Intergenerational nutrition benefits of India’s national school feeding program

Suman Chakrabarti; Samuel P. Scott; Harold Alderman; Purnima Menon; Daniel O. Gilligan

doi:10.1038/s41467-021-24433-w

Intergenerational nutrition benefits of India’s national school feeding program

Chakrabarti, Suman; Scott, Samuel P.; Alderman, Harold; Menon, Purnima; Gilligan, Daniel O. 2021-07-12 00:00:00 ARTICLE https://doi.org/10.1038/s41467-021-24433-w OPEN Intergenerational nutrition beneﬁts of India’s national school feeding program 1 1 1 1 1 Suman Chakrabarti , Samuel P. Scott , Harold Alderman , Purnima Menon & Daniel O. Gilligan India has the world’s highest number of undernourished children and the largest school feeding program, the Mid-Day Meal (MDM) scheme. As school feeding programs target children outside the highest-return “ﬁrst 1000-days” window, they have not been included in the global agenda to address stunting. School meals beneﬁt education and nutrition in par- ticipants, but no studies have examined whether beneﬁts carry over to their children. Using nationally representative data on mothers and their children spanning 1993 to 2016, we assess whether MDM supports intergenerational improvements in child linear growth. Here we report that height-for-age z-score (HAZ) among children born to mothers with full MDM exposure was greater (+0.40 SD) than that in children born to non-exposed mothers. Associations were stronger in low socioeconomic strata and likely work through women’s education, fertility, and health service utilization. MDM was associated with 13–32% of the HAZ improvement in India from 2006 to 2016. 1 ✉ Poverty Health and Nutrition Division, International Food Policy Research Institute, Washington, DC, USA. email: [email protected] NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w lobally, 149 million children are too short for their age disparities advocate for a multi-generation approach that and over half of these children live in Asia . Within India, addresses parental socioeconomic status (SES), child and ado- G38% of children were stunted in 2015–2016 (ref. ). Linear lescent health and development, and young adult’s capacity for growth failure is a marker of chronic undernutrition and multiple planning and future parenting . However, since interventions to pathological changes which, together, have been termed the improve maternal height and education must be implemented ‘stunting syndrome’ . Stunted children are at risk of not reaching years before those girls and young women become mothers, their developmental potential, thus stunting has large implica- empirical assessment of the effectiveness of such programs for tions for human capital and the economic productivity of entire reducing undernutrition among future offspring is challenging. 4–6 societies . The World Health Assembly set the ambitious target This paper studies the intergenerational nutrition beneﬁts of of reducing childhood stunting by 40% from 2010 to 2025 (ref. ), India’s MDM scheme. We use seven population level datasets a target that likely will not be met . Thus, it is imperative to spanning 1993 to 2016, including multiple rounds of National understand how countries can accelerate progress toward stunt- Sample Surveys of Consumer Expenditure (NSS-CES), National ing reduction. Family Household Surveys (NFHS), and India Human Develop- Though much focus has been placed on nutrition-speciﬁc ment Surveys (IHDS). We match cohorts of mothers by state, interventions during the 1000-day period from conception to the birth year, and SES with data on MDM coverage measured as the child’s second birthday, investments across multiple life periods proportion of primary-school-age girls receiving MDM using and which address underlying determinants are also important to data from the NSS-CES. Birth cohort ﬁxed effects and controlled 8,9 achieve stunting reductions . Interventions may work directly interrupted time series models are used to estimate the associa- through maternal–child biological pathways or indirectly through tion of mother’s exposure to the MDM scheme with the nutri- socioeconomic mechanisms. In India, women’s height and edu- tional status of her future children. We ﬁnd that maternal cohorts cational attainment are among the strongest predictors of child living in areas with higher coverage of the MDM scheme are less 10–15 stunting . likely to have stunted children than cohorts living in low coverage In the Indian context, a candidate intervention which poten- areas. This effect is robust to the inclusion of a broad set of tially improves both women’s height and education—and which, controls at multiple levels and ﬁxed effects. Controlled inter- therefore, may lead to reductions in stunting among children rupted time series models conﬁrm that the 14 states which rolled born to these women—is the national school feeding program, out MDM in the late-1990s experienced improvements in child the Mid-Day Meal (MDM) scheme . Launched in 1995 by the height earlier than the rest of the nation, which scaled up MDM Government of India, the MDM scheme provides a free cooked in the 2000s after the Supreme Court mandate. Plausibility is meal to children in government and government-assisted primary supported by our ﬁndings of MDM association with participants’ schools (classes I–V; ages 6–10 years). The mandated minimum education, age at birth, number of children, use of antenatal care, meal energy content is 450 kcal and the meal must contain 12 g of and delivery in a medical facility. protein. In 2016–2017, 97.8 million children received a free cooked meal through the scheme every day, making the MDM Results scheme the largest school feeding program in the world . Program description and motivation. The MDM scheme, Econometric evaluations of India’s MDM scheme have shown a initiated by the central government in 1995, was intended to 18,19 positive association with beneﬁciaries’ school attendance , cover all government schools under the National Programme of 20 21 learning achievement , hunger and protein-energy malnutrition , Nutritional Support for Primary Education . Due to institu- and resilience to health shocks such as drought —all of which tional challenges, only a few states scaled up the program may have carryover beneﬁts to children born to mothers who immediately. NSS-CES data from 1999 show that only 6% of all participated in the program. We are not aware of studies that have girls aged 6–10 years received mid-day meals in school (Fig. 1). explored whether program beneﬁts for the MDM or similar pro- Between 1999 and 2004, program coverage increased in many grams in other countries extend to the next generation. Filling this states, largely due to an order from the Supreme Court of India research gap is critical, as (1) stunting carries over from one gen- directing state governments to provide cooked mid-day meals eration to the next and is therefore optimally studied on a multi- in primary schools . In 2004, 32% of Indian girls aged 6–10 23–26 generational time horizon , (2) school feeding programs are years were covered by the program. Finally, following a sub- implemented in almost every country , and (3) social safety nets stantial increase in the budget allocation for the program in such as India’s MDM scheme have the potential for population- 2006, by 2011, 46% of