Policy Research Working Paper 10009 Unequal Households or Communities? Decomposing the Inequality in Nutritional Status in South Asia Caitlin Brown Eeshani Kandpal Jean Lee Anaise Williams Development Economics Development Research Group April 2022 Policy Research Working Paper 10009 Abstract Half of all undernourished women and children in South a comparison of the effectiveness of targeting undernour- Asia are not found in the bottom 40 percent of wealth-poor ishment using household wealth, a community sanitation households. This paper quantifies the extent to which this infrastructure index, and, separately, the proportion of inequality in nutritional status arises within households improved toilets in a community. The findings show that versus between households. In contrast to previous litera- access to improved toilets, despite its relative simplicity, per- ture, it shows that between-household inequality explains forms almost as well as household wealth and better than the 3.5 times as much of the variation as does inequality within community sanitation index. These findings highlight that households. Within the household, gender, age, and birth (a) inequality between households within the same commu- order are key correlates of nutritional outcomes. At the nities is an overlooked but important driver of inequality household level and accounting for community-level factors, in nutritional status, and (b) community-level sanitation both an index of sanitation infrastructure and the presence infrastructure may be a better indicator of nutritional status of an improved toilet matter independently to household than more complicated household-level targeting measures. wealth for nutritional outcomes. The paper concludes with This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at ekandpal@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Unequal Households or Communities? Decomposing the Inequality in Nutritional Status in South Asia Caitlin Brown Eeshani Kandpal† Jean Lee‡ Anaise Williams§ JEL Codes: I14, I32, I38 Keywords: Nutritional outcomes, household wealth, intra-household inequality, sanitation, targeting, Demo- graphic and Health Surveys, South Asia Department of Economics, University of Manchester † Development Research Group, World Bank ‡ South Asia Chief Economist’s Office, World Bank § Johns Hopkins Bloomberg School of Public Health We thank Harold Alderman, Katy Bergstrom, Benoît Dercef, Francisco Ferreira, Jed Friedman, Emanuela Galasso, Camilla Holmemo, Purnima Menon, Berk Özler, Phuong Nguyen, Martin Ravallion, Krisztina Szabó, Dominique van de Walle, and Roy Van der Weide for helpful discussions and comments. We acknowledge financial support from the World Bank’s Research Support Budget and the South Asia Chief Economist’s Office. 1 Introduction A central challenge for successful antipoverty policy is to correctly identify benefit recipients. This is par- ticularly true of interventions designed to address malnutrition: while household wealth is often used to target nutrition interventions, previous work has found that undernourished individuals can be found in households that are not wealth poor (Brown et al., 2019, 2021). The existence of intra-household in- equality may mean that policies that use household-level measures for targeting will miss poor individuals who reside in non-poor households. However, given a binding budget constraint that precludes universal coverage and missing information on individual-level outcomes, the question remains as to whether tar- geting poor households will still reach a greater proportion of undernourished individuals than alternative targeting methods such as geographic targeting.1 In this paper, we attempt to disentangle the role of the household from that of the community in de- termining nutritional status, with the aim of understanding how policy can best reach undernourished individuals in the developing world.2 We decompose the variation in anthropometric outcomes to show that across South Asia, inequality between households and within communities explains more of the variation in outcomes than inequality within households. We explore the correlates of poor nutritional status to show that i) differences within households appear to stem predominately from age and gender, which could be explained by biological factors in addition to discrimination, and ii) sanitation infrastruc- ture, including the presence of an improved toilet, is a significant predictor of nutritional status even after accounting for household wealth. We highlight the potential gains from sanitation-based targeting over household-level methods for nutrition interventions, though we find that all targeted interventions studied still miss many undernourished individuals. Our focus in this paper is South Asia, the most densely populated region in the world and home to one- fourth of the world’s population, or almost two billion people. The region has experienced rapid economic change over the past three decades, with GDP per-capita rising from $380 (current USD) in 1995 to $1,960 in 2019 as well as an accompanying reduction in poverty, which fell from a rate of 40% at the turn of the century to less than 15% in 2016.3 However, despite massive increases in economic growth in the 1 Here, geographical targeting refers to selecting geographic areas to receive a policy, where all residents within that area have access to the policy benefits. Coady et al. (2004) provide a general overview of targeting methods and their implementation. 2 The importance of reaching undernourished individuals is a significant policy concern: combating undernutrition in developing countries has been a key component of the Millennium Development Goals and features prominently in the Sustainable Development Goals (World Bank, 2008). Of children under age five globally, 21.3% were stunted and 6.9% were wasted in 2019 (UNICEF/WHO/World Bank, 2020). 3 Indicators are drawn from the World Bank’s PovcalNet database, available here. 1 region, rates of undernourishment, particularly among children, remain stubbornly high (Jayachandran and Pande, 2012). South Asia has the second highest rates of stunting at 31.7%, preceded by East Africa at 34.5% and followed by Sub-Saharan Africa at 31.5% (UNICEF/WHO/World Bank, 2020). It also has the highest levels of child wasting, at 14.3% as compared to 6.7% in Sub-Saharan Africa. Two out of five of the world’s stunted children and more than half of all wasted children in the world live in South Asia. Our study draws from the Demographic and Health Surveys (DHS) data, and we therefore focus on countries with recently available data, namely Bangladesh, India, the Maldives, Nepal, Pakistan, and Sri Lanka. Table 1 lists key indicators for each country in our sample, which emphasize the heterogeneity of the region. India dominates in terms of population size and has a substantially higher average poverty rate relative to the other countries. In contrast, the Maldives has almost zero poverty at the current international poverty line of $1.90 per day; Sri Lanka and Pakistan also have relatively low poverty rates at 0.9% and 4.0%, respectively. Table 1: National Wealth Indicators Year Population (m) GDP per capita Gini index Poverty ($1.90) Poverty ($3.10) Bangladesh 2016 158.0 1401.6 32.4 14.5 50.0 India 2012 1266.0 1729.3 35.7 22.5 59.5 Maldives 2016 0.5 9209.3 31.3 0.0 0.2 Nepal 2010 27.0 592.4 32.8 15.0 48.4 Pakistan 2016 203.6 1368.5 33.5 4.0 32.4 Sri Lanka 2016 21.2 3886.3 39.8 0.9 9.7 Note: Data come from the World Bank’s Povcalnet database and the World Development Indicators. This paper makes three main contributions to the literature. First, we show that undernourished individ- uals are found in households across the wealth distribution in South Asia: around 70% of undernourished adults and children not found in poorest 20% of households, echoing similar findings from Brown et al. (2019) for Sub-Saharan Africa.4 Further, we show that obesity, including among children, is more likely to be found in wealthier households than in poorer ones. These findings contrast with existing literature that suggests wealth accumulation drives reductions in undernutrition (e.g., Headey et al., 2015; Headey, 2013; Hong et al., 2006; Behrman and Deolalikar, 1987) and raises the question whether the dual burden of malnutrition is likely to be a serious problem for the region. 4 Brown et al. (2021) perform a similar exercise for Bangladesh using household consumption per-capita and find comparable results. Here, we use the DHS’s wealth index as a proxy for household wealth; see Section 3 for further details. 2 Second, while the literature generally ascribes intra-household inequality in health or nutrition to in- equality in resource allocations within the household, we show that most of the variation in nutritional outcomes for adults and children appears to be found between households and within communities. Between-household inequality explains 3.5 times as much of the variation in nutritional status as does within-household inequality. We also find that within-community inequality (that is, differential out- comes within the same local communities) explains 4 times as much of the inequality as does between- community inequality, suggesting that substantially reducing undernutrition in the region may require community-level interventions in addition to household-level ones. Our finding also corroborates the notion of spatial poverty traps, which is the idea that certain areas are disadvantaged and explain a dis- proportionate amount of poverty (Kraay and McKenzie, 2014). Finally, an open question in the literature pertains to the identification of the alternatives to household- level targeting. Household wealth-based targeting through a proxy means test has often been considered to be the gold standard (Ravallion, 2016; Del Ninno and Mills, 2015). However, the literature also points to errors of both inclusion and exclusion, with the latter particularly salient in the presence of the sort of inequality in outcomes that we show here (Cowell, 2000, 2011). Typically the alternatives to wealth-based targeting also use household-level measures of well-being, including consumption (Brown et al., 2021) or econometric targeting methods (Brown et al., 2018). We show that in most South Asian settings—in particular, the larger, more unequal countries of India, Pakistan, and Bangladesh—a simpler community- based targeting metric, namely the proportion of improved toilets, performs comparably to targeting on household wealth in terms of inclusion and exclusion errors. This result is important as it provides an alternative proxy for deprivation that is easily measured and likely less expensive to update than household-level proxy means test (PMT) rosters. The rest of the paper proceeds as follows. In Section 2, we describe the data used. Section 3 presents key findings on nutritional outcomes and household wealth, while Section 4 explores the extent of variation in outcomes between and within households and communities. Section 5 considers the role of the household and the community in determining nutritional outcomes. Section 6 compares options for household- and community-based targeting. Section 7 concludes. 