Minimum Wages around Birth and Child Health

Abstract

This paper studies the effects of minimum wages in Indonesia around the time of birth on child height-for-age Z scores (HAZ) up to five years of age. Using variations in annual fluctuations in real minimum wages in different Indonesian provinces, it finds that children exposed to increases in minimum wages in their birth years have higher HAZ in the first five years of their lives. The estimated impacts are based on difference-in-differences models with biological-mother fixed effects and year-of-birth fixed effects and are robust to inclusion of multiple time-varying factors. The impacts are prominent particularly among male children.

1. Introduction

Recent surveys in The Lancet estimate that 250 million children under five years of age in low- and middle-income countries (LMICs) are at risk of failing to reach their full developmental potential, with child height-for-age Z scores (HAZ) used as a key marker of child development (Adair et al. 2013; Black et al. 2017).¹ Furthermore, evidence is increasing rapidly that early-childhood nutrition is a key predictor of later-life cognition (e.g., Field, Robles, and Torero et al. 2009; Maluccio et al. 2009; Majid 2015; Almond, Currie, and Duque 2018; Attanasio, Meghir, and Nix 2020) and labor market success (e.g., Behrman and Rosenzweig 2004; Behrman et al. 2009; Almond, Currie, and Duque 2018; Bijwaard et al. 2019; Majid, Behrman, and Mani 2019), suggesting that there are serious long-run consequences of neglect of early-life child development. Increases in parental socioeconomic status (SES) may be key to improving early-life child nutrition, as parental SES is generally understood to be an important determinant of child health with the existence of a gradient between adult SES and child health now well-established (Cutler et al. 2008; Schady et al. 2015). However, credibly studying impacts of parental SES on child health has proved elusive due to concerns about reverse causation and omitted variable bias (Chandra and Vogl 2010; Baker and Stabile 2012; Cesarini et al. 2016). One review article on the causes and consequences of early-childhood health notes: “The number of studies associating poor child outcomes with low SES far exceeds the number that make substantive progress on this difficult question of causality” (Baker and Stabile 2012, 8).

This study evaluates minimum wages (MWs) around the time of birth and their impact on child nutritional status in Indonesia up to five years after birth. Indonesia carries the fifth-highest burden of stunted children in the world, and MWs are an integral part of the social policy debate in Indonesia with regular worker protests for higher MWs (Chun and Khor 2010; Magruder 2013). If MWs influence parental wage income, they may affect parental investments in child health. For example, if wages increase due to higher MWs, families may be more likely to avail themselves of health services and engage in other salutary behaviors that may be particularly effective around the time of birth. However, with higher wages, mothers also may be likely to spend more time in the labor market at the expense of child care-giving activities (Behrman and Rosenzweig 2002). The time around birth is widely understood to be a critical period in shaping child nutritional status, so that changes in parental economic conditions and time devoted to childcare during children's early lives may have particularly large effects on child health and nutrition.

This paper makes two contributions. First, it studies the effects of MWs on children's HAZ in a national context. HAZ in the first five years of life is considered a key marker of overall child development, and a key outcome of interest for ending hunger and malnutrition in the planet as per the United Nation's Sustainable Development Goal 2 for 2025. In cross-country difference-in-difference estimates in the literature (Majid et al. 2016), spatial variation in minimum wage levels comes from a number of countries that may have very different legal, historical, and socioeconomic conditions, whereas in the present study these variations come from different provinces within one country, Indonesia. Second, this paper contributes to the voluminous minimum wage literature (Neumark and Wascher 2010; Belman and Wolfson 2014; Belman et al. 2015) by applying a biological-sibling fixed-effects model. Prior work, even in the United States, has not used biological-sibling fixed-effects models in the context of MWs (Wehby et al. 2016). This paper is uniquely able to do this as it is interested in studying effects on children, whereas most of the MW literature studies effects on adults for whom siblings are harder to identify in the available data.

