Endowments or Returns to Endowments or Both? Deciphering Disparities in Childhood Stunting in Bihar, India

Background Unacceptably high rate of childhood stunting for decades remained a puzzle in the eastern Indian state of Bihar. Despite various programmatic interventions, nearly half of the under-ve children (numerically about 10 million) are still stunted in this resource-constrained state. Data and Methods Using four successive rounds of National Family Health Survey (NFHS) data spread over more than two decades and by employing quantile regressions and counterfactual decomposition (QR-CD), the present study aims to assess effects of various endowments as well as returns to those endowments in disparities in childhood stunting over the period. Results The results show that although child’s height-for-age Z-scores (HAZ) disparity was largely accounted for differing levels of endowments during earlier decade, in the later periods, inadequate access to the benets from various development programmes was also found responsible for HAZ disparities. Moreover, effects of endowments and their returns vary across quantiles. We argue that apart from equalizing endowments, ensuring adequate access to different nutrition-centric programmes are essential to lessen the burden of childhood stunting. Conclusion state nutrition and, policy to harness better


Background
Child undernutrition in India has remained a priority among academician and policy makers. Despite signi cant economic growth during past two decades, prevalence of childhood stunting has dropped only by 27 percent (about 14 points) [1]. UNICEF (2013) has observed that India alone contributed 38 percent of the stunted children in the world in 2011 [2], while Headey [2013] estimated that number of undernourished children in India was higher than in all of Africa [3]. Jose et al. emphasized that despite a moderate decline in child undernutrition during past decade, a large and graded socio-economic disparity in child undernutrition continues to exist [4]. A systematic review on prevalence of child undernutrition in India has also concluded that burden of child undernutrition is still unacceptably high in India and there is an urgent need to understand the risk factors in greater details [5]. It is needless to mention that a rapid reduction of child undernutrition in India is imperative to lessen global burden of child malnutrition.
Recent studies found that women's BMI, education, child's adequate diet, household asset, and sanitation, age at marriage, antenatal care and household size are strong and signi cant predictors of childhood anthropometric outcomes and explains much of the variations across the districts in India [18][19]. Jose et al. have noted that about 83 percent of high stunting prevalence (higher than the national average) districts belong to the eight states located in the north-central, western, and eastern region [4]. Thus, effects of endowments (or covariate per se) were found to be signi cant; however, they vary across space and nature of endowment.
At the same time, some studies have also attempted to document disparities in returns to endowments (or strength of association per se) and their different dimensions, which potentially in uence child nutritional outcomes. For example, how quality of the governance, institutional strength in implementing public policies, reach of public services, bargaining power of the communities, and macro-level political economic context etc. could in uence health and nutritional outcomes have been documented in these studies. In Indian context, disparity in institutional performance (measured in terms of quality of public services such as health, education and public distribution system) was observed between northern and north-central states, and southern states [20][21][22]. Harriss and Kohli investigated in uence of inter-state political and institutional factors on child undernutrition and differentiated between the politics of "clientelism" and "programmatic" politics [23]. They argued that such political spectrum could impinge on worse and better child anthropometric outcomes respectively. Signi cant gaps in implementation regarding nature, coverage and quality of Integrated Child Development Services (ICDS) were found by various researchers in different states [24][25][26][27].
Majority of literature reviewed above have either tried to identify some of the key observable characteristics (or covariates or endowments) that help in explaining variation in child anthropometric outcomes or have emphasized differential strength of relationship (or coe cients or returns to endowments) might also in uence childhood nutritional outcomes. Only a few studies have attempted to quantify the contribution of socio-demographic, economic and ecological variables, individually or at the aggregate (covariate effects) and contribution of the strength of relationship (or coe cient effects) together in the South Asian context [28,39]. In the present context, covariate effects can be de ned as the differences in nutrition outcomes across periods explained by the differences in observed covariates. On the other hand, differences explained by differing strengths of relationships between covariates and outcomes, in other words the "returns" to specific endowments, can be termed coefficient effects. To nd out differentials in child undernutrition in Nepal and Bangladesh, Srinivasan et al. have highlighted that rural-urban disparities in child nutrition are primarily attributable to the difference in the levels of critical endowments such as household a uence, maternal as well as spouse's education, while differences in the strength of association (or returns to endowments) between determinants and nutrition outcomes are of relatively small in magnitude [28].