3 2 Data and Descriptive Statistics Our data on household wealth and nutritional indicators come from the DHS in Bangladesh, India, the Maldives, Nepal, Pakistan, and Sri Lanka. They include anthropometric outcomes for children 5 years and under, women between 15 and 49 years of age, and for some countries, men between 15 and (usually) 49 years of age. From the DHS, we also draw on individual- and household-level indicators such as sanitation measures, household composition, and local geographic characteristics. Table A1 in the Appendix lists the countries and sample sizes included in our analysis. To assess nutritional indicators, we use height and weight data from the DHS for men, women, and children. Following convention, we drop women who report being pregnant as well as those missing values for height or weight.5 For adults, we use body mass index (BMI) and an indicator for whether the person is underweight or not, defined as someone with a BMI value of 18.5 or less. For children, we use two different measures: height-for-age and weight-for-height z-scores. These measures reflect stunting and wasting, respectively (defined as having a z-score of less than –2). Both measures are important in child development, and generally stem from different causes: stunting is often an indicator of longer- term chronic undernutrition and is associated with poor economic conditions. Wasting, on the other hand, tends to reflect shorter-term deprivations, such as lack of food or illness that prevents nutritional absorption, for example. While wasting responds more rapidly to change, stunting is often irreversible; that is, children can gain weight quicker than they can grow tall (Reinhard and Wijeratne, 2000).6 Table 2 reports average values for the nutritional outcomes: across the region, 17% of men are under- weight, 14% of women are underweight, 30% of children are stunted, and 13% of children are wasted. Boys have slightly higher rates of stunting and wasting relative to girls, in line with Wamani et al. (2007) and Thurstans et al. (2020), for example. There is also substantial heterogeneity across countries, with India generally seeing the highest rates of undernutrition among women and children. Nevertheless, it is worth emphasizing that countries that have relatively low poverty rates, namely the Maldives, Pakistan, and Sri Lanka, still have relatively high rates of undernutrition, particularly among children. For example, the Maldives, which has almost zero extreme poverty (measured by the $1.90 per day poverty line), still 5 There is a question of whether to include lactating women; previous studies such as Brown et al. (2019) exclude them in their calculations. When we exclude lactating women, we find our results to be quantitatively similar. Given that 17% of our adult female sample is lactating, we chose to keep these women in our sample. For references, Table A2 list the proportions of pregnant and lactating women by country. 6 There is also an age component: wasting is most prevalent between 12 and 24 months of age, when dietary deficiencies and diarrheal diseases are more common, while the prevalence of stunting increases over time up to the age of 24 or 36 months (WHO, 1986). 4 has 15% of children stunted and 9% wasted; Sri Lanka has similar figures. Table 2: Summary Statistics for Nutritional Indicators Underweight Stunting Wasting Men Women Boys Girls Overall Boys Girls Overall Bangladesh - 18.57 36.88 35.55 36.24 15.08 13.58 14.36 India 19.72 22.85 38.84 37.87 38.37 21.90 20.11 21.04 Maldives 13.82 10.54 15.95 14.35 15.16 10.33 7.94 9.15 Nepal 18.49 17.18 35.81 35.79 35.80 9.54 9.91 9.72 Pakistan - 8.68 37.92 36.88 37.41 7.02 6.63 6.83 Sri Lanka - 8.99 17.96 16.49 17.25 15.41 14.77 15.10 Total DHS 17.34 14.47 30.56 29.49 30.04 13.21 12.15 12.70 Note: Data are drawn from the DHS. The table gives the average values of underweight in women and men between 15 and 49 years and stunting and wasting in children between 0 and 5 years. Underweight is defined as a BMI value of 18.5 or lower; stunting is a height-for-age z-score of –2 or lower; and wasting is a weight-for-height z-score of –2 or lower. Pregnant women and individuals without height or weight data have been dropped. Population weighting and sampling probability is used. To explore the relationship between community-level factors and nutritional outcomes, we use two other data sources: the District Level Households and Facility Survey (DLHS) and the Service Provision Assesss- ments (SPA), both of which are collected by the DHS. Our community-level analysis uses four indexes: community wealth (DHS in all countries), community sanitation (DHS in all countries), community ser- vices (DLHS in India), and health facility quality (DLHS in India and SPA in Bangladesh and Nepal). The community wealth index was created by aggregating the DHS household wealth index to the community level (by community, we mean DHS sampling cluster). To create the other three indexes, we use principal component analysis to incorporate several categorical variables into one index. The sanitation index pulls together DHS variables regarding drinking water source, whether anything is done to make water safe, hand washing station availability, and toilet type at the household level. To create the community-level sanitation index, we take the average of the household sanitation index at the community level. The community services index draws on variables from the DLHS village questionnaire (such that we can only construct it for India). The index is composed of community health facility availability, skilled health provider availability, education facility availability, electrification, drinking water source, drainage, road accessibility, and the presence of a mobile clinic. The facility quality index uses data from the DLHS on district hospitals and the SPA in Bangladesh and Nepal on health facilities, which include district hospitals as well as other health care providers. The variables to construct the health facility quality index are cleanliness levels and the numbers of obstetricians, nurses, and antenatal maternity nurses. 5 To match these community-level variables to the DHS anthropometric and wealth data, we first merge the DLHS district hospital data using state and district name. For the DLHS village-level data and the SPA facility data, we merge facilities to clusters based on GPS coordinates available in both data sets. Here, the nearest facility or village (in the DLHS) with a 50-km radius was matched with each cluster. For India, 540 DLHS hospitals matched with 634 DHS districts, and 494 DLHS villages matched with 7,473 DHS clusters. For the SPA data, 565 of 600 and 382 of 383 DHS clusters in Bangladesh and Nepal, respectively, were matched with SPA health facilities. 3 Nutritional Outcomes and Household Wealth in South Asia The existing evidence regarding the relationship between household income on nutritional outcomes is mixed, and particularly so for South Asia. Well known is the Asian enigma: children in South Asia are shorter, on average, relative to children elsewhere, including those who are poorer, on average (Ramalin- gaswami et al., 1997; Nubé, 2009). Deaton and Drèze (2009), for example, find that per-capita caloric intake is declining in India despite rising incomes across the country. Globally, the income effects on nutritional outcomes have been found to be modest (and often close to zero), particularly in the short run (Behrman and Deolalikar, 1987; Haddad et al., 2003; Smith and Haddad, 2015). On the other hand, there is also a large literature that demonstrates evidence of a wealth effect; that is, the positive relation- ship between nutritional status and household welfare.7 Here, we focus specifically on understanding the relationship between malnutrition (both undernutrition and obesity) and household wealth. 3.1 Undernutrition and Household Wealth We begin by asking how undernourished individuals are distributed across the welfare distribution in South Asia; that is, do most undernourished individuals live in poor(er) households? While the DHS does not include any questions that can be used to directly measure household wealth (such as the value of the household’s assets) or welfare (the DHS does not have a consumption module), it does provide a country-specific household wealth index based on household assets and living conditions (see Filmer and Pritchett, 2001 for details). As the index itself has little economic meaning, we use the index values to construct wealth percentiles, ranking households from poorest to richest. From this, we can construct 7 See, e.g., Behrman and Deolalikar, 1987; Ravallion, 1990, 1992; Pritchett and Summers, 1996; Headey, 2013). For South Asia, Hong et al. (2006) find that children in the poorest 20% of households in Bangladesh are more than three times as likely to suffer from stunting as children from the top 20% of households. This echoes similar findings from Headey et al. (2015); Headey and Hoddinott (2015); Headey et al. (2016) that wealth accumulation, in addition to maternal education and sanitation, is one of the biggest drivers behind the reduction in undernutrition in South Asian countries. 6 concentration curves, which plot the cumulative share of household undernourishment indicators by cu- mulative household wealth percentile (Wagstaff, 2000; Wagstaff et al., 2014; Bredenkamp et al., 2014). The more concave the concentration curve, the higher the proportion of undernourished individuals who are found in wealth-poor households. Figure 1 plots undernutrition concentration curves for women and children across the three nutritional outcomes. As expected, given that stunting is the result of longer term deprivations, there is more cur- vature for stunting relative to the incidence of underweight and wasting. However, it is clear that many undernourished individuals are not in the lower ends of the wealth distribution. There are also differ- ences in curvature across countries, with the Maldives and Nepal showing a smaller correlation between household wealth and nutritional outcomes relative to other countries. Looking at specific points of the distribution (Table 3), we find that 24% of undernourished men and 29% of undernourished women fall in the bottom 20% of the wealth distribution along with 30% of stunted children and 25% of wasted children. The poorest 40% of households have 47% of underweight men, 52% of underweight women, 54% of stunted children, and 48% of wasted children. Despite these findings, we still find evidence of the wealth effect in nutritional outcomes: regressing underweight in women, underweight in men, stunting in children, and wasting in children on the house- hold wealth index yields significant (negative) coefficients (though the coefficients are relatively small in magnitude, at –0.055, –0.046, –0.069, and –0.020, respectively). Nevertheless, most undernourished in- dividuals in South Asia do not live in the poorest end of the distribution: around 50% of undernourished women and children are not found in the poorest 40% of households. These numbers are similar to those found in other work on the topic for Africa (Brown et al., 2019) and Bangladesh (Brown et al., 2021). 3.2 Robustness Our results are robust to several potential confounders, for example, measurement error in nutritional outcomes or the wealth index.8 The first issue we address is in regard to the appropriateness of the cut-off point for undernourishment for South Asia. For children, the two generally accepted international mea- surements of undernourishment are based on height-for-age and weight-for-height z-scores, where the reference population is based on children from a wide variety of ethnic and socioeconomic backgrounds 8 Several of these checks have also been covered in Brown et al. (2019, 2021), where further detail is provided. 7 (a) Bangladesh (b) India (c) Maldives (d) Nepal (e) Pakistan (f) Sri Lanka Note: DHS data. The graphs show concentration curves for the cumulative proportion of women who are underweight and children ages 0–5 who are stunted and wasted at each household wealth percentile. Observations with missing values and pregnant or lactating women have been dropped. The Stata command glcurve is used to construct the curves. Figure 1: Undernutrition Concentration Curves 8 Table 3: Proportion of Undernourished Individuals in Poorest 20% and 40% of Household Wealth Poorest 20% of Households Underweight Stunting Wasting Men Women Boys Girls Overall Boys Girls Overall Bangladesh . 