girls aged 6–10 years beneﬁted from the level stunting reduction as they are implemented at scale and target program. Coverage among boys was similar throughout this multiple underlying determinants in vulnerable groups . period. NSS-CES data show that substantial state variability in At a broader level, a substantial literature documents effects of MDM rollout existed even ten years after the central mandate. cash transfer programs on education of girls in low- and lower- A complete listing of state heterogeneity in program roll-out middle-income countries . While transfer programs clearly can be found in Supplementary Table 1. address food security, their track record on improving anthro- Our empirical exploration of the intergenerational beneﬁts of pometry is mixed at best, possibly because evaluations focus on the MDM scheme was motivated by the observation that stunting 28,30 relatively short-term impacts . However, even in the United prevalence was lower among children aged 0–5 years in 2016 in States, a timely transfer—for example, the Supplemental Nutri- states where MDM coverage was higher in 2005 (Fig. 2). The tion Assistance Program—has been shown to have health beneﬁts ability of historical MDM coverage to predict the prevalence of over time . Other studies document effects of cash transfers, stunting in 2016 suggests that a mother’s exposure to the program health insurance, and other programs for children in beneﬁciary during primary school may have future returns for her children. households on future adult outcomes such as incomes, achieved However, the observed association may be biased because policy 32 33,34 35 schooling , nutritional status , and mortality . variables in observational data are unlikely to be independent of The described literature suggests a potential pathway through latent individual and institutional characteristics . which school feeding programs and other cash transfer or in-kind safety nets focused on education may have intergenerational effects on child nutrition outcomes. Current frameworks for Birth cohort ﬁxed effects analyses. To inform the birth cohort understanding the intergenerational transmission of health ﬁxed effects analysis, we examined coverage and scale-up of the 2 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE Mid-day meal (MDM) program participation among girls aged 6-10y Stunting among children <5y No data 0-9.9% 10-19.9% 20-29.9% 30-39.9% 40-49.9% ≥50% 38% 46% 32% Coverage: 6% Year: 1999 2004 2011 Pathway to impact: MDM is offered at school Adolescent girl (now 12-17y) who Children (<5y) of women (6-10y); school enrolment was exposed to MDM in 2004 has exposed to MDM in 2004 and attendance increases increased years of education and (now 17-22y) are less likely height to be stunted Fig. 1 Overview of study design and proposed pathway. Coverage refers to the proportion of girls aged 6–10 years who received a MDM in school. Source for MDM program coverage data (green maps): NSS-CES 55 (2000), 61 (2005) and 68 (2012). Source for child stunting data (red map): NFHS4 (2016). MDM, mid-day meal. Source data are provided as a Source Data ﬁle. ab 60 -1 Low SES Middle SES High SES -1.2 30 -1.4 -1.6 -1.8 1980 1986 1992 1998 1980 1986 1992 1998 Mother's birth year Mother's birth year Fig. 3 MDM coverage and child HAZ by mother’s birth year and socioeconomic status. Bottom 3 deciles are the poorest households in the 0 10 20 30 40 50 60 70 sample and top 4 deciles are non-poor. MDM exposure of women born State-level MDM coverage in 2005, % between 1980 and 1998 (a) and HAZ of children under 5 years old in 2016 Fig. 2 Association between stunting prevalence among children under 5 of mothers born between 1980 and 1998 (b). Source of MDM coverage years old in 2016 and MDM coverage among girls 6–10 years old in data: NSS-CES 50 (1994), 55 (2000), and 61 (2005). Source of HAZ data: 2005. Each circle represents an individual state in India, with the size NFHS 4 (2016). HAZ height-for-age z-score, MDM mid-day meal. Source representing the state population size. Fit line and shaded 95% conﬁdence data are provided as a Source Data ﬁle. interval are also weighted by state population size. Sources: NFHS 4 (2016) for stunting data and NSS-CES 61 (2005) for MDM coverage data. MDM maternal birth year, wealth, state, and state-speciﬁc-birth-year mid-day meal. Source data are provided as a Source Data ﬁle. ﬁxed effects, as well as a set of child-speciﬁc controls, HAZ in children born to mothers who lived in areas with 100% MDM coverage was 0.40 SD higher than HAZ in children born to MDM scheme and HAZ of children by mother’s birth year and mothers living in areas without the MDM (p < 0.05). The SES. The rate of MDM scale-up across SES deciles moved in inclusion of ICDS and PDS access variables did not attenuate tandem with child HAZ along the mother’s birth year axis this association. The effect of the program varied by SES; children (Fig. 3). Later-born mothers from poor households were more from poor households had the largest effect (0.5 SD, p < 0.05) likely to be exposed to the program than either earlier-born followed by children from middle SES strata (0.33, p < 0.05), mothers or mothers from non-poor households (Fig. 3a). HAZ in relative to children from the wealthiest SES strata. In robustness children also increased with later mother’s birth year and was checks, program access coefﬁcients were slightly attenuated but higher in non-poor households compared to poor households remained signiﬁcant when adding birth year speciﬁc SES ﬁxed (Fig. 3b). The observed trends provide motivation for using effects but were not signiﬁcant after adding birth year and state- MDM rollout by mother’s birth year as a source of variation that speciﬁc SES ﬁxed effects. Further, regressions on subsamples of is time varying and cohort speciﬁc . stunted children showed higher precision but smaller coefﬁcients In the birth cohort model, maternal MDM coverage was for the beneﬁts of MDM coverage on HAZ compared to children associated with future child HAZ (Fig. 4a). After adjusting for who were not stunted (Supplementary Fig. 5). NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 3 Stunting prevalence in 2016, % MDM coverage, % Height-for-age-z-score, SD ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w a b Intervention Control Intervention Control -1 -1.1 MDM coverage 20 -1.2 10 -1.3 0 -1.4 -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4 Event time (0=birth year 1992) Event time (0=birth year 1992) -.6 -.4 -.2 0 .2 .4 .6 .8 c d -.018 -.03 .5 Parallel trends -.021 Parallel trends .00093 -.019 -.028 MDM x poor .039 .046 MDM x middle DID of slopes .042 DID of slopes .055 .045 .029 -.209 -.1 -.05 0 .05 .1 .15 -.1 -.05 0 .05 .1 .15 MDM coverage Height for Age Z-score (SD) per year Height for Age Z-score (SD) per year -.6 -.4 -.2 0 .2 .4 .6 .8 District RE District FE State FE Low SES Middle SES High SES HAZ(SD) difference at full MDM coverage Fig. 5 Relationship between MDM and future child HAZ: controlled Fig. 4 Relationship between MDM coverage and future child HAZ: birth interrupted time series analyses. All models exclude Kerala and Tamil cohort ﬁxed effects model. Panel a shows the relationship between MDM Nadu. Panel a shows MDM coverage by event time across intervention and coverage and future child HAZ in the birth cohort model (Eq. 1) while panel control states. The program begins between event time 0 and 1. Panel b b shows the relative association across wealth strata (Eq. 2). The circles shows the local polynomial of HAZ of children in 2016, born to women represent the point estimates and whiskers are 95% conﬁdence intervals. belonging to birth cohorts, before and after the start of the program in each Point estimates are interpreted as the difference in HAZ due to 100% state. The shaded gray area indicates the 95% conﬁdence interval. Panel c exposure to the MDM scheme during primary school years for the relevant shows the coefﬁcient on γ (parallel trends) and γ (DID) from Eq. (3). 