However, studies conducted in India found that large disparities in child nutritional outcomes across states are modestly explained by the differences in critical endowments, while returns to endowments or implementation of nutrition-relevant policies and programmes play an important role in explaining such disparity [29].
During past two decades, India and its states have witnessed substantial changes in endowments (covariates) and also experienced enormous policy changes (coe cients) which could have direct or indirect bearing on child nutritional outcomes. Apart from expanding scope and coverage of ICDS, many states have also come up with many state-speci c schemes and emphasized multisectoral nutrition intervention. For example, Maharashtra, Madhya Pradesh and Karnataka have implemented State Nutrition Missions and placed special emphasis on nutrition surveillance, district planning, and district-level monitoring with the goal of reducing undernutrition at a desirable extent.
The present study intends to nd out changing relative contribution of different covariates and coe cients resulting disparities in childhood stunting in different intervals between 1992-93 and 2015-16 in the state of Bihar. The state of Bihar, located in the eastern part of India and a resource-constrained state, having the highest prevalence of childhood stunting in India for several past decades. The proportion of childhood stunting has declined by 21 percent (or by 13 percentage-points) during last twenty years -implying an annual average decline of just one percent [1]. Numerically, about 10 million children in Bihar are stunted. Notably, Bihar alone contributes around 15 percent of stunted children in India. More importantly, out of 100 districts, where prevalence of stunting is the highest, one-quarter belonging to Bihar. It was estimated that malnutrition (maternal and child malnutrition together) continued to be the largest risk factor driving the most death and disability since 1990s [30].
Changes in the basic socio-demographic and economic indicators during last two decades are given in Table 1. To note, the state of Bihar has undergone territorial changes following Bihar Reorganization Act (2000) (Government of India, 2000) and a separate state of Jharkhand was created from the districts of south Bihar.
--< Table 1: Some important demographic and health indicators of Bihar in 1992-93 and 2015-16>-- The study contributes to the literature of childhood stunting by applying a recently developed advanced statistical technique, namely, quantile regression-based counterfactual decomposition (QR-CD) method, which allows a more nuanced approach to disentangle the effects of endowments (or covariates) and returns to endowments (or coe cients). The study would also like to enquire whether changing contribution of covariate and coe cient effects are different at the lower tail of the distribution of height-for-age zscores (HAZ), where severe stunting is likely to be prevalent, compared to the middle and higher tail. Such insights would be of utmost value in a policy atmosphere where targeting most vulnerable is considered imperative. The primary hypothesis is that the period-wise changes across the HAZ distribution arises from covariate, rather than coe cient effects. Disparities at the lower tail of the distribution is of particular interest of the present study. A secondary hypothesis is that, even if a covariate or a coe cient dominates, there are important differences across the HAZ distribution in the relative contributions of covariate and coe cient effects to period-wise changes.

Methods
[Please see the supplementary les section to view the methods.] Without loss of generality, one can note that absolute increase in overall HAZ scores was the highest between second and third rounds of NFHS (i.e. between 1998-99 and 2005-06) followed by third and fourth rounds i.e. between 2005-06 and 2015-16. Child's HAZ scores largely remained at the same level between 1992-93 and 1998-99. Absolute increase of child's HAZ scores was remarkable for the bottom quantiles between 1998-99 and 2005-06 nationally and in Bihar, in particular. In Bihar, there was even decline of HAZ scores at the top quantile. However, between 2005-06 and 2015-16, absolute increase in the HAZ scores was observed at the top quantile nationally as well as in Bihar. In other words, nutritionally better-off children gained more compared to the severely stunted during last decade.