0.32 0.33 0.29 0.31 0.25 0.30 0.27 India 0.26 0.29 0.34 0.36 0.35 0.29 0.30 0.30 Maldives 0.25 0.18 0.25 0.22 0.23 0.20 0.19 0.20 Nepal 0.22 0.19 0.27 0.30 0.28 0.19 0.18 0.19 Pakistan . 0.44 0.30 0.36 0.33 0.30 0.33 0.31 Sri Lanka . 0.35 0.31 0.32 0.32 0.23 0.28 0.26 Total 0.24 0.29 0.30 0.32 0.30 0.24 0.26 0.25 Poorest 40% of Households Underweight Stunting Wasting Men Women Boys Girl Overall Boys Girls Overall Bangladesh . 0.58 0.54 0.52 0.53 0.51 0.47 0.49 India 0.51 0.54 0.58 0.61 0.60 0.52 0.53 0.53 Maldives 0.45 0.35 0.50 0.45 0.48 0.50 0.33 0.43 Nepal 0.45 0.43 0.49 0.55 0.52 0.41 0.38 0.40 Pakistan . 0.64 0.54 0.58 0.56 0.54 0.61 0.57 Sri Lanka . 0.60 0.52 0.54 0.53 0.48 0.50 0.49 Total 0.47 0.52 0.53 0.54 0.54 0.49 0.47 0.48 Note: Data are drawn from the DHS. The table gives the conditional probabilities of being undernourished given that the individual lives in a household in the poorest 20% and 40% of the household wealth distribution. Population weighting and sampling probability is used. internationally (Waterlow, 1972; Waterlow et al., 1977; WHO, 1986).9 For adults, generally no reference scales (such as z-scores) are used, and the BMI cut-off point of 18.5 is based on the level at which adults begin suffering from chronic energy deficiency (see, e.g., James et al., 1988; James and Francois, 1994; Kurpad et al., 2005). Nevertheless, for both children and adults, the cut-off points for being classified as undernourished (i.e., underweight, stunted, or wasted) are essentially arbitrary, where lower values of the respective mea- surements indicate more severe undernutrition.10 As such, we also consider conditional probabilities for 9 Habicht et al. (1974) find that ethnic differences in height and weight for nourished preschool children are relatively small; social background, on the other hand, matters substantially. 10 More specifically, to our knowledge, there is no scientific literature that indicates that a z-score equal to 90% and 80% of the reference median of height-for-age and weight-for-height distributions physiologically implies undernourishment; rather, it seems to be a target for policy that encourages healthy childhood growth. Waterlow et al. (1977), among others, provide discussion on the topic. 9 severely undernourished individuals: severe stunting is defined as three standard deviations below the median for height-for-age, and severe wasting is a z-score of –3 or lower for weight-for-height; for adults we use a BMI of 17 or lower. Table A3 lists the conditional probabilities for severe undernourishment. As expected, the wealth effect is stronger, particularly for stunting, but a large proportion of severely undernourished individuals remain outside the bottom end of the wealth distribution. While child weight-for-height z-scores and adult BMI will be robust to measurement errors in age, the same is not true for height-for-age z-scores, which explicitly depend on the correct measurement of a child’s age. The likelihood of mismeasurement is particularly problematic for very young children, whose height may be difficult to measure and even slight discrepancies in months of age can yield very different nutritional conclusions (Ulijaszek and Kerr, 1999; Larsen et al., 2019; Agarwal et al., 2017). Similarly, in regards to BMI, teenagers are still experiencing growth and may not be representative of adult nutritional status more generally. We therefore redo our conditional probabilities excluding children less than 18 months of age and adults less than 18 years of age (Table A4). We do not find this to substantially alter our main findings. Measurement in the wealth index is another concern. To account for this, we regress the wealth index and all the components of the wealth index on the nutritional outcomes for adults and children, essentially re-weighting the value of the wealth index. We can then use the predicted values from these regressions and assign individuals as undernourished if this value is below the standard cut-off point for undernour- ishment. After using this variable to calculate the conditional probabilities (see Table A5), we arrive at very similar results to our main findings. Lastly, we can verify whether nutrition-related illnesses show similar patterns to what we observe above. We construct concentration curves for whether the child has anemia and whether the child reports having a fever or diarrhea in the past two weeks. Figure A1 provides curves for each country (note, however, that not every country contains information for each variable), which, in line with the nutrition-based curves, are relatively flat. We also find a strong correlation between health outcomes for mothers and their children: Table A6 shows that underweight and anemic mothers are more likely to have stunted, wasted, or anemic children or have children who report having had diarrhea in the past two weeks (on the other hand, fever is not correlated with the child’s mother’s nutritional status and has a small negative correlation with mothers who are found to be anemic). 10 3.3 Obesity and Household Wealth In addition to undernutrition, the rising incidence of obesity among adults is fast becoming a concern, particularly in Asia (Yoon et al., 2006; Ramachandran and Snehalatha, 2010; Helble and Francisco, 2017). In our data, defining obesity as an adult with a BMI of 30 or greater, less than 5% of adults are obese in Bangladesh, India, and Nepal; however, we find much higher numbers for the Maldives (20% of women and 8% of men), Pakistan (19% of women), and Sri Lanka (12% of women). For children, obesity seems to be of a lesser concern across countries: using similar data to us, Bishwajit and Yaya (2020) find that less than 2% of children in South Asia are obese.11 In this section, we therefore concentrate on the relationship between adult obesity and household wealth. A priori, the effect of household wealth on the incidence of obesity is unclear. On the one hand, the poorer the household, the more constrained they are in the quantity and quality of food they can consume, in addition to affording other necessary inputs (such as clean water and proper sanitation infrastructure) for good health. A large literature shows that poorer households tend to consume lower quality calo- ries, driven in part by the higher relative prices of “healthy” foods (Headey and Alderman, 2019).12 On the other hand, existing work has found a positive relationship between obesity and household wealth, particularly for South Asia (Bishwajit, 2017; Al Kibria et al., 2019; Ahmad et al., 2020). In Figure A2, we plot the incidence of obesity at each household wealth percentile by country. In contrast to our findings on undernutrition, we see a much higher incidence of obesity at higher wealth percentiles: in Pakistan, for example, around 5% of women are obese at the lowest household wealth percentiles, while the incidence is more than 30% for the highest percentiles. Countries with a relatively low incidence of obesity overall, such as Bangladesh and Nepal, tend to have even higher concentrations of obesity among the top 40% of households. These findings suggest that for South Asia, adult obesity is much more likely to be an issue among wealthier households. 11 Other studies using smaller and non-representative samples such as Kaur et al. (2008); Jafar et al. (2008) and Karki et al. (2019) have found higher rates of childhood obesity in the region, which seems to be highly concentrated among wealthier population groups. 12 Ravallion (2016) discusses potential causes for the relationship between poverty and obesity in the United States, for example, the existence of “food deserts,” proximity to fast food restaurants, and the changing nature of work toward more sedentary occupations (p. 351–353). Hirvonen et al. (2020) find the cost of a standard healthy diet is not affordable for most of the world’s poor; that is the cost of the reference diet exceeded household per-capita income for at least 1.58 billion people. 11 4 Variation in Nutritional Outcomes between Groups In the previous section, we found that around 70% of undernourished adults and children are not in the poorest 20% of households in the wealth distribution, and around 50% of undernourished individuals are not in the poorest 40%. However, these numbers do not tell us whether undernourished individuals are in the same households (i.e., everyone in the household is or is not undernourished) or whether these individuals are spread across households (i.e., only some people in the household are undernourished). To answer this, we first check the extent to which the incidence of undernourishment is correlated within households. Table 4 shows that households with at least one undernourished man, woman, or child are significantly more likely to have other undernourished household members (the exception is for within- household stunting and wasting, which are not significantly correlated). Table 4: Correlation Matrix: Within-Household Nutrition At least one man underweight At least one woman underweight At least one child stunted At least one woman underweight 0.1805*** At least one child stunted 0.1089*** 0.1071*** At least one child wasted 0.0603*** 0.1055*** 0.0025 Note: Data are drawn from the DHS. We next create a dummy variable equal to one if an adult is underweight or if a child is either stunted or wasted and then calculate the average value of this variable for each household; if everyone in the household is (or is not) undernourished, the mean undernourishment will equal one (or zero) and take a value between zero and one if there is any variation in nutritional status within the household. We find that 63% of households have no undernourished household members. The remaining 37% of households have at least one member who is undernourished, with 8% of households having all members who are undernourished. Figure 2 plots average household undernourishment using the aforementioned undernourishment indi- cator by household wealth percentile. Here, the wealth effect on nutrition is clear, yet a non-negligible share of household members are undernourished even at the higher ends of the wealth distribution. A similar relationship is evident when we consider the share of households that have at least one house- hold member who is undernourished—while we expect to see a large share of households in the bottom wealth percentiles with undernourished individuals (particularly given that poorer households tend to be 12 larger), even at the 80th wealth percentile (within a county), almost 30% of households have at least one household member who is undernourished. (a) Average household undernourishment (b) At least one member undernourished Note: DHS data. The first figure shows the relationship between household wealth and average household nutritional status, which is the household average of an indicator variable equal to one if an adult is underweight or a child is stunted or wasted and zero otherwise. The second is an indicator variable equal to one if at least one household member is undernourished and zero otherwise. Observations with missing values and pregnant women have been dropped. Figure 2: Average Nutritional Status by Wealth Percentile To explore further how much variation there is in nutritional status between households, Figure A3 plots the distribution of undernourishment at the (a) household- and (b) community-level. As expected, we find a much greater dispersion of undernourishment across communities than across households: while more than 60% of households have no undernourished members, only 6% of communities can say the same. On the other hand, we find that there are fewer communities with “high” rates of undernourishment; for example, 10% of communities have an average incidence of undernourishment greater than 40%, while only 20% of households do. To quantify how much of the differences that we observe in nutritional outcomes occur within versus between households, we can use an inequality measure that is perfectly decomposable (into between and within components) while also satisfying other desirable properties, such as scale and population size independence, which are important in our setting given the different sample sizes and outcomes used across adults and children.13 We draw from the class of generalized entropy measures, which take the form n θ 1 1 yi GE (θ ) = −1 θ (θ − 1) n i =1 ¯ y / (0,1), where yi is equal to the outcome for individual i who is part of a population with size for θ ∈ 13 For a review of inequality measures, see, for example, Cowell, 2000, 2011. 