6 7 sample. Point estimates in panel b for MDM × poor and MDM × middle are Coefﬁcients from three models are speciﬁed as Eq. (3) plus random effects the relative effect of 100% MDM coverage for that SES stratum compared and ﬁxed effects for district and state. Panel d: γ (parallel trends) and γ 6 7 to the average effect of 100% MDM coverage for the wealthiest four (DID) from Eq. (3) with state ﬁxed effects run on a subset of low (SES 1–3), deciles (MDM coverage). MDM coverage is the proportion of girls born middle (SES 4–6), and high (SES 7–10) households. The squares/diamonds between 1980 and 1998, within state-speciﬁc socioeconomic status deciles, represent the point estimate and whiskers are 95% conﬁdence intervals. who reported receiving at least 10 meals free of cost at school in the The DID coefﬁcient can be interpreted as the difference in the average rate previous month. All models control for child age, sex, birth order, maternal of change in HAZ, per-year, before versus after MDM started, in the antenatal care (4+ visits), institutional birth, residence (urban/rural), intervention compared to control states All models control for child age, religion, caste, access to services from the Integrated Child Development sex, birth order, maternal antenatal care (4+ visits), institutional birth, Services (dummies for receiving take home rations, child health check-ups, residence (urban/rural), religion, caste, access to services from the pre-school education, weight measurements, and nutrition counseling) and Integrated Child Development Services (dummies for receiving take home the Public Distribution System (household has a Below Poverty Line card to rations, child health check-ups, pre-school education, weight obtain subsidized food). The models include ﬁxed effects for mother’s birth measurements, and nutrition counseling) and the Public Distribution year, state, household wealth, and for state × mother’s birth year. All System (household has a Below Poverty Line card to obtain subsidized models cluster standard error estimates at the district level. Sources: NFHS food). All models cluster standard error estimates at the state level. 4 (2016) for outcome and covariates. NSS-CES 50 (1994), 55 (2000), and Sources: NFHS 4 (2016) for outcome and covariates. NSS-CES 50 (1994), 61 (2005) for MDM coverage data. HAZ height-for-age z-score, MDM 55 (2000), and 61 (2005) for MDM coverage data. FE ﬁxed effects, MDM mid-day meal, SES socioeconomic status. Source data are provided as a mid-day meal, RE random effects, SES socioeconomic status. Source data Source Data ﬁle. are provided as a Source Data ﬁle. Controlled interrupted time series analyses. The controlled Program pathways. When examining factors that the MDM may interrupted time series model exploits variation in the timing of work through to inﬂuence child HAZ, full MDM coverage during the expansion of the MDM program to estimate program beneﬁts primary school years was a meaningful predictor of all factors relative to a reference period (event time 0 = birth year 1992). examined (Table 1). Full MDM coverage predicted 3.9 years of MDM expansion between event time 0 and 4 (birth years attained maternal education in years, delaying age in years at ﬁrst 1992–1996 capturing the short run impact of the program) dif- birth by 1.6 years, having a fewer (−0.8) children, a higher fered substantially across intervention and control states (Fig. 5a). probability of having at least four antenatal care visits (22%), and Trends in child HAZ were parallel between event time −4 and 0 giving birth in a medical facility (28%) (all p < 0.001). Full MDM across intervention and control states (Fig. 5b). After event time coverage predicted higher adult height among direct beneﬁciaries 0, intervention states saw a larger change in child HAZ compared (0.51 cm) but the association was not statistically signiﬁcant. to control states. In regression models, the coefﬁcient for parallel trends was not signiﬁcant, conﬁrming that trends in child HAZ were statistically similar across intervention and control states Regression decomposition. Our ﬁndings can be put into context before the intervention (Fig. 5c). The estimated association was by considering changes in HAZ among children under 5 years of similar across all three speciﬁcations, 0.038, 0.041, and 0.044 SD age reported in the National Family Health Surveys. HAZ per year (p < 0.05). Relative to wealthier households, the effect improved by 0.4 SDs between 2006 and 2016, on average. Using estimate of the MDM in intervention states was larger among Eq. (4), with an average MDM coverage of 32% in 2004 at the poor and middle-income households at 0.044–0.055 SD per year national level (NSS-CES 61) multiplied by the effect size of 0.166 (p < 0.10) (Fig. 5d). In robustness checks, effect coefﬁcients were SD (raw data model) to 0.401 SD (smoothed data model), we stable when excluding Gujarat, Odisha, and Chhattisgarh (some estimate the MDM explains 0.053–0.128 SD or 13.3–32.1% of districts in these states adopted MDM after Tamil Nadu and average change in HAZ. Using Eq. (5), with an average of 2.6 Kerala) from treatment states (Supplementary Fig. 2). years of exposure multiplied by the effect size of 0.044 SD per 4 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications MDM coverage, % Height-for-age z-score, SD NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE Table 1 Relationship between MDM and direct beneﬁciary education, height, fertility, and health service use in Indian women born between 1980 and 1998. Education, years Height, cm Age at ﬁrst Children, number Antenatal Institutional birth, years care, binary birth, binary Coefﬁcient 3.95 0.51 1.62 −0.80 0.22 0.28 Standard error (0.46) (0.36) (0.20) (0.07) (0.03) (0.02) P value <0.000 0.163 <0.000 <0.000 <0.000 <0.000 R 0.38 0.12 0.32 0.36 0.24 0.15 N 218,810 215,812 218,810 218,810 218,528 218,218 Coefﬁcients are from Eq. (1). Point estimates are interpreted as the difference in the outcome due to 100% exposure to the MDM scheme during primary school years. MDM coverage is the proportion of girls born between 1980 and 1998, within state-speciﬁc socioeconomic status deciles, who reported receiving at least 10 meals free of cost at school in the previous month. All models control for residence (urban/rural), religion, and caste. The models include ﬁxed effects for mother’s birth year, state, household wealth, and for state-speciﬁc mother’s birth year. All models cluster standard error estimates at the district level. Sources: NFHS 4 (2016) for outcome and covariates. NSS-CES 50 (1994), 55 (2000), and 61 (2005) for MDM coverage data. MDM mid-day meal, SES socioeconomic