Descriptive statistics
--< Table 2: Percentile of HAZ score in NFHS 1, NFHS 2, NFHS 3 and NFHS 4 in Bihar and India>-- Table 3 depicts socio-demographic and economic characteristics of the samples in four rounds of NFHS. It has been observed that initiation of early breastfeeding (within one hour of birth) has improved dramatically -more than 14-times -between 2005-06 and 2015-16. Although number of siblings of the index child has been declined in the recent past, it still indicates fertility in the state is high. Notably, bene t received from ICDS services increased by more than 7-fold between 2005-06 and 2015-16. Similar is the case for institutional delivery of mothers. Mother's age at rst child has increased by nearly two years during the study period. Majority of the respondents in the sample was Hindu and non-SC/ST, including OBCs. It is surprising to nd out that proportion of economically marginalized households in the sample has increased from 1998-99, in spite of the state's higher economic growth during these periods, particularly after 2005 [36]. Being the least urbanised state of the country (among the major states), overwhelming proportion of the sample belong to the rural areas of Bihar. --<

Unconditional RIF quantile regression results
The estimates derived from unconditional RIF quantile regressions (QR) separately for all the survey periods were shown in Tables 4 and 5. It has been observed that child age has negative and signi cant in uence with child's HAZ scores across quantiles. If one moves from the lower tail to the upper tail, this effect increases, indicating that children starting with better nutritional status stand to lose more through faltering as they grow older. Although such observation holds for second and third rounds of survey, said observation con rms up to 75 percent quantile for rst and fourth rounds. Girls were found to have signi cantly better HAZ outcomes compared to boys across quantiles; however, strength of association varies across quantile and period of survey. Child's size at birth (proxy for birth weight) was found to have varying association with HAZ scores across quintiles during rst two rounds, in third and fourth rounds, size of the birth of children did not have any signi cant effect on HAZ scores. Early initiation of breastfeeding found to have positive and signi cant effect on HAZ scores in rst round, while such effect weakened during the last three rounds. Higher sibling size has negative signi cant in uence on child's HAZ scores, particularly those belong to the lower quantiles in third and fourth rounds of survey. Receipt of any bene t from ICDS found to be negatively associated with child's HAZ scores and such effect increases when we go from lower tail to higher tail of the HAZ distribution in the last round of survey.
Institutional delivery of mother, which is an important indicator for contact with health personnel, has positive and signi cant in uence on child's HAZ scores across quantiles, particularly at the lower and middle quantile in varying degree except during the third round of the survey. Signi cant positive effect of higher age of mother's rst birth on child's HAZ outcomes was found in the higher quantiles during the rst and the latest rounds of survey, but not in other rounds. Notably, signi cant positive in uence of maternal education on child's HAZ scores decreased with rounds.
---< Table 4: Unconditional Re-centred In uence Function (RIF) Quantile Regression Results for NFHS 1 (1992-93) and NFHS 2 (1998-99) in Bihar >------< Table 5 Differentials with respect to religion and caste a liation were found in child's HAZ scores during rst round of the survey; however, the relationship weakened thereafter. Signi cant positive in uence of household a uence on child's HAZ outcomes was found during rst and third round of survey and observation suggests the effect is higher among those belonging to higher quantiles. The results also revealed that rural-urban differentials in child's HAZ outcomes diminished over the period in Bihar.