To identify impacts, this study exploits timing of births relative to patterns in MWs: some children happen to be born in years with higher MWs whereas others are born in years with lower MWs. Comparisons of siblings born from the same biological mother helps address concerns about potential roles of unobserved factors that lead some mothers to time births relative to changes in MWs and helps to control for unobserved fixed factors common to siblings that might cause serial correlation and omitted-variable bias. The paper further strengthens this basic strategy with a differences-in-differences (DiD) framework that compares cohort differences among biological siblings in provinces with high versus provinces with low MWs. It uses data on child anthropometrics (0–5) from the 2007–2008 wave of the Indonesian Family Life Survey (IFLS, wave 4), a rich nationally representative longitudinal dataset. A panel of policy data, from the Indonesian Bureau of Statistics, was constructed for province-specific MWs as well as other time-varying factors at the provincial level that covers five years (2002–2008) and was merged with the IFLS.

The results show that children exposed to higher MWs in the year of birth have higher HAZ in the first five years of their lives. Effects are particularly prominent for male children. The estimated impacts are evident with differences-in-differences models with biological-sibling fixed effects and year-of-birth fixed effects and are robust to inclusion of measures of child characteristics (birth order, gender) and parental characteristics (ethnicity, age, and schooling attainment, household assets), as well as community covariates (province-specific time trends, urban status, and urban-rural-specific time trends, provincial real GDP per capita, lagged provincial unemployment rate). The study does not know the labor-market outcomes for parents of most children in the sample at time of birth (children were born between 2002 and 2008 but labor-market data were collected in 2007/2008 only). Even so, data on parental labor-market outcomes from up to five years after births can shed some light on associations between minimum wages and parental labor-market outcomes.

MW effects of child HAZ are prominent (at the 5 percent level of statistical significance), particularly among children whose fathers earn in the bottom half of the wage distribution, whereas no effects are found for fathers with earnings in the top half of the wage distribution. Effects are also found on child HAZ for fathers working in the informal sector. Although labor-force participation by mothers is low, leading to smaller samples sizes for mothers than for fathers, the study finds that mothers who work (in formal or informal sectors), those who work in part-time and full-time jobs in small and larger firms all have children with higher HAZ from higher MWs. However, it also finds that higher MWs lead to lower employment and salaries for some mothers, whereas no negative employment effects are found for fathers.

The rest of the paper is organized as follows.²Section 2 reviews the data. Section 3 presents a simple conceptual framework. Section 4 presents the econometric models used for estimation. Section 5 presents results and discusses their robustness and their economic/health significance. Section 6 concludes.

2. Data

This study has considerable data requirements. It necessitates multilevel data linking provincial MWs to individual data on parents, children, and communities. The study needs data on children's health and on parental labor-market outcomes, and it needs to disentangle mechanisms by assuring that the data on children are linked to data on their actual biological parents at the time of birth. Furthermore, the study needs panel data on other confounding variables with provincial and time variation.

A panel of policy data for minimum wages was constructed for provincial MWs that covers the post initiation-of-decentralization period 2002–2008 for each province. Similarly, data on the consumer price index (CPI) at the province level from the Indonesian Bureau of Statistics and Indonesian Ministry of Manpower were assembled to compute a measure of real provincial MWs. Most LMICs have national MWs only, but Indonesia is one of the few LMICs that have rich subnational cross-sectional and time variation in MWs (see fig. 1). In contrast in the United States minimum wages are not updated frequently, with the U.S. federal minimum wage last updated in 2009 (Chun and Khor 2010).

To obtain information on child health and other variables relating to household- and individual- levels, this study uses the Indonesian Family Life Survey (IFLS). Since 1993, this longitudinal survey, which is publicly available at the RAND website, has been collecting data that pertain to individuals, their households, and their communities. The IFLS represents 83 percent of the Indonesian population across 13 out of 26 provinces. The present study obtains information from the 2007–2008 wave of the IFLS. The child health outcome of interest is HAZ. WHO 2006 guidelines were used to determine HAZ based on anthropometric data on child height in centimeters, child age, and child sex. It also assembles parental variables, such as gender, age, education, employment, and income. Moreover, the study connects information between each child and their parents so that the connection represents the family relationship at the time of the child's birth.³

To understand effects of minimum wages on preschool children, the study restricts analyses to those aged 0–5 in the 2007–2008 wave of the IFLS. Furthermore, the IFLS sampling strategy is such that only two randomly selected children, aged 0 to 14, of the head and spouse are typically interviewed, except for “origin” households that are being followed from previous rounds (IFLS 1, IFLS 2, IFLS 3), in which case everyone is interviewed. The age restriction and the sampling design of IFLS mean that about one-third of the households in the full sample are in the sibling sample for the analysis of IFLS 4 data for 0–5 year-olds.