status. year, we estimate the MDM can explain 0.114 SD or 28.6% of expenditure, assets, and parental education. We ﬁnd that girls in average change in HAZ. The range estimated contributions are government schools are, on average, 0.89 cm shorter than those in similar in magnitude and relatively substantial, considering that private schools (p < 0.001). When we add a dummy variable HAZ is dependent on a large set of determinants, of which each indicating the receipt of MDM during primary school for these can individually only explain a small part of total variation in girls (identiﬁed in IHDS-1), we ﬁnd that MDM is associated with South Asian countries . a higher height of 1.3 cm on average (p < 0.001), while govern- ment school attendance is associated with 1 cm lower height (p < 0.001). This suggests that selection effects from program place- Migration. A possible concern for our estimates is susceptibility ment in government schools are likely to bias our estimates to the effects of migration. Since we measure MDM exposure at downward and that MDM is the driver of higher height among the state level in the past and associate it with child nutrition in government school beneﬁciaries. the future, attribution of the estimated role of MDM exposure would be weakened in the presence of substantial migration across states. A recent study allays this concern by providing Testing ﬁxed effects models with raw MDM coverage data. The estimates on migration in India. Although 30% of India’s popu- MDM coverage estimate from a regression model using Eq. (1) lation has ever migrated, two-thirds are intra-district migrants, and the raw coverage data is statistically signiﬁcant but attenuated more than half of whom are women migrating for marriage .In to 0.166 SD as expected (Supplementary Table 5, model 1). We 2001, only 4% of India’s population migrated across state also speciﬁed a second set of regressions using only the 2004 NSS borders . Therefore, migration is not a major concern for mis- data, and matched MDM coverage by district and SES. Again, we classiﬁcation of treatment status in our models. ﬁnd an attenuated but signiﬁcant coefﬁcient of 0.115 SD (Sup- plementary Table 5, model 2) and, as expected, a larger coefﬁcient Discordant SES matching between NSS-CES and NFHS. of 0.189 SD among poor households (p < 0.05) (Supplementary Overall, we ﬁnd a 78% concordance between expenditure-based Table 5, model 3). As the district level exposure does not have SES deciles measured in 2005 and asset-based SES deciles mea- temporal variation by birth year, this model is not directly sured in 2012 at the state level using India’s IHDS (Supplemen- comparable with the birth cohort model. However, it does tary Fig. 3). Given this 22% discordance, we cannot rule out that demonstrate that MDM coverage variation by district and SES is our estimates are somewhat biased due to imperfect classiﬁcation strongly correlated with HAZ of children of mothers born by SES status. However, the degree of bias is likely to be small between 1993 and 1997. because mobility across deciles is limited (the IHDS shows that a The MDM coverage estimate from a regression model using household generally only moves up by one or two SES deciles Eq. (1) and the log-linear smoothed coverage data are statistically over 7 years, if they move at all) and MDM coverage within states signiﬁcant but attenuated to 0.261 SD (Supplementary Table 6, does not ﬂuctuate greatly with small increments of SES classes (in model 1). However, attenuation here is smaller in magnitude 2005, coverage in the IHDS sample ranged between 53 and 62% compared to those using raw data. The model using Eq. (2) shows in the bottom four SES deciles). Moreover, non-differential that children from poor households had the largest effect (0.468 misclassiﬁcation as a form of measurement error generally tends SD, p < 0.01) followed by children from middle SES strata (0.296, to bias estimates towards the null . p < 0.05), relative to children from the wealthiest SES strata (Supplementary Table 6, model 2). Overall, we conclude that both the smoothed and raw data Matching by caste and religion. To test the sensitivity of our models provide evidence of an effect of maternal MDM coverage estimates to demographic measures of socioeconomic position on child anthropometry, though the size of the effect depends on that are less likely to change over time, we matched MDM cov- the preferred model. We have provided evidence that this effect is erage by state of residence, caste, and religion. Similar to SES robust to varying model speciﬁcations, and that the effect of matching, adjusted full maternal MDM coverage using caste and MDM coverage is largest among the poorest households. religion matching was associated with an improvement in HAZ Moreover, the control interrupted time series models do not among children aged 0–59 months (Supplementary Fig. 4). use smoothed coverage but provide qualitatively similar estimates. Selection bias from program placement in government schools. Using 2011 data (IHDS-2) we tested whether girls aged 11–17 years in government schools are shorter than those in private Discussion schools. We ﬁt a model with state ﬁxed effects that controls for We have shown that investments made in school meals in pre- child age, urban residence, occupation, household size, household vious decades were associated with improvements in future child NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w linear growth. The plausibility of this ﬁnding is supported by an received free meals at school from the National Sample Survey of Consumer 44–47 Expenditure (NSS-CES) (1993, 1999, and 2004 rounds) . These data are association between MDM exposure and underlying determi- combined with data on child height-for-age z-scores in 2016 from wave four of nants of child linear growth: women’s education, fertility, and India’s Demographic Health Survey, the National Family Health Survey (NFHS) . health service use. As the analysis covers a large nationally Both are large nationally representative surveys, which make it possible to match representative sample of households, the results reﬂect a program exposure to MDM by cohorts of girls born between 1980 and 1998 at the district level with data on mean child height in the same locations in 2016. The 2016 implemented at scale, with all its ﬂaws, and not a pilot program NFHS4 sample included 217,940 women with 196,310 children under 5 years of designed to provide proof of concept. This, of course, comes at a age. NSS-CES data from 2011 were also used for generating maps for coverage but cost; we could not follow a randomized cohort of girls from not for the primary analyses. Our interest was examining next generation beneﬁts primary school to childbearing. We put the magnitude of the on child stunting and our hypothesis was that intergenerational effects work through ﬁrst generation improvements in education, height, fertility, and access to association into context by using regression decomposition to 14,43,49–52 health services . We expected larger inﬂuence of maternal coverage estimate the share of the actual HAZ improvement explained by compared to paternal coverage given previous evidence showing larger program the predicted MDM effect on HAZ. impacts on girls than on boys . We support our main ﬁndings by using the 2004 