Quartile regression Oaxaca Blinder counterfactual decomposition (QR-CD)
The estimated QR-CD results at the aggregate level of child, maternal, household and spatial characteristics were presented in the Tables 6-8, while a detailed breakdown of contribution of these characteristics were given in the Appendix Tables A1-A3. Before interpreting the results, it should be kept in mind that the negative sign of the observed raw gap in HAZ scores between two successive periods re ects the fact that raw HAZ scores of the later period was lower than the previous period in all quantiles, except at the highest quantile between second and third rounds. Additionally, it must also be kept in mind that the direction of effect of contribution of characteristics as shown in the Tables 6-8 -negative gures imply a contribution to increase in the disparity in HAZ scores over time, while positive gures show a contribution to diminish it. A careful look to these tables reveals certain pattern of covariate effects and coe cient effects across quantiles and over the periods.
---< Table 6: Oaxaca Blinder Decomposition of HAZ Scores of NFHS 1 and NFHS 2 in Bihar >------< Table 7: Oaxaca Blinder Decomposition of HAZ Scores of NFHS 2 and NFHS 3 in Bihar >------< Table 8: Oaxaca Blinder Decomposition of HAZ Scores of NFHS 3 and NFHS 4 in Bihar >---It may be observed that between the periods 1992-93 and 1998-99 covariate (or endowments) effects contributed signi cantly to enhance disparities in child HAZ outcomes, at the 10 th , 50 th and 75 th quantiles, while coe cient (returns to endowments) effects dominates over covariate effect in enhancing disparities in child's HAZ outcomes at 90 th quantile (see Table 6). Lower panel of the Table 6 suggests that child endowments alone contributed 36.5 percent at 90 th quantile to 270.8 percent at 25 th quantile in explaining disparities in child's HAZ outcomes. Effect of mother's characteristics (or mother's endowments) in explaining such disparities was found to be relatively small and varies between -8.  Table 7), while such effects vary between 117 -168.7 percent between the same quantile (see Table 8). Additionally, between the said periods, coe cient effects enhanced disparity in child's HAZ outcomes even at the 90 th quantile. The lower panels of the Tables 7 and 8 revealed that between 1998-99 and 2005-06, coe cient effects of child characteristics signi cantly increased disparities across quantiles, while said effects of mother's characteristics have tried to reduce it except at 25 th and 50 th quantiles. Further, coe cient effects of the household attributes have tried to increase disparities in HAZ outcomes signi cantly at 25 th and 75 th quantiles between 1998-99 and 2005-06 and 10 th to 50 th quantiles between 2005-06 and 2015-16. Additionally, during the last period, positive and signi cant covariate effects were observed at the higher tails of HAZ distribution.
If covariate effects and coe cient effects of different attributes are looked in more disaggregated manner during the study period (as given in the Appendix A1-A3), it has been found that these effects vary across quantiles, periods and nature of endowment. For example, delivery in institutions was found to have signi cant effect in enhancing disparities, particularly between lower tails of the HAZ distribution between 1992-93 and 1998-99 (Appendix Table A1), coe cient effects of mother's height and BMI, and, media exposure have tried to reduce disparities across quantiles between 1998-99 and 2005-06 (Appendix Table   A2). During the same period, covariate effect of institutional delivery has contributed signi cantly in increasing disparities. Between 2005-06 and 2015-16, both covariate and coe cient effects of the receipt of ICDS services were found to be signi cantly associated with reduction of HAZ disparities among children (Appendix Table A3).