13 n and ¯ y is the average outcome in the population considered. We use two common cases, namely the n yi yi Theil index and the mean log deviation (MLD), which are equal to GE (1) = 1 n i =1 ¯ y ln ¯ y and G E (0) = 1 n yi n i =1 ln ¯ y , respectively.14 Letting y in this setting represent nutritional outcomes, decomposing the MLD yields between and within components of the total index (see the Appendix in Brown et al., 2021 for further details): J nj 1 M LD = ln ¯ y− ln yi j n j =1 i =1 J jn J 1 ¯ yj 1 ¯ y = ln + n j ln , n j =1 i =1 yi j n j =1 yj Within Between where each individual i belongs to group (e.g., a household) j . Each group j has a total of n j members and an average nutritional outcome of y j . J and n are the total number of groups and individuals, respectively. Two main issues arise: 1) nutritional outcomes are different for adults and children and 2) our under- nourishment indicator variable cannot be used with either the Theil index or the MLD. Given this, we first re-scale the height-for-age and weight-for-height z-scores (note that the Theil index and the MLD are scale-independent inequality measures so re-scaling will not affect the results) and estimate overall in- equality as well as the between and within components separately for each outcome. Given the different outcomes for adults and children, and that men’s nutritional status is only observed for some countries, we separate our decompositions between adults and children (and for adults, between male and females). Table 5 lists the overall values for the Theil index and the MLD as well as the decomposition of the MLD for between and within households and between and within communities (Table A7 provides estimates by country). We find similar estimates for our inequality measures—women’s BMI and children’s height-for-age z- scores exhibit the most inequality and weight-for-height the least. Nevertheless, the inequality decompo- sitions are remarkably consistent across the four groups: between-household inequality is substantially larger than within-household inequality, and within-community inequality is substantially larger than between-community inequality.15 14 Generally, Theil’s index is more sensitive to differences at the top of the distribution, while MLD is more sensitive to differences at the lower end of the distribution. 15 Similar results are found when we decompose the Theil index. The results are available upon request. 14 Table 5: Inequality Decompositions Households Communities Theil MLD Between (%) Within (%) Between (%) Within (%) Women’s BMI 0.018 0.017 78 22 19 81 Men’s BMI 0.015 0.014 79 21 22 78 Height-for-age 0.019 0.019 80 20 22 78 Weight-for-height 0.012 0.012 81 19 24 76 Note: Data are drawn from the DHS. Between and within refer to decompositions of mean log deviation (MLD). Pregnant women and individuals without height or weight data have been dropped. BMI is measured for adults between 15 and 49 years. Men’s BMI is only available for India, Maldives, and Nepal. Height-for-age and weight- for-height z-scores are for children five years and younger. Coefficients of variation squared coefficients are 0.019, 0.015, 0.019, and 0.012. 5 Household- and Community-Level Variation in Outcomes The above discussion suggests that within-community factors seem to be important for nutritional out- comes, though household-level factors certainly seem non-negligible. In Table A8, we explore how com- munity and household factors might covary with predicted undernourishment using a Kitagawa-Blinder- Oaxaca decomposition on residual undernourishment after accounting for observable household and in- dividual characteristics that might be correlated with undernourishment. These characteristics include age, sex, educational attainment of both the individual and the household head. The decomposition of residuals shows that household wealth covaries with predicted undernourishment to a comparable de- gree as a simple binary measure of community-level sanitation– the proportion of improved toilets in the village. That difference in undernourishment rates between individuals in the bottom 20% of communi- ties with improved toilet access is 5% while the same difference among individuals in the bottom 20% of household wealth is 7%, suggesting that community level disease environment factors may explain as much undernourishment as household wealth. We return to a comparison of these in the two variables as metrics for targeting undernourished individuals in Section 6. The rest of this section delves further into various correlates of the household- and community-level variation in outcomes to understand the drivers of undernourished status. We first explore the household-level factors that covary with nutritional status and then explore community-level factors. 5.1 Intra-Household Inequality An extensive literature details intra-household inequality in resources and outcomes and discrimination against certain individuals within the household. Lambert et al. (2014) find substantial consumption in- 15 equality between men and women in Senegal, which is largely attributable to differences in education and inheritance. Dercon and Krishnan (2000) use adult nutrition data to show that women in poorer households are more likely to bear the brunt of a negative shock than men. Other work by Calvi (2020) shows that women’s household bargaining power and ability to access household resources diminishes with age, meaning older women have higher poverty rates as compared to their male household coun- terparts. Son preference has also been well documented in Asia, resulting in a large and growing gender imbalance in countries such as China and India.16 Birth order has also been found to affect the alloca- tion of resources within households: in India, Behrman (1988) find a bias in the allocation of nutrients favoring earlier-born children, and Jayachandran and Pande (2017) show a relative height disadvantage for later-born children that appears to be driven by a preference for eldest sons. Underlying any discussion of intra-household inequality in health outcomes are varying sex- and age- based biological predispositions.17 Wells (2000), for example, show naturally higher rates of male infant mortality than female mortality, with female infants having stronger vitality related to nutrition, growth, and resilience to environmental stress.18 Weight generally tends to be an increasing function of age until mid-adulthood (Chumlea et al., 2002). Given that height changes more slowly over time, older children are therefore less likely to be wasted but more likely to be stunted. Gender differences in nutritional out- comes persist through adulthood, driven by factors such as the retention of postpartum weight (Johnston, 1991; Olson et al., 2003) and societal norms (Garawi et al., 2014). In Table 6, we investigate how the above characteristics noted in the literature are associated with dif- ferences in nutritional outcomes within the household using a fixed effects regression model. After ac- counting for time-invariant household-level factors, we find that women have a higher BMI but are also more likely to be underweight, suggesting an inverted U relationship between gender and adult nutri- tional outcomes. In line with existing evidence, girls have higher height-for-age and weight-for-height z-scores relative to boys, and are less likely to be stunted or wasted. Age is an increasing function of BMI, and older adults are less likely to be underweight. Given that children grow taller slower than they grow older, older children are more likely to be stunted; but, given that weight typically increases with age, 16 In the past, this preference has led to excess female infant mortality as well as poorer health outcomes for girls as compared to boys; see, for example Das Gupta (1987); Behrman and Deolalikar (1988); Garg and Morduch (1998); Das Gupta and Shuzhuo (1999); Pande (2003); Pande and Yazbeck (2003); Oster (2009); Willis et al. (2009) and Jayachandran and Kuziemko (2011). 17 We note here the important distinction between sex, which is biological, and gender, which is a socially determined construct; see, for example, Heidari et al. (2016) for further reading. Each of these may have differential impacts on health outcomes. 18 This is a direct test of the Trivers-Willard theory, which states that natural selection necessarily means that females in poorer conditions produce lower ratios of males to females (Trivers and Willard, 1973). Excess male infant mortality has also been documented in Naeye et al. (1971); McMillen (1979); Garenne (2003), and Drevenstedt et al. (2008), among many others. 16 they are less likely to be wasted. Table 6: Intra-Household Inequality in Nutritional Outcomes (1) (2) (3) (4) (5) (6) BMI Underweight Height-for-age Stunted Weight-for-height Wasted Female 0.33*** 0.01*** 0.10*** -0.02*** 0.06*** -0.02*** (18.35) (3.47) (7.77) (-4.99) (5.68) (-6.40) Age 0.13*** -0.01*** -0.23*** 0.04*** 0.04*** -0.03*** (181.93) (-119.67) (-53.17) (33.81) (10.67) (-30.28) Education 0.02*** -0.00*** (12.67) (-10.50) Son preference 0.02 -0.01 -0.04 0.01 (0.67) (-0.76) (-1.26) (1.23) First born 0.16*** -0.05*** 0.02 0.00 (9.34) (-8.97) (1.64) (0.02) Male first born -0.03 0.00 -0.02 -0.00 (-1.49) (0.39) (-0.79) (-0.11) Constant 17.21*** 0.51*** -1.21*** 0.35*** -1.05*** 0.27*** (451.93) (126.98) (-92.24) (84.70) (-93.18) (79.76) N 504150 504150 120173 120173 120173 120173 R2 0.634 0.532 0.622 0.579 0.607 0.559 HH FE Yes Yes Yes Yes Yes Yes Note: Data are drawn from the DHS. Age is measured in years for adults and children. Education is measured in years. Son preference is defined as if the mother prefers to have more boys than girls. Household fixed effects are included in all regressions. Standard errors are clustered at the community-level. t-statistics are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01 In regards to socio-demographic factors, adult education is found to be positively associated with BMI and is negatively associated with being underweight – given that household-level factors such as wealth are controlled for, this may suggest that either these individuals have more bargaining power to access household resources or more knowledge regarding health behaviors. We find no significant association of son preference for child nutritional outcomes; however, we do find that first-born children have higher height-for-age z-scores and are less likely to be stunted (no effects are found for weight-for-height and wasting). Including a gender interaction term yields insignificant coefficients; that is, neither male nor female first-born children are driving these results. This is consistent with Brown et al. (2021), who find that first-born children in Bangladesh have higher shares of household resources relative to later- 17 born children with no significant gender difference. It is also consistent with findings from a broader literature on child outcomes demonstrating a firstborn height advantage due to, among other things, parental behavior (Price, 2008; Lehmann et al., 2018) and early maternal investments (Buckles and Kolka, 2014). 