While others have examined the effects of school feeding and 2011 rounds of Indian Human Development Surveys for descriptive analyses 53,54 programs on education and nutrition in beneﬁciaries themselves, and robustness checks . IHDS provides a wide array of variables that are not available in the NSS or the NFHS and offers supportive evidence on the main to our knowledge our paper is the ﬁrst to demonstrate an inter- estimates and model assumptions. Our study was a secondary analysis of existing generational transmission of beneﬁts. This ﬁnding provides evi- public survey data; hence, no ethical approval was required for our study. All dence that, when intergenerational effects are considered, the surveys complied with ethical norms with appropriate approvals and consent taken complete beneﬁt of school feeding programs at scale for linear at the time of survey. Summary statistics for the primary and secondary outcomes examined in this paper are shown in Supplementary Table 3. Summary statistics growth is much larger than previously understood. The result that for the covariates from NFHS are shown in Supplementary Table 4. a school feeding program is related to the nutritional status of children in the next generation also has important implications Identiﬁcation strategy. In an ideal experiment, children would be randomly for other transfer programs. The literature generally focuses on assigned access to free lunches from the MDM program in primary school and we investments in nutrition during the 1000-day period to reduce would compare the average HAZ outcomes for the children of the MDM bene- childhood stunting; our ﬁndings suggest that intervening during ﬁciaries and of the MDM non-beneﬁciaries when the original children in the the primary school years can make important contributions to experiment reached adulthood. In the absence of randomized treatment allotment, we chose to use panel data techniques from repeated cross-sections to exploit the reducing future child stunting, particularly given the cumulative strengths of the available data for identiﬁcation—the fact that the data cover birth exposure that is possible through school feeding programs. cohorts over a long period and that MDM coverage varies by state of residence and School meal programs are often motivated by their potential to SES. SES was calculated using a principal component analysis of household assets, increase schooling, particularly that of girls. While enrolment including cooking fuel, ﬂoor and wall materials, land and house ownership; and the possession of assets, including a mattress, pressure cooker, chair, bed, table, fan, parity is within reach in primary schooling –between 2000 and TV, sewing machine, phone, computer, fridge, watch, bicycle, motorbike, car; and 2015, the number of primary school-age children not in school the possession of animals, including cows, goats, and chickens. declined globally from 100 million to 61 million —there is a larger goal of primary and post primary school completion. Very Birth cohort ﬁxed effects analyses. Year of birth, SES decile, and state of resi- little in the literature on school meal programs can quantify dence were used to determine an individual’s exposure to the program. In India, program contribution to total years of schooling completed. children are expected to attend primary school between the ages of 6 and 10 years. Moreover, evidence that the scale-up of school meals is associated The NSS-CES provide data on the age of all household members and whether they received free meals at school in the past 30 days. Of all the girls aged 6–10 years in with increased heights of women—in a population in which the 2005 NSS sample who reported receiving any free meals at school (N = 8873), stunting has been historically linked with maternal under- 95.6% reported receiving at least 10 meals in the previous month. We used a nutrition—provides a new perspective on the contribution of minimum of 10 meals per month to ensure that our coverage estimates were for such programs. This reinforces an increased attention to seeking children who received the program with ﬁdelity. Models were run separately using any MDM access (at least 1 meal) and comparable results were obtained. We use opportunities to improve nutrition in the “next 7000 days” , that this information to calculate the percentage of all girls aged 6–10 years covered by is, to ﬁnd means of addressing undernutrition should efforts in the program for cohorts born between 1980 and 1998. This period gives us an the high priority period prior to a child’s second birthday not be approximately equal number of birth cohorts who were born before and after the fully successful. The results here show that school meals may introduction of the MDM scheme. Since the MDM scheme was introduced in 1995, those born after 1989 would be able to receive free meals in primary school. In contribute to education, nutrition (height), later fertility deci- addition, the NSS-CES provide measures of SES and state of residence, which sions, and access to health care; by doing so, school meals may allowed us to calculate coverage rates for all girls aged 6–10 years, speciﬁc to each reduce the risk of undernutrition in the next generation. In its SES strata in all Indian states. current form, India’s MDM scheme has the potential to address For any cohort, MDM exposure is a function of the number of years an average multiple underlying determinants of undernutrition. Improving child spends in primary school and when the program started in the school they attended. In an ideal data setting, to obtain an accurate coverage estimate for a the quality of meals provided and extending the program beyond birth cohort, we would have data from ﬁve cross-sections surveyed consecutively. primary school might further enhance its beneﬁts , though we For example, to obtain an estimate of MDM coverage for the 1994 cohort, we could not empirically test these hypotheses given the ideally would have coverage data on 6-year-old children measured in 2000, 7-year- available data. olds in 2001, 8-year-olds in 2002, 9-year-olds in 2003, and 10-year-olds in 2004. We would then average these ﬁve coverage estimates into a single estimate, The MDM is mandated by the Supreme Court of India as a representing average MDM exposure assuming a typical ﬁve-year period in social protection program addressing food insecurity. The social primary school for the 1994 cohort. The averaging is necessary because any single protection role of addressing hunger and food insecurity may be a year does not accurately reﬂect exposure for all 5 years in primary school. justiﬁcation by itself for school-based transfers in many settings . Each NSS-CES repeated cross-section, conducted within 5 year intervals, provides MDM coverage by child age as measured in the survey year. We used However, evidence such as presented here depict these programs linear interpolation to estimate a smoothed continuous exposure indicator that as contributing to both food security and to improved outcomes varies by maternal birth year, state, and SES. For example, using coverage estimates in the next generation, thus contribute to the policy framework for 6 year olds in 1999 NSS-CES (birth year 1993) and 6 year olds in the 2004 NSS- for school-based interventions. CES (birth year 1998), we ﬁrst