Last, the study obtains data on covariates from the World Bank Development Indicators and the Indonesian Bureau of Statistics. The province-level variables include unemployment rates and real provincial GDP per capita.

3. Conceptual Framework

Consider a life-cycle framework, with emphasis on two stages: (1) the period during pregnancy and up to birth and (2) subsequent early childhood (i.e., before school-entry age up to five years). Conceptualize children as starting life with a vector of genetic and environmental endowments or “inputs” (Y0). Conditional on these endowments, early childhood health is directly affected by investments that mediate the relationship between any MWs at birth and child health at birth and between child health at birth and after birth up to age five years. Investments in human development include: (1) familial investments such as provision of nutritious foods, supplements, and vaccines; and (2) public investments such as public expenditures on health care (Behrman, Rosenzweig, and Taubman 1994; Heckman 2006).

Figure 2 presents an overview of this framework. MWs potentially increase the cost of production for firms, who may decide to fire workers or to retain them. Workers who are fired may respond by switching occupations from the formal sector to the informal sector. In contrast, MWs may increase employment and wages if workers with higher pay consume more products and services, leading to more jobs with higher pay and possible shifts from informal to formal work. Based on the literature reviewed in section S1 of the supplementary online appendix, firm size (small or medium/large) and sector (formal or informal) where parents worked prior to MW changes may play important roles in mediating the effects of MWs on parental earnings/wages. To the extent that MWs influence parental wage incomes, they might affect parental investments in child health. For example, if parental wages increase, families might more likely avail themselves of health services and engage in other salutary behaviors (eating more nutritious foods) that may be particularly effective around the time of birth. However, mothers may also be more likely to spend more time in the labor market at the expense of care-giving activities. Familial and public investments—in the box at bottom left (figure 2)—may moderate not only the impact of changes in parental wages/earnings, and their impacts on child health at birth, but also how outcomes at birth influence life after birth.

4. Econometric Models

The following equation estimates the reduced-form effects of minimum wages at birth on subsequent preschool-age child health:

where HAZ_impt is the HAZ for child i of mother m living in province p and born in year t. MW_pt is a vector of real minimum wages by provincial levels p in year t in which child i was born. The sample is restricted to include families with two or more children per biological mother and controls for γ_m, biological-mother fixed effects. All time-invariant unobserved factors that determine differences between children's HAZ across mothers will be controlled for. For instance, if mothers time their births to take advantage of MWs, all time-invariant factors that shape composition of births will be controlled by comparing children of the same biological mother. The study can thus control for variables like parents' age, education, height, genes, ethnicity, and permanent income as well as community- and province-specific factors that shape determination of labor legislation and health and well-being of mothers and children; g(t) represents birth-year fixed effects to control for other unobserved time-related characteristics. Because each province takes into account its own economic conditions when deciding its minimum wage level, the study accounts for heterogeneity from any correlation between these economic factors.

Biological-mother fixed effects control for any time-invariant differences among not only mothers, but also among provinces, which may bias the estimated effects of minimum-wage levels on health if they are not controlled. This identification strategy essentially compares differences in HAZ across biological siblings and compares such differences across provinces in Indonesia with varying levels of MWs. In the basic model, X_impt includes child gender. However, in addition to equation (1), this study also estimates models with wide ranges of additional covariates. These include child's birth order and parental characteristics (such as schooling attainment and measures of household assets), as well as community covariates (province-specific time trends, urban-rural specific time trends, provincial real GDP per capita, lagged provincial unemployment rate).⁴ε_impt is an error term that includes measurement error in HAZ.

Models are also estimated to assess heterogeneous treatment effects in (1), where subgroup effects are shown by parental labor-market outcomes. Furthermore, an alternate to model (1) is estimated that replaces biological-mother fixed effects with provincial fixed effects to study how MWs affect parental labor-market outcomes (e.g., employment)⁵ for the parents of the same children aged zero to five for whom this paper studies HAZ in IFLS 4.