5.2 Community-Level Factors A large literature highlights how how community-level factors, such as access to public health facilities and sanitation infrastructure may affect women’s and children’s nutritional status. Village-level sanita- tion programs have been found to have significant impacts on child stunting, even among children of households where no household changes were made (Hammer and Spears, 2013). Several studies have explained the Asian enigma at least in part by the high levels of open defecation in South Asia; over half of global open defecation occurs in India, for example (Coffey et al., 2015). Even when controlling for literacy and water supply, the association between stunting and sanitation remain (Spears, 2020). In- deed, Spears (2020) suggests that most if not all of excess stunting in India can be explained by open defecation. Geruso and Spears (2018) identify community sanitation and open defecation to be a main determinant factor explaining why Muslim infant mortality is lower than Hindu infant mortality in India despite increased wealth among Hindu families. It is less clear how wealth and sanitation matter independently to undernutrition. In Table 7, we explore this variation at both the household and community level with undernutrition using a regression model with different levels of fixed effects. For undernutrition, we again use our indicator equal to one if an adult is underweight or if a child is stunted or wasted; our sanitation and village services indices are described in Section 2. Here, the community wealth and sanitation variables are calculated by taking the community-level average of the household-level equivalents. We also include a separate and important measure of sanitation, the presence of an improved toilet, which is more straightforward to measure and observe in comparison to the sanitation index.19 We also include community-level education, as measured by the share of childbearing age women with at least a primary education. As expected, household wealth is significantly and negatively correlated with undernutrition across all specifications. However, we also find a significant effect of our household sanitation index in addition to 19 As per DHS standards, an improved toilet is defined as (1) flush latrines (flushed to piped sewer systems or septic tanks), (2) ventilated pit latrines or those with a slab, or (3) compostable toilets. Our sanitation index includes household toilet type, so this variable should be seen as an independent effect of the type of toilet a household has. 18 wealth, indicating that sanitation is providing additional explanatory power for undernutrition. Further- more, we also find a strong effect for whether or not the household has an improved toilet, and this is also in addition to the sanitation index. In the specifications with community fixed effects (columns (1)–(4)), household wealth and sanitation are both important correlates with undernutrition within communities. Moving to our specifications with district fixed effects (columns (5) and (6)), both community and house- hold wealth are important covariates of undernutrition; however, community wealth becomes insignifi- cant once average female education in a community is included, which may highlight the important role of information for nutritional status. Interestingly, after accounting for household-level wealth and sani- tation and community wealth and education, community sanitation yields a weakly positive coefficient, while the share of the community with an improved toilet plays a negligible role. Finally, in column (7), we consider district-level variables such as village services and health facility quality that are available only for a subset of countries at the district level.20 Neither index appears significantly associated with the incidence of undernutrition, and except for community education, all other covariates maintain their direction and significance as in the other models. 6 Targeting Undernourished Individuals: Community or Household? In the previous section, we found that household sanitation in addition to wealth (both at the household and community levels) is significantly associated with undernutrition. Here, we ask how this may in- form targeting of health policy. Broadly, there are two related components to targeting. The first is the unit at which programs should be targeted, for example, whether a program should include individual-, household-, or community-level targeting. Often this is driven by the type of program to be implemented, that is, whether it is a program for individuals, households, or communities. The second component is the information available for targeting purposes, which is often restricted to the household or community level. For example, targeting undernourished individuals would require information on nutritional status (or a highly correlated proxy variable) to determine eligibility, which is unlikely to be available in large populations. As a result, nutrition policies are often targeted using a hybrid of household and geographic targeting, where all individuals in a household or community are exposed to at least some components of the program.21 For example, conditional cash transfer schemes are targeted at the household level; on 20 Village services are available for India, Nepal, and Bangladesh. We use the SPA, which is only available for the latter two, and DLHS for India. 21 Note that our use of the term community-level targeting is aligned with geographic targeting, where policymakers target based on geographic location. This differs from community-based targeting, where leaders in a community are tasked with allocating a program to households or individuals within 19 Table 7: Multilevel Effects on the Relationship between Wealth, Sanitation, and Undernutrition (1) (2) (3) (4) (5) (6) (7) Household wealth -0.04*** -0.04*** -0.04*** -0.04*** -0.04*** -0.04*** -0.04*** (-87.84) (-75.60) (-73.29) (-69.30) (-84.95) (-67.02) (-26.40) Household sanitationa -0.05*** -0.04*** -0.03*** -0.01 (-12.66) (-8.01) (-7.04) (-1.03) Improved toilet -0.02*** -0.01*** -0.01*** -0.01*** (-13.41) (-9.35) (-9.04) (-3.54) Community wealth -0.01*** -0.00 -0.00 (-15.80) (-0.28) (-0.34) Community sanitationa 0.02* -0.03 (2.25) (-1.56) Community education -0.06*** -0.05*** (-17.49) (-5.81) Share of community with improved toilet -0.00 -0.01* (-1.23) (-1.70) Village servicesd -0.01 (-1.54) Health facility qualityc -0.02 (-1.59) Individual controls Yes Yes Yes Yes Yes Yes Yes Community fixed effects Yes Yes Yes Yes No No No District fixed effects No No No No Yes Yes No State fixed effects No No No No No No Yes N 1083053 1079826 1083053 1079826 1044752 1025727 171209 R2 0.198 0.199 0.199 0.199 0.161 0.161 0.151 Note: Data are from the DHS. Age and education is measured in years. Education is measured in years. Data for health facility quality are only available for India (DLHS survey), Bangladesh (SPA survey), and Nepal (SPA survey). a) Logged index: water source, treatment and distance, hand washing unit, toilet type; b) logged index: facility availability, distance to district hospital, distance to public health facility, education availability, drinking water source, hours of electrification, facilities connected to all-weather roads, number of family welfare camps, skilled provider availability, program availability; c) logged index: facility cleanliness measures and number of obstetricians, nurses, and antenatal maternity nurses; d) logged index: community health facility availability, skilled health provider availability, education facility availability, electrification, drinking water source, drainage, road accessibility, and the presence of a mobile clinic. Individual controls include age in years, sex, education in years (for children, this is mother education), age of household head, household size, and sex of household head. t-statistics are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01 20 the other hand, larger infrastructure programs such as clean water and sanitation initiatives tend to be at the community level.22 In this section, we focus on two different levels of targeting, namely household- and community-level targeting. We use three feasible targeting variables for which we have data from all countries in our sample—(1) the community services index, (2) the community sanitation index, and (3) the share of households with an improved toilet in the village. Our aim is to compare both the level of targeting as well as the variables used for targeting.23 Here, we treat the wealth index variable as our benchmark and the sanitation variables in comparison to that. Wealth, in practice, will not be observable in large populations; our sanitation variables, on the other hand, serve a more practical purpose; for example, whether or not a household has an improved toilet is straightforward to measure. To compare targeting performance between our household- and community-level measures, we first con- struct concentration curves that compare the share of undernourished, underweight, stunted, or wasted individuals targeted using one of the three metrics to the share of individuals covered using household wealth-based targeting. Figure 3 shows the results using our indicator for undernourishment, which is equal to one if a man or woman is underweight or if a child is stunted or wasted. We first find that house- hold wealth is more effective at covering the share of the malnourished individuals than the community services index. On the other hand, our index for community-level sanitation infrastructure performs as well as household wealth. However, constructing a sanitation index may be data intensive and introduce a new source of measurement error. As a result, we explore a third alternative for targeting, which is the share of households in the community with an improved toilet. This indicator performs as well as the sanitation index, suggesting that community sanitation infrastructure—including a simple categorical measure—may be a feasible alternative to household wealth for targeting nutrition interventions. Figure A4 provides concentration curves for the three targeting variables separately for underweight adults as well as stunted and wasted children. The results are generally similar, with the exception of child wasting, for which all three indices significantly outperform household wealth as a targeting measure. Next, we calculate inclusion and exclusion errors, focusing on the bottom 20% of the distribution. In the community. 22 Such geographic targeting can be sensible in the presence of spatial poverty traps: as noted by Kraay and McKenzie (2014), in many low- and middle-income countries a few regions or areas explain a disproportionate share of national poverty. 23 Note that here we observe these variables with precision because these are representative sub-samples—at scale, there is likely to be substantial measurement error as these metrics are typically unobservable in large populations. 