used linear interpolation to estimate the average rate of increase in MDM coverage for 6 year olds for the years (2000, 2001, 2002, and 2003) with no NSS-CES data (these correspond with birth years 1994, 1995, 1996, Methods and 1997). Next, we performed similar interpolation for 7-, 8-, 9-, and 10-year-old Data sources. This paper relies on evidence from seven rounds of three publicly children. This provided smoothed coverage estimates for children born in 1993 for available nationally representative surveys (Supplementary Table 2). The primary the survey years 1999, 2000, 2001, 2002, and 2003—the years the 1993 cohort analysis in this paper uses data on whether children born between 1980 and 1998 would have aged from 6 to 10 years. We take the average coverage for these 5 years 6 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE as the ﬁnal estimate of coverage experience of a speciﬁc birth year (Supplementary households to attend primary school (and to improve nutrition); therefore, the Fig. 1). This process of smoothing (i) estimates the relationship between maternal estimates in Eq. (1) are likely to mask heterogeneity of response to the program. school meals exposure and annual child HAZ outcomes under an assumption of a Masking is anticipated because outcome data from children sampled from non- linear trend in exposure and (ii) reduces probable bias due to measurement error poor households, who would be more likely to opt out of the government school present in the raw data by moving extreme values closer to the center of the system in favor of private schools, would inﬂuence average effect sizes . We expect distribution. that mothers who were enrolled in government schools during their childhood Throughout the paper, we use the term MDM coverage, which refers to an would have worse nutritional outcomes and this might place a downward bias on estimate of state-by-year average program exposure during primary school for the our estimates. To investigate the existence of such heterogeneity, we compared birth cohorts in the sample, under the assumption that coverage increases in a associations across SES groups. We created SES deciles and grouped women in the linear fashion within age groups of children in primary school surveyed in the years bottom three (poor), middle three, and top four (non-poor) deciles to create two 1993, 1999, 2004, and 2011. It is almost certain that exposure in the interval wealth strata. We estimated models for differential associations for poor, middle between 2 years lies between the values in the end points; the assumption that the versus non-poor households by modifying Eq. (1) as follows: expansion is linear is a plausible pattern of program roll out. We assume that Y ¼ γ þ γ MDM Poor þ γ MDM Middle þ γ MDM within 5-year intervals, the duration of primary school, age-speciﬁc trends in iwst wst wst wst wst wst 0 1 2 3 ð2Þ coverage would have increased gradually. Gradual rollout is typical of at-scale þ γ T þ γ S þ γ S T þ γ W þ γ C þ γ P þ ε 4 t 5 s 6 s t 7 w 8 iwst 9 iwst iwst programs in developing countries with numerous implementation, ﬁnancing, and bureaucratic challenges . However, in sensitivity analyses, we subject the where Poor and Middle dummy variables for bottom three and middle (4–6) wst wst assumption of a linear scale-up to an additional robustness check where we smooth SES deciles, respectively, with the top four SES deciles serving as the reference non- MDM coverage using a log-linear process. poor group. γ and γ measure if poor and middle SES households beneﬁtted more 1 2 Next, using birth year, SES deciles and state of residence, we match NFHS data from MDM coverage compared to non-poor households. We expect γ to be larger with NSS-CES data for the percentage of girls covered by the MDM for cohorts than γ , and if these coefﬁcients are statistically signiﬁcant and of a large order, born between 1980 and 1998. NFHS data provide anthropometric measurements then we have evidence that MDM program beneﬁts differ by SES. Note that SES for the last three births for each mother. We use data for all available children with here is current, and mother’s SES may have differed in childhood. To this end, we valid anthropometric measurements. We calculate HAZ using the “zscore06” offer evidence in our sensitivity analyses that SES mobility is likely modest. STATA routine which automatically excludes outlier measurements. We specify the following model: Controlled interrupted time series models using state rollout timing. The birth cohort model exploits variation in treatment measured as the proportion of children Y ¼ γ þ γ MDM þ γ T þ γ S þ γ S T þ γ W þ γ C þ γ P þ ε iwst 0 1 wst 2 t 3 s 4 s t 5 w 6 iwst 7 iwst iwst covered by the program within a birth year, state, and across SES strata. It allows us ð1Þ to express the relationship between MDM and HAZ as a function of coverage. However, it comes at the cost of potential for endogeneity because MDM coverage where Y is the height-for-age z-score for child i belonging to SES strata w in iwst could potentially be associated with changes in living conditions that vary within state s in mother’s birth year t. MDM is a continuous indicator coded as the wst cohorts deﬁned by state, birth year, and SES strata. An alternate model exploits the proportion of mothers covered by the MDM as children and ranges between 0 and differential timing of MDM rollout across Indian states as a robustness check on 1. T represents birth-year ﬁxed effects which forces identiﬁcation of within birth- the birth cohort model. This alternative can reveal insights for the short-term year effects and controls for time-varying national level economic changes, cumulative beneﬁts of the program . programs, and policies. Examples of these are national programs such as the States implemented the program at different times; de-facto, the program was National Health Mission introduced in 2005 (ref. ) and changes in national GDP, rolled out in the three phases (Supplementary Table 1). According to the NSS data, which has shown robust growth . MDM coverage patterns by state and birth year show that Tamil Nadu and Kerala, We estimated Eq. (1) using MDM coverage at the state level disaggregated by i.e. “phase 1” states, had average coverage greater than 20% for maternal birth year wealth strata. W represents the wealth-decile ﬁxed effects and provides controls 1988). These states initiated school feeding programs well before the central for all unobserved time-invariant factors associated with household wealth and government funded MDM. Following the central government order, in phase 2, MDM coverage. S is the state ﬁxed effects which controls for all for time-invariant other states—Odisha, Himachal Pradesh, Uttaranchal, Haryana, Rajasthan, Sikkim, differences across states with high and low MDM exposure. S * T or state-birth- s t Tripura, West Bengal, Chhattisgarh, Madhya Pradesh, Gujarat, Maharashtra, year ﬁxed effects controls for unobserved state-speciﬁc time-varying factors that Andhra Pradesh, and Karnataka—implemented the program at scale with coverage could be correlated with the outcome such as the state’s political climate, varying increasing by more than 10% between maternal birth years 1992–1996. In the degrees of implementation of welfare programs, agricultural policies, and remaining states (phase 3), MDM