, Jan Willem Gunning, Marleen Dekker, Sudarno Sumarto and two anonymous referees 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 remaining errors are my own.
|
Endnotes |
|---|
|
TopAbstractIntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
1 See, for example, Alderman (2001
, 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 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.
4 The JPS further included 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.
5 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) further question to 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 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% of the households in the sample as poor. But according to per capita consumption, only 57% of these households rank with the poorest 49% of the population.
6 See, for example, Conning and Kevane (2002) or Coady et al. (2004).
7 An alternative specification would be to include the observed health card coverage in districts. However, this variable would be endogenous to the outcome variable.
8 Since the utilization 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. Joint estimation of the two equations, by means of Heckman probit, yields identical results, showing no correlation between the error terms. The results are reported in the supplementary appendix, which is available upon request.
9 Nevertheless, there may be reasons why the BKKBN indicators could still be correlated with utilization. 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 ). 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.
10 The errors are computed as 
where H and L are the upper and lower truncation points, respectively, 
reflects the standard normal cdf and u is drawn from a uniform distribution with a value between 0 and 1.
11 Household expenditures are adjusted to account for regional price differences. Further details on construction of the household expenditure variable are available upon request.
12 The 13 districts in the province of East Timor are excluded from the analysis due to incomplete data for the JPS module.
13 IDT refers to the Inpres Desa Tertinggal programme, an anti-poverty programme for economically less developed villages. For this programme, each village or urban precinct in Indonesia has been classified as either developed or less developed.
14 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.
15 See BPS (2000) for details.
16 Note that the district is the administrative unit for geographic targeting. Nevertheless, for ease of presentation, the table reports poverty measures for 26 provinces. In the remainder of the paper the unit of analysis for geographic targeting is the district.
17 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%, slightly above the observed 10.6% 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 normalized to 100% for all counterfactual regimes. The coverage results are not reported here but are available upon request.
|
References |
|---|
|
TopAbstractIntroductionThe Indonesian economic crisis...Methodology and dataResults and discussionConclusionEndnotesReferences |
|---|
Alatas V, Pradhan MP. Geographic targeting in Indonesia (2003) Jakarta: World Bank. Unpublished document.
Alderman H. Multi-tier targeting of social assistance: the role of intergovernmental transfers. The World Bank Economic Review (2001) 15:33–53.[Abstract]
Alderman H. Do local officials know something we don't? Decentralization of targeted transfers in Albania. Journal of Public Economics (2002) 83:375–404.[CrossRef][Web of Science]
Ananta A, Siregar R. Social safety net policies in Indonesia: objectives and shortcomings. ASEAN Economic Bulletin (1999) 16:344–59.[CrossRef]
Bardhan P, Mookherjee D. Capture and governance at local and national levels. American Economic Review (2000) 90:135–9.[Web of Science]
Bardhan P, Mookherjee D. Decentralizing antipoverty program delivery in developing countries. Journal of Public Economics (2005) 89:675–704.[CrossRef][Web of Science]
BPS. (2000) Tingkat Kemiskinan Kabupaten. Jakarta. Unpublished document.
Cameron LA. Survey of recent developments. Bulletin of Indonesian Economic Studies (1999) 35:2–40.
Coady D, Grosh ME, Hoddinott J. The targeting of transfers in developing countries: review of experiences and lessons (2004) Washington, DC: World Bank.
Conning J, Kevane M. Community-based targeting mechanisms for social safety nets: a critical review. World Development (2002) 30:375–94.[CrossRef][Web of Science]
Daly A, Fane G. Anti-poverty programs in Indonesia. Bulletin of Indonesian Economic Studies (2002) 38:309–29.[CrossRef][Web of Science]
Dhanani S, Islam I. Poverty, vulnerability and social protection in a period of crisis: the case of Indonesia. World Development (2002) 30:1211–31.[CrossRef][Web of Science]
Frankenberg E, Thomas D, Beegle K. The real costs of Indonesia's economic crisis: preliminary findings from the Indonesian Family Life Survey. (1999) Labor and Population Program Working Paper 99–04: Santa Monica: RAND.
Frankenberg E, Smith JP, Thomas D. Economic shocks, wealth, and welfare. Journal of Human Resources (2003) 38:280–321.
Friedman J, Levinsohn J. The distributional impact of Indonesia's financial crisis on household welfare: a "rapid response" methodology. The World Bank Economic Review (2002) 16:397–423.[Abstract]
Galasso E, Ravallion M. Decentralized targeting of an antipoverty program. Journal of Public Economics (2005) 89:705–27.[CrossRef][Web of Science]
Knowles JC, Pernia EM, Racelis M. Social consequences of the financial crisis in Asia. (1999) Economic Staff Paper Number 60: Manila: Asian Development Bank.
Knowles JC, Marzolf JR. Health financing for the poor in Indonesia. (2003) Washington, DC: World Bank. Unpublished document.
Lanjouw P, Pradhan MP, Saadah F, Sayed H, Sparrow R. Poverty, education and health in Indonesia: who benefits from public spending? In: Education and health expenditures, and development: the cases of Indonesia and Peru—Morrisson C, ed. (2002) Paris: Development Centre Studies, OECD Development Centre. 17–78.
Pradhan MP, Saadah F, Sparrow R. Did the Health Card Program ensure access to medical care for the poor during Indonesia's economic crisis? The World Bank Economic Review (2007) 21:125–150.
Pritchett L, Sumarto S, Suryahadi A. Targeted programs in an economic crisis: emperical findings from the experience of Indonesia. (2002) SMERU working paper: Jakarta: SMERU.
Skoufias E, Suryahadi A, Sumarto S. Changes in household welfare, poverty and inequality during the crisis. Bulletin of Indonesian Economic Studies (2000) 36:97–114.[CrossRef][Web of Science]
Smith JP, Thomas D, Frankenberg E, Beegle K, Teruel G. Wages, employment and economic shocks: evidence from Indonesia. Journal of Population Economics (2002) 15:161–93.[CrossRef][Web of Science]
Soelaksono B, Budiyati S, Hastuti, et al. 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. (1999) Jakarta: SMERU. Unpublished document.
Strauss J, Beegle K, Dwiyanto A, et al. Indonesian living standards: before and after the financial crisis (2004) Santa Monica: RAND.
Sumarto S, Wetterberg A, Pritchett L. The social impact of the crisis in Indonesia: results from a nationwide Kecamatan survey. (1998) SMERU working paper: Jakarta: SMERU.
Sumarto S, Suryahadi A, Pritchett L. Safety net or safety ropes? Dynamic benefit incidence of two crisis programs in Indonesia. World Development (2003) 31:1257–77.[CrossRef][Web of Science]
Suryahadi A, Sumarto S, Pritchett L. The evolution of poverty during the crisis in Indonesia. Asian Economic Journal (2003) 17:221–41.[CrossRef]
Suryahadi A, Suharso Y, Sumarto S. Coverage and targeting in the Indonesian social safety net programs: evidence from 100 village survey. (1999) SMERU working paper: Jakarta: SMERU.
Waters H, Saadah F, Pradhan MP. The impact of the 1997–98 East Asian economic crisis on health and health care in Indonesia. Health Policy and Planning (2003) 18:172–81.
Accepted for publication 16 January 2008.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||