5. Results

Validity of Biological-Mother Fixed-Effects Model to Study MWs

One concern with studying effects of MWs through a biological-mother fixed-effects model is that MWs may affect fertility so that families facing higher MWs are deferentially likely to have two or more children in the relevant age range than families facing lower MWs. In that case, samples with only biological siblings may suffer from selection bias. To address this concern, table 1, in contrast to equation (1), expands the sample to all children zero to five in IFLS 4 for whom there are valid entries of child HAZ. This study also replaces biological-mother fixed effects with provincial fixed effects, to allow comparisons across families with and without siblings. The dependent variable in column (1) is a dummy for being in the sibling sample (i.e., families with two or more children in the zero-to-five age range from the same biological mother versus only one child from the same biological mother). The results show that higher MWs are not associated with the likelihood of being a single child versus one with a biological sibling. Columns (2)–(6) show estimates where the dependent variables are mothers’ ages, mothers’ highest school grades attained, fathers’ ages, fathers’ highest school grades attained, and household asset indices, respectively. The key variable for the models presented in columns (2)–(6) is the interaction between MW and the dummy for being in the sibling sample. The controls include MW, provincial fixed effects, birth-year fixed effects, child gender, dummy for sibling-sample and interactions of sibling-sample dummy with provincial fixed effects, interactions of sibling-sample dummy with birth-year fixed effects and interactions of sibling-sample dummy with child-gender dummy. None of the coefficients of interest in table 1 are statistically significant at the 10 percent level of significance. Table 1, thus, provides suggestive evidence that biological-mother fixed-effects models do not suffer from selection bias and therefore may be relevant for understanding effects of minimum wages in families with children aged 0–5.

Summary Statistics

Consider summary statistics for the key variables from the analyses from families with two or more biological siblings between ages zero to five in table 2. The average level of the real monthly provincial minimum wage is 38.9 U.S. dollars (USD) per month, but with a wide range between 10.54 to 105.8. Real values are calculated by using provincial CPI values in 2008 USD. Most papers use national CPIs because provincial CPIs are unavailable.⁶ The average of HAZ is 0.028, which implies that the average child in the IFLS sample is not undernourished but has a height-for-age at about the median for a well-nourished population, though there is a large SD of 1.878.⁷ There are a little over two children per biological mother who are aged zero to five (in 2007–2008) in the sample, with a maximum of four children in some families. The average mother is 30 years old, has nine grades of schooling, whereas the average father is 34 years old and has nine grades of schooling; 46 percent of mothers work, whereas 99 percent of fathers work. Of the working mothers, only 31 percent work in the formal sector, and 52 percent work in full-time jobs, and mothers earn about USD 1,140 on average in the year before the 2007–2008 interview.⁸ Of the working fathers, 42 percent work in the formal sector, and 76 percent work in full-time jobs, and fathers earn on average about USD 1480 in the year before the 2007–2008 interview; 75 percent of Indonesian households own a TV, nearly everyone owns a stove, but only 64 percent own a private toilet with septic tank, and only 44 percent own a refrigerator. Monthly per capita expenditures average 85 USD and total assets average 1410 USD per capita. The average unemployment level is 10.3 percent, with significant variation between ∼4 percent to ∼18 percent across provinces.

Main Results

Table 3 presents estimates of equation (1) and shows the effects of higher MWs on children's HAZ. Column (1) presents results for an OLS model without biological-mother fixed effects. Column (2) shows results with biological-mother fixed effects for boys and girls combined. Column (3) is the same specification as column (2) but for male children only. Column (4) is the same specification as column (2) but for female children only. This study presents separate results for boys and girls because there are competing hypotheses about gender effects in the literature: (1) the male-vulnerability hypothesis (Kraemer 2000; Kruger and Nesse 2004) assert that, particularly early in life, males are biologically more vulnerable to changes; (2) the male-bias-in-resource-allocation hypothesis that claims that males are favored in resource allocations (Sen 1984; Gupta 1987); (3) the girls-are-luxuries hypothesis claims that boys are treated as basic necessities and girls as luxuries so marginal changes in resources are experienced more by girls (Behrman and Deolalikar 1990). This study finds that a one USD higher monthly real minimum wage in year of birth leads to a 0.028 increase in HAZ overall. However, boys tend to be the primary beneficiaries of higher MWs relative to girls (who register a smaller and statistically insignificant increase), which is consistent with both the male vulnerability and male-bias-in-resource-allocation hypothesis, but inconsistent with the girls-are-luxuries hypothesis, as boys tend to benefit more from marginal changes in MWs.