coverage was below 5% and increased by less educational subsidies. A concern for a model estimated without this parameter is than 10% between maternal birth years 1992–1996. that states that introduced free meals in primary school at different times and rates These roll-out patterns lend themselves to analysis using a controlled of coverage expansion could be systematically different. For example, states with 63–65 interrupted time series design (CITS) . Conceptually, the CITS is a residents who had lower education or poorer nutritional status on average may combination of the difference-in-differences and interrupted time series models. It have been more likely to introduce the MDM. Similarly, states with better includes a within group before–after comparison, and a between-group governance may have been better equipped to implement the MDM program at comparison, strengthening the control for potential confounders. The ﬁrst scale. In either case, the correlation between outcomes and MDM implementation difference is the change in the outcome trend within each group, comparing the could be confounded with unobserved state-speciﬁc time-varying factors. period before MDM to the period after (slope change). The second difference is the C represents a vector of individual, household and survey-speciﬁc controls, iwst difference in slope changes in the control group compared to the intervention including child age, sex, birth order, mothers antenatal care status during group (difference-in-differences of slopes). The CITS reduces bias due to other pregnancy, birth in a medical facility, and household characteristics at the time the interventions or events occurring around the same time as the MDM intervention outcome was measured. The vector includes SES, caste, religion, and residence and allows comparison groups to start at different levels of the outcome. Moreover, (urban or rural). P represents a vector of individual and household-speciﬁc iwst the CITS controls for the improvement in HAZ that would be expected without the programmatic controls, including access to services from the Integrated Child MDM and tests for parallel trends within the model. Development Services (dummies for receiving take home rations, child health We exclude Tamil Nadu and Kerala from CITS analysis as they were early check-ups, pre-school education, weight measurements, and nutrition counseling) MDM implementers and both states have better nutrition outcomes compared to and the Public Distribution System (household has a Below Poverty Line card to other states in India. We focus on maternal birth years 1988 to 1996, when we have 59,60 obtain subsidized food) . Controlling for these variables reduces possible a pre intervention period with no MDM across all states, and a post intervention confounding from government interventions that could beneﬁt current child period when some states introduced the program while others did not. Phase nutritional status. All standard error estimates were clustered at the district level. 2 states form the intervention group and phase 3 states serve as the control group. Clustering adjusts standard error estimates after accounting for intra-district We parameterize the CITS model using Eq. (3). correlations and assumes that residuals are independent across districts . The coefﬁcients estimated by Eq. (1) are intent-to-treat (ITT) estimates because Y ¼ γ þ γ Int þ γ T þ γ Post þ γ Post T þ γ Post Int ist 0 1 s 2 t 3 t 4 t t 5 t t the MDM coverage variable measures “potential exposure” to the program on ð3Þ þ γ T Int þ γ T Int Post þ γ C þ γ P þ ε t s t s t ist ist ist entire birth cohorts. Our ITT estimates are a policy-relevant parameter for an ex- 6 7 8 9 post analysis of the effects of a large program on the entire population (birth In state s in mother’s birth year t, Int is a dummy for the intervention states, T s t 21,22 cohorts) . Our models, based on population representative MDM coverage, is the event time, a discrete variable (for maternal birth years 1988–1996) that is estimate the magnitude of improvement in child undernutrition that can be centered at 1992 and ranges between −4 and 4. Post is a dummy for maternal expected if a cohort is potentially treated. birth years 1993–1996. In Eq. (3), γ tests the null hypothesis of parallel pre- intervention trends; if not signiﬁcant, we reject this hypothesis and conclude that pre-interventions differed between intervention and control groups. γ is the Testing for differential beneﬁts for the poor. IHDS data show that 80% of all MDM beneﬁciaries in 2004 attended government schools and that two-thirds of coefﬁcient of interest, and represents a “difference-in-difference of slopes” between children attending government schools were from low-income households (bottom the intervention and control states. If γ is statistically signiﬁcant, the change in six SES deciles), suggesting that MDM was primarily implemented in government HAZ slope for intervention states differs from the change in HAZ slope for control schools rather than in private schools as an incentive for children from poor states. In other words, it tests for faster gains in child linear growth for states with NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w MDM. To account for spatial heterogeneity, we run three speciﬁcations of Eq. (3) use only the 2004 NSS raw coverage data, matched by district and SES. These by adding district random effects, district ﬁxed effects, and state ﬁxed effects. models specify the same level of coverage to SES groups within districts for birth To explore heterogeneity, we investigate differential associations by household cohorts 1993 to 1997. SES by running models within subsamples of poor (SES deciles 1–3), middle (SES deciles 4–6), and non-poor (SES deciles 7–10) households. For robustness, we Testing ﬁxed effects models with MDM coverage smoothed using log-linear check sensitivity of coefﬁcients to exclusion of Gujarat, Odisha, and Chhattisgarh process. To test the sensitivity of our estimates using linearly smoothed coverage, from the intervention group. These states had greater than 10% coverage at event we use an alternative log-linear smoothing process. This process assumes an time 0 and thus could arguably be placed in the phase 1 category. exponential growth in MDM coverage within 5-year intervals. We then ﬁt models with Eqs. (1) and (2). Regression decomposition. To test the plausibility of our results we performed regression-based decomposition with our estimates from Eqs. (1) and (3) (ref. ). Reporting summary. Further information on research design is available in the Nature From Eq. (1), we estimated the population level effect of the program between 2006 Research Reporting Summary linked to this article. and 2016 with Eq. (4). γ MDM2004 ð4Þ Data availability ΔHAZ 44–48,53,54 The conclusions of this article are based on publicly available datasets . Source where γ is the coefﬁcient of MDM from Eq. (1), MDM is the MDM coverage 2004 data are provided with this paper. The cleaned and merged dataset is available on the in 2004 and ΔHAZ change in HAZ between 2006 and 2016. 