21 (a) Community services (b) Community sanitation (c) Share improved toilets Note: DHS data. The graphs show concentration curves for the cumulative proportion of undernourished adults and children, where undernourishment is equal to one if a man or woman is underweight or a child is stunted or wasted. Figure 3: Targeting Undernourishment Using Household Wealth versus Community-Level Indicators this context, inclusion errors refer to the proportion of those targeted who are not undernourished, and exclusions errors quantify the proportion of undernourished who are not targeted. Generally, the broader the coverage (where coverage refers to the proportion targeted), the higher the inclusion errors and the lower the exclusion errors.24 We focus here on the bottom 20% of households and communities. Table 8 lists inclusion and exclusion errors for the three targeting levels and metrics that we use as well as coverage rates and the proportion of the sample that we ideally would like to reach (i.e., the proportion undernourished). Note that for the improved toilet indicator, we exclude both the Maldives and Sri Lanka, as these countries have almost universal usage of improved toilets; on the other hand, less than 60% of our sample for India have access to an improved toilet. Given its indicator status, we cannot create household-level percentiles, but for comparison purposes we include this variable as a targeting outcome in itself; specifically, whether or not the household does not have an improved toilet, along with the share 24 See Brown et al. (2018) for further discussion regarding inclusion and exclusion errors. 22 of households within the community that do have an improved toilet (whereby we target the communities with the lowest shares). Table 8: Inclusion and Exclusion Errors for Household- and Community-Level Targeting of Bottom 20% Targeting on Wealth Household-level Community-level % Undernour Coverage Incl. Errors Excl. Errors Coverage Incl. Errors Excl. Errors Bangladesh 0.266 0.199 0.590 0.694 0.197 0.620 0.719 India 0.285 0.196 0.571 0.705 0.207 0.595 0.705 Maldives 0.104 0.174 0.857 0.760 0.195 0.884 0.783 Nepal 0.211 0.179 0.733 0.774 0.197 0.746 0.764 Pakistan 0.243 0.198 0.601 0.674 0.230 0.587 0.541 Sri Lanka 0.157 0.205 0.765 0.694 0.206 0.777 0.708 Total 0.277 0.195 0.582 0.706 0.207 0.606 0.706 Targeting on Sanitation Household-level Community-level % Undernour Coverage Incl. Errors Excl. Errors Coverage Incl. Errors Excl. Errors Bangladesh 0.266 0.202 0.654 0.736 0.201 0.663 0.745 India 0.285 0.198 0.602 0.722 0.203 0.615 0.726 Maldives 0.104 0.201 0.924 0.850 0.215 0.928 0.851 Nepal 0.212 0.208 0.690 0.697 0.207 0.690 0.696 Pakistan 0.243 0.198 0.694 0.743 0.210 0.632 0.600 Sri Lanka 0.157 0.203 0.818 0.756 0.213 0.822 0.759 Total 0.277 0.198 0.615 0.725 0.203 0.627 0.727 Targeting on Improved Toilets Household-level Community-level % Undernour Coverage Incl. Errors Excl. Errors Coverage Incl. Errors Excl. Errors Bangladesh 0.266 0.283 0.636 0.622 0.203 0.629 0.716 India 0.285 0.419 0.625 0.439 0.209 0.603 0.708 Nepal 0.104 0.162 0.648 0.783 0.220 0.696 0.685 Pakistan 0.212 0.154 0.598 0.774 0.225 0.619 0.572 Total 0.282 0.410 0.626 0.451 0.209 0.605 0.691 Note: Data are drawn from the DHS. % Undernour refers to the proportion of the sample that is undernourished. Incl. Errors refers to inclusion errors, defined as the proportion of those who are covered who are not undernourished. Excl. Errors refers to exclusion errors, defined as the proportion of those who are undernourished who are not covered. Undernourishment is an indicator variable equal to one if a man or woman is underweight or a child is stunted or wasted. Sanitation is the sanitation index as described in Section 2. Improved toilets is an indicator variable equal to one if the household has an improved toilet and zero otherwise. At the household- level, we target households without an improved toilet. Community-level variables are calculated by taking the within-community household averages. For community-level improved toilets specifically, we take the proportion of households in the community with an improved toilet. Coverage across the different targeting levels of course varies: the number of individuals in the bottom 20% of households is unlikely to be the same as the number in the bottom 20% of communities. Nev- 23 ertheless, coverage across targeting levels and metrics generally hovers around our aim of 20%, with the exception of the household-level improved toilet indicator (which is simply the share of households without and improved toilet). Given the rates of undernourishment, with perfect targeting reaching 20% of the sample, we should only expect inclusion errors for the Maldives (which has an undernourishment rate of only 10%) and Sri Lanka (16%). For the remaining countries, our exclusion error rates will be greater than zero, as not all of the undernourished will be included given our cut-off point. Comparing the results between household- and community-level targeting, we find relatively few differ- ences in terms of inclusion and exclusion error rates. For wealth (our benchmark), targeting communities has slightly higher inclusion error rates but very comparable exclusion errors. Targeting based on the san- itation outcomes also results in relatively similar results between the household and community level: at the household level, sanitation results in somewhat lower inclusion errors comparatively but results in similar exclusion errors, on average. Using household-level wealth as the benchmark, we find that sanitation, both at the household and com- munity level, performs almost as well and in some cases better in terms of both inclusion and exclusion errors. The share of improved toilets in a community, an arguably straightforward targeting metric, has commensurate error rates with the other sanitation metrics, suggesting this may be an effective tool for practical purposes. Even more straightforward is the simple indicator as to whether the household has an improved toilet or not: for countries with coverage rates around 20% (i.e., all countries except India), inclusion and exclusion errors are not too dissimilar to the other household-level variables. Nevertheless, in absolute terms, both inclusion and exclusion error rates are very high for wealth- and sanitation-based targeting. Our results suggest that while sanitation may be a better targeting metric than household wealth, targeted interventions more generally are likely to find it difficult to capture many undernour- ished individuals. 7 Conclusion Wealth poverty is often assumed to underlie deficits in women’s and children’s nutritional status. And where there are undernourished individuals in households that are not wealth poor, the literature and policymakers typically ascribe such inequality to intra-household factors such as asymmetry in bargaining power. As a result, household-based wealth or consumption targeting is commonly suggested to address 24 individual deprivation. We show that half of all undernourished women and children in South Asia do not live in wealth-poor households. This finding, by itself, is not new: Brown et al. (2018) document this pattern across Sub- Saharan Africa. However, we go beyond the previous literature by showing that such inequality in nu- tritional outcomes covaries to a significantly greater degree with community characteristics than with household-level factors. Between-household and within-community inequality explains three to four times as much of the variation in nutritional outcomes across the wealth distribution as does within- household or between-community inequality. Our findings thus highlight the importance of within- community inequality as a driver of inequality in women’s and children’s nutritional status. We also show that community-level sanitation infrastructure is a key correlate of the probability of being undernourished. We conclude with a comparison of the effectiveness of targeting using household wealth versus community-level infrastructure. Our findings suggest that a simple categorical variable like the proportion of improved toilets in the village may provide an inexpensive and effective alternative to household wealth-based targeting. We show that for many measures of nutritional status, the share of improved toilets in the community performs at least as well as household wealth-based targeting in terms of capturing undernourished individuals. This is particularly true for the larger, more unequal South Asian countries of India, Pakistan, and Bangladesh. This paper thus provides evidence suggesting that targeting communities with poor sanitation infrastruc- ture is likely a viable alternative to household wealth-based targeting across South Asia. While a full cost-effectiveness analysis of targeting approaches is beyond the scope of this paper, data on a single, clearly defined proxy for sanitation infrastructure—the proportion of improved toilets—are likely less ex- pensive to collect than the full set of household characteristics required for consumption- or wealth-based targeting or indeed the data needs of a proxy means test (Brown et al., 2018, 2021). In turn, updating such a measure is also likely less expensive than updating a household-level PMT database. Finally, be- cause this measure relies on a single outcome, it is less likely to suffer from measurement error (Grosh and Baker, 1995). 25 References AGARWAL, N., A. AIYAR, A. BHATTACHARJEE, J. CUMMINS, C. GUNADI, D. SINGHANIA, M. TAYLOR, AND E. WIGTON-JONES (2017): “Month of Birth and Child Height in 40 Countries,” Economics Letters, 157, 10–13. AHMAD, K., T. KHANAM, S. A. KERAMAT, M. I. ISLAM, E. KABIR, AND R. KHANAM (2020): “Interaction between the place of residence and wealth on the risk of overweight and obesity in Bangladeshi women,” PloS one, 15, e0243349. AL KIBRIA, G. M., K. SWASEY, M. Z. HASAN, A. SHARMEEN, AND B. DAY (2019): “Prevalence and factors associated with underweight, overweight and obesity among women of reproductive age in India,” Global Health Research and Policy, 4, 1–12. BEHRMAN, J. (1988): “Nutrition, Health, Birth Order and Seasonality,” Journal of Development Economics, 28, 43–62. BEHRMAN, J. R. AND A. B. DEOLALIKAR (1987): “Will Developing Country Nutrition Improve with Income? A Case Study for Rural South India,” Journal of Political Economy, 95, 492–507. ——— (1988): “Health and Nutrition,” Handbook of Development Economics, 1, 631–711. BISHWAJIT, G. (2017): “Household wealth status and overweight and obesity among adult women in Bangladesh and Nepal,” Obesity Science & Practice, 3, 185–192. BISHWAJIT, G. AND S. YAYA (2020): “Overweight and obesity among under-five children in South Asia,” Child and Adolescent Obesity, 3, 105–121. BREDENKAMP , C., L. BUISMAN, AND E. VAN DE POEL (2014): “Persistent Inequalities in Child Undernutrition: Evidence from 80 Countries, from 1990 to Today,” International Journal of Epidemiology, 43, 1328– 1335. BROWN, C., R. CALVI, AND J. PENGLASE (2021): “Sharing the Pie: An Analysis of Undernutrition and Individual Consumption in Bangladesh,” Journal of Public Economics, 200, 104460. BROWN, C., M. RAVALLION, AND D. VAN DE WALLE (2018): “A Poor Means Test? Econometric Targeting in Africa,” Journal of Development Economics, 134, 109–124. ——— (2019): “Most of Africa’s Nutritionally-Deprived Women and Children are Not Found in Poor Households,” Review of Economics and Statistics, 104. BUCKLES, K. AND S. KOLKA (2014): “Prenatal investments, breastfeeding, and birth order,” Social Science & Medicine, 118, 66–70. CALVI, R. (2020): “Why are older women missing in India? The age profile of bargaining power and poverty,” Journal of Political Economy, 128, 2453–2501. CHUMLEA, W. C., S. S. GUO, R. J. KUCZMARSKI, K. M. FLEGAL, C. L. JOHNSON, S. B. HEYMSFIELD, H. C. LUKASKI, K. FRIEDL, AND V . S. HUBBARD (2002): “Body composition estimates from NHANES III bio- electrical impedance data,” International journal of obesity, 26, 1596–1609. 