Suggestive Mechanisms

The labor-market outcomes for parents of all children are unknown in the sample at the time of birth (children were born between 2002 and 2008, but labor-market data were collected in 2007–2008 only), but data from up to five years after births on parental labor-market outcomes can shed some light on associations between MWs and parental labor-market outcomes because there is probably some serial correlation in such behaviors. Table 4 explores how MW effects on children's HAZ vary by their fathers' labor-market outcomes and schooling. All columns show results from biological-mother fixed-effects models. Each column shows heterogeneity of MW effects on children's HAZ by fathers’ labor-market outcomes. Samples in each column are restricted to (1) fathers earning less than the median salary, (2) fathers earning more than or equal to the median salary, (3) fathers having less than nine grades of schooling, (4) fathers having nine or more grades of schooling, (5) fathers working in the formal sector, (6) fathers working in the informal sector, (7) fathers working full time, (8) fathers working part time, (9) fathers working in firms with less than 20 employees, ⁹ and (10) fathers working in firms with 20 or more employees.

MW effects on children's HAZ are prominent (at the 5 percent level of statistical significance), particularly among children whose fathers earn in the bottom half of the wage distribution, whereas there are no effects for fathers with earnings in the top half of the wage distribution. There are positive effects of higher MWs on children's HAZ for fathers working in the informal sector and for fathers who work part time or in medium-to-large firms (at the 10 percent level of statistical significance).¹⁰

Table 5 conducts a similar exercise as table 4 except that it analyses heterogeneity in treatment effects by mothers' labor-market outcomes. In contrast to men who work universally, 46 percent of mothers with children aged zero to five in IFLS 4 work, so this study first shows effects by mothers’ labor-force participation. Although labor-force participation by mothers is relatively low, leading to small sample sizes when heterogeneity is explored, this study finds that mothers who work (in formal or informal sectors), who work in part-time or full-time jobs, in small and larger firms, all have children with higher HAZ if MWs are higher (at the 5 percent level of statistical significance). Furthermore, children of mothers with earnings more than the median wages and those with more than nine grades of schooling also have higher HAZ when MWs are higher (at the 10 percent level of statistical significance).

Table 6 presents estimates of the effects of MWs on fathers' and mothers' employment and earnings for the sibling sample (OLS model) to explore further likely mechanisms at play. There is no negative impact on fathers’ employment, earnings on average or on likelihood of working in the formal sector. However, there are adverse effects on mothers' employment status and earnings, though no effects on working in the formal sector, conditional on working.

Robustness

There are several challenges in investigating relations for poor populations between MWs and children's health. Minimum-wage legislation may be introduced along with other legislation and polices, or the effects may be driven by changes in birth composition / selective births: This study deals with these issues by comparing within-province, across biological-sibling exposure, to changes in levels of MWs rather than the introduction of new legislation (which may be bundled with other laws). Another concern may be that MWs can increase prices and reduce investments in children through inflation: this study estimates real MWs, which control for price changes using provincial CPI data.¹¹ A third concern is measurement error. To the extent that there is classical measurement error in MWs, study estimates are biased downwards. For nonclassical measurement error, the study is able to control for time-invariant unobserved factors through sibling fixed effects—comparing changes in children's health among children from the same biological mother—so any time-invariant component of error in, say, parental health-investments reports or parental economic background (e.g., status/class) are washed out.

To provide additional evidence on robustness of the main estimates, in table 3A in the supplementary online appendix, this study adds a wide range of time-varying controls at child, parental, provincial, and regional levels. It is necessary to be cautious with these results, since some of these may be poor controls (e.g., province-specific time trends), as they may present pathways through which MW affects child HAZ. Nonetheless, if the main results remain qualitatively similar, it is somewhat reassuring. Column (1) shows results from an OLS model that controls for provincial fixed effects, birth-year fixed effects, child gender, birth-order dummies, urban/rural status at time of interview, and urban-rural-specific time trends, provincial time trends, provincial real GDP per capita, lagged provincial unemployment rate, mothers’ ages in years, mothers' grades of schooling attained, fathers' ages, fathers' grades of schooling attained, dummies for mothers' ethnicity and dummies for fathers' ethnicity, asset index for household, television, refrigerator, toilet, stove, per capita expenditure, and total assets per household member. Column (2) shows results from a biological mother fixed-effects model that controls for biological-mother fixed effects in addition to all the controls in column (1). Column (3) is the same specification as (2) but for the male children. Column (4) is the same specification as (2) but for the female children. These estimates generally are noisier but remain broadly robust. Estimated effects of MWs on male children's HAZ are larger and remain statistically significant at the 5 percent level, whereas effects of MWs on female children's HAZ become smaller and remain statistically insignificant.