Discussion
The QR-CD method provides speci c insight into the drivers of disparities across child's HAZ distribution.
The understanding of factors resulted in disparities in the lower quantiles of HAZ scores would be useful in designing interventions aimed at the vulnerable households with children of the highest levels of stunting.
In order to assess the contribution of the 'returns' to various interventions in reducing child HAZ disparities during last two decades, such quanti cation of the contribution of different socio-demographic, economic and cultural determinants seemed to be imperative for the state of Bihar.
This study indicates that although between 1992-93 and 1998-99 child's HAZ disparity at the bottom quantile of the distribution was largely accounted for differing levels of endowments, in the later periods such differences weakened statistically. In other words, between 1992-93 and 1998-99, at the lowest quantile, reducing disparity in childhood stunting was a matter of equalizing endowments; however, between 1998-99 and 2015-16, both -unequal endowments as well as dissimilar access to the bene ts of implementation of government sponsored schemes -were largely responsible for childhood HAZ disparity.
At the higher quantiles, particularly between 50 -75 th quantile, although unequal endowments were responsible for such disparities between 1992-93 and 1998-99, inadequate access to bene ts from outcomes for the bottom quintiles, though at the aggregate level in uences of endowments were statistically weak. According to the current estimates, much of the reduction of disparities at the lowest quantile can be achieved by maintaining regularity of ICDS services, early initiation of breastfeeding, reduction in sibling size (proxy for fertility size), increasing mother's age at rst birth, mass media exposure, educational attainment and employability. Additionally, access to the programmes pertaining to initiation of early breastfeeding, securing access and reducing gender-gap in receipt of ICDS services, reduction of early childbearing, improving mother's nutritional status, and creation of household wealth found to be imperative to the households having the highest level of stunting. Because coe cient effects indicate allinclusive returns to endowments, arguably, not only the 'reach' of these programmes, but also ensuring 'quality' of these programmes also could enhance child nutritional status. Although earlier studies have also demonstrated the in uence of these characteristics to lowering stunting [37][38], these studies could not able to quantify the contribution of reach of various policies and programmes in reducing stunting. institutional strengthening through capacity building of staff, improved infrastructure and outreach; strengthening child-relevant resources and facilitating uptake of principal schemes and services etc. The state plans for action also emphasized 'breaking the intergenerational cycle of malnutrition' by provisioning take-home ration and ensuring safe health and hygiene practices through better outreach services, particularly in the lower performing districts. The said action plan must also accommodate the issue of intersectoral coordination in implementation of these programmes in order to harness better dividend of these schemes.
Some limitations of the study should be acknowledged. First, NFHS sampling frame of 1998-99 does not allow to separate districts from the states. However, because of unavailability of any other comparable dataset, it was compelling to segregate districts of undivided Bihar. This may under-or over-estimate the QR-CD results to a disproportionate extent. Secondly, CD exercise can provide reliable estimates only if the primary quantile regression includes all the important factors of child nutrition and is well-speci ed [28]. To note, the choice of determinants has been constrained by the coverage of NFHS, key variables considered by the previous literature were included in the present study [18,[28][29]. However, in such situation, the issue of endogeneity cannot be entirely ruled out, though necessary tests were carried out to get rid of this.
Thirdly, providing clinical interpretations of the effect size of the variables are beyond the scope of the present study. Finally, the 'coe cient effects' in such comparisons lump several potential effects together and not informative about speci c factors or actions [29]; thus, interpretations of coe cient effects are speculative. Nonetheless, this research helps to highlight important dimensions to child nutritional improvement during last two decades for the state of Bihar.

Conclusions
Inconspicuous presence of child nutrition in Millennium Development Goals (MDG) framework with an imperfect measure of child undernutrition (i.e. underweight) was criticised. However, the issue has gained considerable momentum in the Sustainable Development Goals (SDGs) as the ambition to 'end hunger, achieve food security and improved nutrition and promote sustainable agriculture' is captured in SDG 2. Further, at least 12 of the 17 Goals contain indicators, which are highly relevant to nutrition because of the fact that without adequate and sustained investments in good nutrition, the SDGs would not be realised.
Results of the present study suggest that child undernutrition in Bihar is not just from a lack of su cient and adequately nutritious and safe food, but from a host of intertwined factors linking healthcare, women's education and work, household wealth (including water, sanitation and hygiene) and more. In addition to scaling-up proven nutrition-speci c interventions in other Indian states, the state of Bihar, must focus on policy processes and their political underpinnings reduce the risk of child undernutrition. The study used fourth round of National Family Health Survey (NFHS) data, which is available publicly available. Before conducting the survey NFHS had taken ethical approval. For the present study, ethical approval is not required.

Consent for publication
Not applicable for this study.

Availability of data and materials
The datasets generated and/or analysed during the current study are available in the [DHS] repository,