26 COADY, D., M. GROSH, AND J. HODDINOTT (2004): Targeting Transfers in Developing Countries: Review of Lessons and Experience, World Bank, Washington D.C. COFFEY, D., A. GUPTA, P. HATHI, D. SPEARS, AND N. SRIVASTAV (2015): “Culture and the health transition: Understanding sanitation behavior in rural north India,” International Growth Center, Working Paper, April. COWELL, F. (2011): Measuring inequality, Oxford University Press. COWELL, F. A. (2000): “Measurement of inequality,” Handbook of income distribution, 1, 87–166. DAS GUPTA, M. (1987): “Selective Discrimination Against Female Children in Rural Punjab, India,” Popu- lation and Development Review, 13, 77–100. DAS GUPTA, M. AND L. SHUZHUO (1999): “Gender Bias in China, South Korea and India 1920-1990: Effects of War, Famine and Fertility Decline,” Development and Change, 30, 619–652. DEATON, A. AND J. DRÈZE (2009): “Food and Nutrition in India: Facts and Interpretations,” Economic and Political Weekly, 42–65. DEL NINNO, C. AND B. MILLS (2015): Safety Nets in Africa: Effictive Mechanisms to Reach the Poor and Most Vulnerable, World Bank, Washington D.C. DERCON, S. AND P . KRISHNAN (2000): “In Sickness and in Health: Risk Sharing within Households in Rural Ethiopia,” Journal of Political Economy, 108, 688–727. DREVENSTEDT, G. L., E. M. CRIMMINS, S. VASUNILASHORN, AND C. E. FINCH (2008): “The rise and fall of excess male infant mortality,” Proceedings of the National Academy of Sciences, 105, 5016–5021. FILMER, D. AND L. PRITCHETT (2001): “Estimating wealth effects without expenditure data—or tears: an application to educational enrollments in states of India,” Demography, 38, 115–132. GARAWI, F., K. DEVRIES, N. THOROGOOD, AND R. UAUY (2014): “Global differences between women and men in the prevalence of obesity: is there an association with gender inequality?” European Journal of Clinical Nutrition, 68, 1101–1106. GARENNE, M. (2003): “Sex differences in health indicators among children in African DHS surveys,” Journal of biosocial science, 35, 601–614. GARG, A. AND J. MORDUCH (1998): “Sibling Rivalry and the Gender Gap: Evidence from Child Health Outcomes in Ghana,” Journal of Population Economics, 11, 471–493. GERUSO, M. AND D. SPEARS (2018): “Neighborhood sanitation and infant mortality,” American Economic Journal: Applied Economics, 2, 125–162. GROSH, M. AND J. L. BAKER (1995): “Proxy means tests for targeting social programs,” Living standards measurement study working paper, 118, 1–49. HABICHT, J.-P., C. YARBROUGH, R. MARTORELL, R. MALINA, AND R. KLEIN (1974): “Height and weight standards for preschool children: how relevant are ethnic differences in growth potential?” The Lancet, 303, 611–615. 27 HADDAD, L., H. ALDERMAN, S. APPLETON, L. SONG, AND Y. YOHANNES (2003): “Reducing Child Malnutri- tion: How Far Does Income Growth Take Us?” The World Bank Economic Review, 17, 107–131. HAMMER, J. S. AND D. SPEARS (2013): “Village sanitation and children’s human capital: evidence from a randomized experiment by the Maharashtra government,” World Bank Policy Research Working Paper. HEADEY, D. (2013): “Developmental Drivers of Nutritional Change: A Cross-Country Analysis,” World Development, 42, 76–88. HEADEY, D., J. HODDINOTT, D. ALI, R. TESFAYE, AND M. DEREJE (2015): “The Other Asian Enigma: Ex- plaining the Rapid Reduction of Undernutrition in Bangladesh,” World Development, 66, 749–761. HEADEY, D., J. HODDINOTT, AND S. PARK (2016): “Drivers of nutritional change in four South Asian coun- tries: a dynamic observational analysis,” Maternal & child nutrition, 12, 210–218. HEADEY, D. D. AND H. H. ALDERMAN (2019): “The relative caloric prices of healthy and unhealthy foods differ systematically across income levels and continents,” The Journal of Nutrition, 149, 2020–2033. HEADEY, D. D. AND J. HODDINOTT (2015): “Understanding the rapid reduction of undernutrition in Nepal, 2001–2011,” PloS one, 10, e0145738. HEIDARI, S., T. F. BABOR, P. DE CASTRO, S. TORT, AND M. CURNO (2016): “Sex and gender equity in research: rationale for the SAGER guidelines and recommended use,” Research Integrity and Peer Review, 1, 1–9. HELBLE, M. AND K. FRANCISCO (2017): “The imminent obesity crisis in Asia and the Pacific: First cost estimates,” . HIRVONEN, K., Y. BAI, D. HEADEY, AND W. A. MASTERS (2020): “Affordability of the EAT–Lancet reference diet: a global analysis,” The Lancet Global Health, 8, e59–e66. HONG, R., J. E. BANTA, AND J. A. BETANCOURT (2006): “Relationship Between Household Wealth Inequal- ity and Chronic Childhood Under-Nutrition in Bangladesh,” International Journal for Equity in Health, 5, 1–10. JAFAR, T., Z. QADRI, M. ISLAM, J. HATCHER, Z. A. BHUTTA, AND N. CHATURVEDI (2008): “Rise in childhood obesity with persistently high rates of undernutrition among urban school-aged Indo-Asian children,” Archives of disease in childhood, 93, 373–378. JAMES, W., A. FERRO -LUZZI, AND J. C. WATERLOW (1988): “Definition of chronic energy deficiency in adults. Report of a working party of the International Dietary Energy Consultative Group.” European Journal of Clinical Nutrition, 42, 969–981. JAMES, W. AND P. FRANCOIS (1994): “The choice of cut-off point for distinguishing normal body weights from underweight or’chronic energy deficiency in adults,” European Journal of Clinical Nutrition, 48, S179. JAYACHANDRAN, S. AND I. KUZIEMKO (2011): “Why Do Mothers Breastfeed Girls Less Than Boys? Evidence and Implications for Child Health in India,” The Quarterly Journal of Economics, 126, 1485–1538. 28 JAYACHANDRAN, S. AND R. PANDE (2012): “Introduction to Issue on Malnutrition,” CESifo Economic Stud- ies, 58, 253–255. ——— (2017): “Why Are Indian Children So Short? The Role of Birth Order and Son Preference,” Amer- ican Economic Review, 107, 2600–2629. JOHNSTON, E. (1991): “Weight changes during pregnancy and the postpartum period.” Progress in Food & Nutrition Science, 15, 117–157. KARKI, A., A. SHRESTHA, AND N. SUBEDI (2019): “Prevalence and associated factors of childhood over- weight/obesity among primary school children in urban Nepal,” BMC Public Health, 19, 1–12. KAUR, S., H. SACHDEV, AND U. KAPIL (2008): “Prevalence of overweight and obesity amongst school children in Delhi, India,” Asia Pacific Journal of Clinical Nutrition, 17, 592. KRAAY, A. AND D. MCKENZIE (2014): “Do poverty traps exist? Assessing the evidence,” Journal of Economic Perspectives, 28, 127–48. KURPAD, A., S. MUTHAYYA, AND M. VAZ (2005): “Consequences of inadequate food energy and negative energy balance in humans,” Public Health Nutrition, 8, 1053–1076. LAMBERT, S., M. RAVALLION, AND D. VAN DE WALLE (2014): “Intergenerational Mobility and Interpersonal Inequality in an African Economy,” Journal of Development Economics, 11, 327–344. LARSEN, A. F., D. HEADEY, AND W. A. MASTERS (2019): “Misreporting Month of Birth: Implications for Nutrition Research,” Demography, 56, 707–728. LEHMANN, J.-Y. K., A. NUEVO -CHIQUERO, AND M. VIDAL-FERNANDEZ (2018): “The early origins of birth order differences in children’s outcomes and parental behavior,” Journal of Human Resources, 53, 123– 156. MCMILLEN, M. M. (1979): “Differential mortality by sex in fetal and neonatal deaths,” Science, 204, 89– 91. NAEYE, R. L., L. S. BURT, D. L. WRIGHT, W. A. BLANC, AND D. TATTER (1971): “Neonatal mortality, the male disadvantage,” Pediatrics, 48, 902–906. NUBÉ, M. (2009): “The Asian enigma: predisposition for low adult BMI among people of South Asian descent,” Public health nutrition, 12, 507–516. OLSON, C., M. STRAWDERMAN, P . HINTON, AND T. PEARSON (2003): “Gestational weight gain and postpar- tum behaviors associated with weight change from early pregnancy to 1 y postpartum,” International Journal of Obesity, 27, 117–127. OSTER, E. (2009): “Does increased access increase equality? Gender and child health investments in India,” Journal of Development Economics, 89, 62–76. PANDE, R. (2003): “Can Mandated Political Representation Increase Policy Influence for Disadvantaged Minorities? Theory and Evidence from India,” American Economic Review, 93, 1132–1151. 29 PANDE, R. AND A. YAZBECK (2003): “What’s in a Country Average? Wealth, Gender, and Regional Inequal- ities in Immunization in India,” Social Science & Medicine, 57, 2075–2088. PRICE, J. (2008): “Parent-Child Quality TIme: Does Birth Order Matter?” Journal of Human Resources, 43, 240–265. PRITCHETT, L. AND L. H. SUMMERS (1996): “Wealthier is Healthier,” Journal of Human Resources, 31. RAMACHANDRAN, A. AND C. SNEHALATHA (2010): “Rising burden of obesity in Asia,” Journal of Obesity, 2010. RAMALINGASWAMI, V., U. JONSSON, AND J. RHODE (1997): Malnutition: A South Asia Enigma, UNICEF, New York. RAVALLION, M. (1990): “Income Effects on Undernutrition,” Economic Development and Cultural Change, 38, 490–515. ——— (1992): “Does Undernutrition Respond to Incomes and Prices: Dominance Tests for Indonesiah,” World Bank Economic Review, 6, 109–124. ——— (2016): The Economics of Poverty: History, Measurement, and Policy, New York: Oxford University Press. REINHARD, I. AND K. B. S. WIJERATNE (2000): The use of stunting and wasting as indicators for food inse- curity and poverty, Poverty Impact Monitoring Unit. SMITH, L. AND L. HADDAD (2015): “Reducing Child Undernutrition: Past Drivers and Priorities for the Post-MDG Era,” World Development, 68, 180–204. SPEARS, D. (2020): “Exposure to Open Defecation Can Account for the Indian Enigma of Child Height,” Journal of Development Economics, 146, 102–277. THURSTANS, S., C. OPONDO, A. SEAL, J. WELLS, T. KHARA, C. DOLAN, A. BRIEND, M. MYATT, M. GARENNE, R. SEAR, ET AL. (2020): “Boys are more likely to be undernourished than girls: a systematic review and meta-analysis of sex differences in undernutrition,” BMJ global health, 5, e004030. TRIVERS, R. L. AND D. E. WILLARD (1973): “Natural selection of parental ability to vary the sex ratio of offspring,” Science, 179, 90–92. ULIJASZEK, S. AND D. KERR (1999): “Anthropometric Measurement Error and the Assessment of Nutri- tional Status,” British Journal of Nutrition, 82, 165–177. UNICEF/WHO/WORLD BANK (2020): “Levels and Trends in Child Malnutrition: Key Findings of the 2020 Edition of the Joint Child Malnutrition Estimates,” . WAGSTAFF, A. (2000): “Socioeconomic Inequalities in Child Mortality: Comparisons Across Nine Devel- oping Countries,” Bulletin of the World Health Organization, 78, 19–29. WAGSTAFF, A., C. BREDENKAMP , AND L. BUISMAN (2014): “Progress on Global Health Goals: Are the Poor Being Left Behind,” World Bank Research Observer, 29, 137–162. 30 WAMANI, HENRY AND ÅSTRØM, A. N., S. PETERSON, J. TUMWINE, AND T. TYLLESKÄR (2007): “Boys Are More Stunted Than Girls in Sub-Saharan Africa: A Meta-Analysis of 16 Demographic and Health Sur- veys,” BMC Pediatrics, 7, 1–10. WATERLOW, J. (1972): “Classification and Definition of Protein-calorie Malnutrition.” British Medical Jour- nal, 3, 566. WATERLOW, J. C., R. BUZINA, W. KELLER, J. LANE, M. NICHAMAN, AND J. TANNER (1977): “The presentation and use of height and weight data for comparing the nutritional status of groups of children under the age of 10 years,” Bulletin of the world Health Organization, 55, 489. WELLS, J. (2000): “Natural Selection and Sex Di4erences in Morbidity and Mortality in Early Life,” Journal of theoretical Biology, 202, 65–76. WHO (1986): “Use and interpretation of anthropometric indicators of nutritional status,” Bulletin of the World Health Organization, 64, 929. WILLIS, J., V. KUMAR, S. MOHANTY, P . SINGH, V. SINGH, A. BAQUI, S. AWASTHI, J. SINGH, M. SANTOSHAM, AND G. DARMSTADT (2009): “Gender Differences in Perception and Care-seeking for Illness of Newborns in Rural Uttar Pradesh, India,” Journal of Health, Population and Nutrition, 27, 62–71. WORLD BANK (2008): “Poverty Assessment for Bangladesh: Creating Opportunities for Bridging the East- West Divide,” Development Series Paper No. 26. Poverty Reduction, Economic Management, Finance & Private Sector Development Sector Unit South Asia Region. YOON, K.