Magnitude

The point estimates suggest a sizable effect of MWs in the year of birth. In contrast to cross-country estimates, which suggested that increases in real monthly MWs were associated with harmful effects on child health (Majid et al. 2016), minimum wages in Indonesia have had a positive and sizable effect on children's height for age. If estimates from table 3 are taken as the benchmark, a one USD monthly increase in real MWs leads to 0.028 to 0.044 increase in HAZ for zero-to-five years olds, which is equivalent to an increase of 0.55–0.73 of a SD in HAZ in a well-nourished population (see footnote 1) in response to a one SD increase in MW. Wehby et al. (2016) do not study effects beyond birth, but that study finds rather modest effects on birth weight—a 1000 USD increase in annual household income (83.3 USD monthly) leads to a 8.5 gm increase in birth weight.

6. Conclusion

A large literature establishes “gradients” between parental SES and children's health. However, relatively little is known about casual effects of parental SES on children's health. Furthermore, despite a voluminous literature on MWs in developed and developing countries, not much is known about health effects of MWs. Using variation in annual fluctuations in real MWs in different provinces of Indonesia, this study finds that children exposed to higher MWs in their years of birth have higher height-for-age Z scores (HAZ) in their first five years of life. Furthermore, this study uses data on parental wages to focus on children of parents for whom MWs are most likely to be binding—those whose parents are below the median of the wage distribution. The estimated impacts are evident with differences-in-differences models with biological-mother fixed effects and year-of-birth fixed effects and are robust to inclusion of measures of child characteristics (age, gender, birth order) and parental characteristics (age and schooling attainment, household income and assets) as well as community covariates (province-specific time trends, urban-rural-specific time trends, provincial GDP and unemployment rates).

This study found that the effects of MWs on child HAZ tend to be concentrated among boys rather than girls. The gender results are consistent with both the male vulnerability (Kramer 2000; Kruger and Nesse 2004) and male-bias-in-resource-allocation hypothesis (Sen 1984; Gupta 1987), but inconsistent with the hypothesis that parents view boys as basic necessities and girls as luxuries that are so marginal that changes in resources are experienced by girls (Behrman and Deolalikar 1990).

This paper contributes to filling the current critical gap in knowledge on the impacts of income-transfer polices, especially minimum wages, at birth in a poor population on development of an important measure of child development and nutrition- height-for-age Z scores (HAZ) up to five years after birth. This gap is critical because about 250 million children under five years of age in LMICs are at risk of failing to reach their full developmental potential as indicated primarily by low HAZ (stunting) and secondarily by living in extreme poverty. Indonesia, the country of interest in this study, has the fifth-greatest burden of stunting in the world. Central to this paper is an innovative identification strategy and rich panel data connecting annual provincial minimum wages to individual data on a key child anthropometric measure, embedded in a life-cycle framework motivated by recent conceptual work (Heckman 2006). The U.S. National Institute of Health considers research on social factors shaping child development, multilevel interaction and inputs to human health and to health and disease across the lifespan to be high priorities (NICHD 2018). This paper contributes to filling this knowledge gap by evaluating the impact of provincial MWs during a critical period (year of birth) on child nutritional status up to five years after birth, paying attention to differential effects by parental background.

Future research should pay greater attention to mechanisms as well as to implications of increases in MWs on broader measures of quality and quantity of children.

Data Availability

The data underlying this article will be shared on reasonable request to the corresponding author.

Footnotes

Notes

Muhammad Farhan Majid (corresponding author) is a Senior Economist at American Institutes for Research; his email address is [email protected]. Jere R. Behrman is the William R. Kenan Jr. Professor of Economics at the University of Pennsylvania; his email address is [email protected]. The research for this article was supported by the National Institutes of Health (NIH) grant# 1 R03 HD097425-01. The authors thank the journal reviewers and editor and participants at the 2019 Annual Meeting of the American Economic Association and the Seventh Annual Conference of the American Society of Health Economists for excellent comments and suggestions. The authors also thank Zoe Pham and Marwa Wazzi for excellent research assistance. A supplementary online appendix is available with this article at the World Bank Economic Review website.

References