-H., J.-H. LEE, J.-W. KIM, J. H. CHO, Y.-H. CHOI, S.-H. KO, P . ZIMMET, AND H.-Y. SON (2006): “Epidemic obesity and type 2 diabetes in Asia,” The Lancet, 368, 1681–1688. 31 A Appendix: Additional Tables and Figures (a) Bangladesh (b) India (c) Maldives (d) Nepal (e) Pakistan (f) Sri Lanka Note: DHS data. The graphs show concentration curves for the cumulative proportion of children who report suffering from various illnesses. The Stata command glcurve is used to construct the curves. Figure A1: Concentration Curves for Child Anemia, Fever, and Diarrhea 32 (a) Bangladesh (b) India (c) Maldives (d) Nepal (e) Pakistan (f) Sri Lanka Note: DHS data. The graphs show lowess curves for the proportion of women and men who are obese in each household wealth percentile. Observations with missing values and pregnant or lactating women have been dropped. Figure A2: Incidence of Obesity by Household Wealth 33 (a) Household (b) Community Note: DHS data. Both figures show histograms with the distribution of undernourishment at the household and community level, respectively. Un- dernourishment is an indicator variable equal to one if an adult is underweight or a child is stunted or wasted and zero otherwise. Observations with missing values and pregnant women have been dropped. Figure A3: Average Nutritional Status by Wealth Percentile 34 Community services index (a) Underweight women (b) Stunted children (c) Wasted children Community sanitation index (d) Underweight women (e) Stunted children (f) Wasted children Proportion improved toilets (g) Underweight women (h) Stunted children (i) Wasted children Note: DHS data Figure A4: Targeting on Household Wealth versus Community-Level Indicators 35 Table A1: Sample Sizes and Survey Years Survey Year Women Children 0-5 Men India 2016 655,156 225,002 126,543 Maldives 2017 6,835 2,350 6,759 Nepal 2016 6,165 2,365 6,589 Pakistan 2018 4,681 4,099 Sri Lanka 2016 17,054 7,655 Bangladesh 2014 16,624 6,965 Total 706,515 248,436 139,891 Note: Data are drawn from the DHS. Pregnant women and individuals with- out height or weight data have been dropped. Table A2: Proportion of Pregnant and Lactating Women by Country Percent women pregnant Percent women breastfeeding Bangladesh 5.64 20.95 India 5.54 16.51 Maldives 3.63 16.26 Nepal 5.01 23.03 Pakistan 12.43 27.19 Sri Lanka 4.53 25.48 Total 5.56 16.97 Note: Data are drawn from the DHS. 36 Table A3: Proportion of Severely Undernourished Individuals in Poorest 20% and 40% of Household Wealth Poorest 20% of Households Underweight Stunting Wasting Men Women Boys Girls Overall Boys Girls Overall Bangladesh . 0.34 0.39 0.34 0.37 0.27 0.29 0.28 India 0.25 0.29 0.40 0.42 0.41 0.30 0.31 0.30 Maldives 0.26 0.19 0.38 0.25 0.32 0.13 0.32 0.21 Nepal 0.19 0.20 0.37 0.35 0.36 0.31 0.28 0.30 Pakistan . 0.45 0.39 0.48 0.43 0.25 0.24 0.25 Sri Lanka . 0.39 0.30 0.36 0.33 0.23 0.35 0.28 Total 0.23 0.31 0.37 0.37 0.37 0.25 0.30 0.27 Poorest 40% of Households Underweight Stunting Wasting Men Women Boys Girl Overall Boys Girls Overall Bangladesh . 0.60 0.60 0.56 0.58 0.54 0.41 0.49 India 0.50 0.54 0.63 0.67 0.65 0.52 0.54 0.53 Maldives 0.47 0.36 0.65 0.45 0.56 0.49 0.39 0.44 Nepal 0.47 0.45 0.55 0.58 0.56 0.48 0.49 0.49 Pakistan . 0.69 0.62 0.68 0.65 0.64 0.62 0.63 Sri Lanka . 0.63 0.54 0.62 0.57 0.52 0.57 0.54 Total 0.48 0.54 0.60 0.59 0.60 0.53 0.50 0.52 Note: All data are drawn from the DHS. The table gives the conditional probabilities of being severely undernourished given that the individual lives in a household in the poorest 20% and 40% of the household wealth distribution. Severe undernourishment is defined as 3 standard deviations below median for height-for-age or weight-for-height scores for children or with a BMI less than 17 for adults. Population weighting and sampling probability is used. 37 Table A4: Proportion of Undernourished Individuals in Poorest 20% and 40% of Household Wealth Ex- cluding Women Less than 18 Years and Children Less than 18 Months of Age Poorest 20% of Households Underweight Stunting Wasting Women Boys Girls Overall Boys Girls Overall Bangladesh 0.33 0.34 0.30 0.32 0.23 0.27 0.25 India 0.30 0.35 0.36 0.35 0.29 0.30 0.30 Maldives 0.17 0.24 0.25 0.24 0.21 0.13 0.20 Nepal 0.19 0.26 0.30 0.28 0.19 0.31 0.21 Pakistan 0.44 0.30 0.38 0.34 0.24 0.25 0.26 Sri Lanka 0.35 0.33 0.33 0.33 0.24 0.23 0.26 Total 0.30 0.30 0.32 0.31 0.24 0.25 0.25 Poorest 40% of Households Underweight Stunting Wasting Women Boys Girl Overall Boys Girls Overall Bangladesh 0.59 0.56 0.54 0.55 0.50 0.45 0.48 India 0.55 0.59 0.61 0.60 0.52 0.54 0.53 Maldives 0.34 0.45 0.49 0.47 0.54 0.37 0.47 Nepal 0.45 0.50 0.55 0.52 0.52 0.43 0.48 Pakistan 0.64 0.54 0.60 0.57 0.52 0.56 0.54 Sri Lanka 0.60 0.55 0.58 0.57 0.49 0.52 0.51 Total 0.53 0.53 0.56 0.55 0.52 0.48 0.50 Note: Data are drawn from the DHS. The table gives the conditional probabilities of being undernourished given that the individual lives in a household in the poorest 20% and 40% of the household wealth distribution, excluding children younger than 18 months of age and women younger than 18 years of age. Population weighting and sampling probability is used. Table A5: Mean Conditional Probabilities Using Predicted Wealth from the Augmented Regressions Model 1: Household Model 2: Household and Individual Bottom 20% Bottom 40% Bottom 20% Bottom 40% Underweight women 0.214 0.425 0.209 0.412 Stunted children 0.201 0.409 0.200 0.400 Wasted children 0.204 0.404 0.197 0.387 Note: The table gives proportions of the underweight who fall into the poorest 20% and 40% of the distribu- tion of predicted values from regressions of nutrition indicators on wealth, controlling for household covariates (model 1) and both household and individual covariates (model 2). Means are population weighted. House- hold covariates: electricity, radio, TV, fridge, number of rooms, telephone, bank account, age of head, education of health, female-headed household, and number of household members. Individual covariates: works, marital status, education, age (woman), girl child, and age (child). 38 Table A6: Correlation Matrix: Mother and Child Nutrition Stunted Wasted Diarrhea Anemia Fever Mother underweight 0.0949*** 0.0897*** 0.0113*** 0.0574*** 0.0013 Mother anemia 0.0407*** 0.0197*** 0.0056*** 0.1356*** -0.0075*** Note: Data are drawn from the DHS. 39 Table A7: Mean Log Deviation (MLD) between and within Variation Women’s BMI Households Communities Overall Between (%) Within (%) Between (%) Within (%) Bangladesh 0.017 89 11 18 82 India 0.017 78 22 19 81 Maldives 0.024 65 35 7 93 Nepal 0.015 78 22 22 78 Pakistan 0.023 88 12 29 71 Sri Lanka 0.019 95 5 21 79 Total 0.017 78 22 19 81 Men’s BMI Households Communities Overall Between (%) Within (%) Between (%) Within (%) India 0.014 79 21 23 77 Maldives 0.018 77 23 10 90 Nepal 0.013 76 24 18 82 Total 0.014 79 21 22 78 Height-for-age z-scores Households Communities Overall Between (%) Within (%) Between (%) Within (%) Bangladesh 0.013 88 12 15 85 India 0.020 80 20 22 78 Maldives 0.010 80 20 15 85 Nepal 0.013 83 17 25 75 Pakistan 0.021 72 29 26 74 Sri Lanka 0.009 89 12 33 67 Total 0.019 80 20 22 78 Weight-for-height z-scores Households Communities Overall Between (%) Within (%) Between (%) Within (%) Bangladesh 0.008 89 11 12 88 India 0.012 81 19 24 76 Maldives 0.010 80 20 13 87 Nepal 0.007 82 18 24 76 Pakistan 0.009 72 28 27 73 Sri Lanka 0.009 89 11 34 66 Total 0.012 81 19 24 76 Note: Data are drawn from the DHS. Pregnant women and individuals without height or weight data have been dropped. 40 Table A8: Kitagawa-Blinder-Oaxaca decomposition of the variation in predicted undernourishment Predicted undernourishment rates Targeting metric: improved toilets Targeting metric: wealth Threshold: bottom 20% Threshold: bottom 40% Threshold: bottom 20% Threshold: bottom 40% Overall Above threshold 0.28*** 0.28*** 0.27*** 0.27*** 0.28*** 0.28*** 0.27*** 0.27*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Above threshold 0.34*** 0.34*** 0.33*** 0.33*** 0.35*** 0.35*** 0.33*** 0.33*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Difference -0.05*** -0.05*** -0.05*** -0.05*** -0.07*** -0.07*** -0.07*** -0.07*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Endowments Household wealth -0.04*** -0.04*** -0.06*** -0.07*** (0.000) (0.000) (0.000) (0.000) Proportion of improved toilets -0.01* -0.01*** -0.01*** -0.01*** (0.079) (0.000) (0.000) (0.000) Coefficients Household wealth -0.00*** -0.00 0.00** 0.00*** (0.002) (0.197) (0.025) (0.000) Proportion of improved toilets -0.00 -0.00*** 0.00*** 0.00*** (0.110) (0.000) (0.000) (0.000) Interaction Household wealth 0.00*** 0.00 -0.00** -0.01*** (0.002) (0.197) (0.025) (0.000) Proportion of improved toilets -0.01 -0.01*** 0.01*** 0.00*** (0.110) (0.000) (0.000) (0.000) Observations 946,128 947,713 946,128 947,713 946,128 947,713 946,128 947,713 Note: Data are drawn from the DHS. Standard errors are clustered at the household-level in the decomposition. t-statistics are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01 41 Table A9: Conditional Probabilities: Actual and Simulated without Intra-Household Inequality Women Boys Girls Actual Simulated Actual Simulated Actual Simulated Poorest 20% of households Bangladesh 0.33 0.33 0.18 0.33 0.24 0.23 India 0.31 0.36 0.39 0.48 0.40 0.48 Maldives 0.18 0.12 0.25 0.17 0.20 0.20 Nepal 0.19 0.23 0.28 0.25 0.30 0.24 Pakistan 0.61 0.80 0.40 0.91 0.43 0.81 Sri Lanka 0.39 0.38 0.35 0.34 0.36 0.37 Total 0.34 0.37 0.31 0.41 0.32 0.39 Poorest 40% of households Bangladesh 0.54 0.56 0.34 0.50 0.41 0.57 India 0.56 0.62 0.64 0.72 0.66 0.74 Maldives 0.34 0.24 0.44 0.32 0.45 0.41 Nepal 0.43 0.44 0.49 0.45 0.56 0.56 Pakistan 0.71 0.80 0.54 0.91 0.57 0.81 Sri Lanka 0.65 0.62 0.49 0.46 0.58 0.58 Total 0.54 0.55 0.49 0.56 0.54 0.61 Note: Using DHS data from countries where male head BMI is available, the table displays the proportions of underweight women and children by wealth distribution. Actual: conditional probabilities using actual nutrition values. Simulated equality: conditional probabilities when each woman and child is given the malnutrition status of their household head. Statistics are population weighted. 42 Table A10: Inequality in Nutritional Outcomes (1) (2) (3) (4) (5) (6) BMI Underweight Height-for-age Stunted Weight-for-height Wasted Female 0.27*** 0.01*** 0.09*** -0.02*** 0.05*** -0.02*** (20.23) (6.01) (12.86) (-9.32) (8.32) (-12.30) Age 0.13*** -0.01*** -0.01*** 0.00*** 0.00 -0.00*** (269.44) (-167.46) (-17.14) (9.55) (1.48) (-10.30) Wealth index 0.95*** -0.06*** 0.27*** -0.08*** 0.12*** -0.02*** (121.61) (-73.02) (42.78) (-41.59) (23.23) (-14.71) Female head -0.01 0.00** 0.03** -0.01*** 0.03*** -0.00 (-0.59) (2.34) (2.55) (-3.41) (2.86) (-1.13) Age of head -0.01*** 0.00*** 0.00*** -0.00*** 0.00*** 0.00 (-15.97) (11.76) (16.26) (-14.11) (5.28) (0.30) Education head 0.02*** -0.00*** 0.02*** -0.00*** 0.00*** -0.00*** (16.49) (-6.98) (18.14) (-18.05) (5.87) (-3.39) Education 0.03*** -0.00*** (24.30) (-18.57) First born -0.01 -0.01*** 0.04*** -0.02*** (-1.45) (-4.70) (6.65) (-7.94) Son preference -0.04*** 0.01*** -0.01* 0.00 (-4.68) (4.50) (-1.69) (0.26) Constant 17.60*** 0.47*** -1.42*** 0.41*** -1.09*** 0.27*** (533.14) (132.48) (-48.32) (46.57) (-44.70) (37.09) N 815046 815046 246586 246586 246586 246586 R2 0.289 0.159 0.250 0.220 0.254 0.202 Community FE Yes Yes Yes Yes Yes Yes Note: Data are drawn from the DHS. Age is measured in years for adults and months for children. Education is measured in years. Household fixed effects are included in all regressions. t-statistics are in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01 43