67 Harvard Dataverse at [https://doi.org/10.7910/DVN/JTN87W] . Source data are From the controlled interrupted time series model, we estimated the effect of provided with this paper. exposure to the program using Eq. (5). γ PostEventTime Code availability ð5Þ ΔHAZ The analysis code that reproduces all the tables and ﬁgures in the manuscript is available where γ is the coefﬁcient from Eq. (3), PostEventTime is the average event time 5 on the Harvard Dataverse at [https://doi.org/10.7910/DVN/JTN87W] . years before and after the start of the program, and ΔHAZ is the change in HAZ between 2006 and 2016. Estimates from Eqs. (4) and (5) are proportions and are Received: 21 February 2021; Accepted: 7 June 2021; expected to be less than 1 because the predicted difference in HAZ explained by MDM must be less than the total change in HAZ observed between 2004 and 2016. Program pathways. We next investigated plausible pathways that might support intergenerational links between the MDM program and child nutrition. We used Eq. (1) to investigate the association of the MDM with six factors that may be References related to the MDM program and which, in turn, correlate with child HAZ: 1. Development Initiatives. 2020 Global Nutrition Report—Global Nutrition mother’s education and height, mother’s age at ﬁrst birth, total number of children Report. https://globalnutritionreport.org/reports/2020-global-nutrition-report/ per mother, number of antenatal care visits attended by the mother during preg- (2020). nancy and if the child was born in a medical facility. We recognize that this is a 2. Ministry of Health and Family Welfare Government of India. National Family plausibility analysis and cannot isolate causality. Health Survey-4, 2015-2016, India Fact Sheet. http://rchiips.org/nfhs/pdf/ NFHS4/India.pdf (2017). Discordant SES matching between NSS-CES and NFHS. We matched MDM 3. Prendergast, A. J. & Humphrey, J. H. The stunting syndrome in developing coverage by state of residence and SES decile between the NSS-CES and NFHS. countries. Paediatr. Int. Child Health 34, 250–265 (2014). This assumes that (1) mobility across SES strata over time is minimal and (2) SES 4. Development Initiatives. Global Nutrition Report 2017: Nourishing the SDGs deciles in NFHS correspond well with those in the NSS. We therefore use panel (2017). data from the IHDS to assess the concordance of expenditure-based SES deciles 5. Black, M. et al. Early childhood development coming of age: science through measured in 2005 and asset-based SES deciles measured in 2012. Since IHDS the life course. Lancet 389,77–90 (2017). follows the same individuals over 7 years, we can track their mobility across SES 6. de Onis, M. & Branca, F. Childhood stunting: a global perspective. Matern. strata over time and then compare their status on both SES measurements. Child Nutr. 12,12–26 (2016). 7. de Onis, M. et al. The World Health Organization’s global target for reducing Matching by caste or religious group. To test the sensitivity of our estimates to childhood stunting by 2025: rationale and proposed actions. Matern. Child demographic measures of socioeconomic position, we matched maternal MDM Nutr. 9,6–26 (2013). coverage by birth year, state of residence and households’ caste/religious groups in 8. Bundy, D. A. P. et al. Investment in child and adolescent health and the NSS-CES and NFHS. The social groups used to match households were development: key messages from Disease Control Priorities, 3rd Edition. scheduled caste (Hindu), scheduled tribe (Hindu), Muslim, Christian, and others. Lancet https://doi.org/10.1016/S0140-6736(17)32417-0 (2017). Similar to the SES model, this model works by assigning a probability of exposure 9. Black, R. E. et al. Maternal and child undernutrition and overweight in low- to MDM for maternal birth cohorts that varies by state, religion, and caste. While income and middle-income countries. Lancet 382, 427–451 (2013). social groups do not follow strict income hierarchies across states, they have the 10. Corsi, D. J., Mejía-Guevara, I. & Subramanian, S. V. Risk factors for chronic advantage of being largely time invariant and thus do not introduce biases that undernutrition among children in India: estimating relative importance, result from income mobility. population attributable risk and fractions. Soc. Sci. Med. 157, 165–185 (2016). 11. Headey, D., Hoddinott, J. & Park, S. Drivers of nutritional change in four Testing ﬁxed effects models with raw MDM coverage data. To test the sen- South Asian countries: a dynamic observational analysis. Matern. Child Nutr. sitivity of our estimates using smoothed coverage, we offer an additional alternative 12, 210–218 (2016). using raw coverage data from NSS. These coverage estimates are from cross- 12. Alderman, H. & Headey, D. D. How important is parental education for child sections at speciﬁc points in time and are not smoothed using the age proﬁles of nutrition? World Dev. 94, 448–464 (2017). children in the NSS rounds. We ﬁrst created a scatter plot of smoothed coverage 13. Cavatorta, E., Shankar, B. & Flores-Martinez, A. Explaining cross-state estimates against the raw coverage data to gauge the degree and direction of the disparities in child nutrition in rural India. World Dev. 76, 216–237 (2015). smoothing process (Supplementary Fig. 6). 14. Kim, R., Mejía-Guevara, I., Corsi, D. J., Aguayo, V. M. & Subramanian, S. V. The maps in Fig. 1 show a discrete jump in coverage from 6% in 1999 to 32% in Relative importance of 13 correlates of child stunting in South Asia: insights 2004. The smoothed data attempt to ﬁll in data gaps on coverage for the years 2000 from nationally representative data from Afghanistan, Bangladesh, India, to 2003. The scatterplot of the smoothed coverage data against the raw data shows Nepal, and Pakistan. Soc. Sci. Med. 187, 144–154 (2017). that the smoothed data are less extreme than the raw data, which has many 0 and 15. Chen, Y. & Li, H. Mother’s education and child health: is there a nurturing 100% coverage estimates. These extreme values present in the raw data likely reﬂect effect? J. Health Econ. 28, 413–426 (2009). measurement error for cohort-speciﬁc coverage because they do not capture the 16. Raghunathan, K., Chakrabarti, S., Menon, P. & Alderman, H. Deploying the transition of increasing coverage for the initial years of program implementation, power of social protection to improve nutrition what will it take? Econ. Polit. so that coverage for any observation reﬂects only a single year during a time of Wkly 52, (2017). program expansion despite the fact that a student will have spent more than a 17. Ministry of Human Resource Development. Mid Day Meal Scheme. single year in school. To test our hypothesis that measurement error in the raw Department of School Education & Literacy, Government of India. http:// coverage data would attenuate results compared to those from the models using mdm.nic.in/ (2017). smoothed data in keeping with standard expectation with random errors in 18. Afridi, F. The impact of school meals on school participation: evidence from variables, we ran our primary birth cohort model with raw coverage matched by rural India. J. Dev. Stud. 47, 1636–1656 (2011). state, SES, and birth years. We also ran a second test of sensitivity with models that 8 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE 19. Drèze, J. & Kingdon, G. School participation in rural India. London School of 49. Bhutta, Z. A. et al. Evidence-based interventions for improvement of maternal Economics and Political Science (1999). and child nutrition: what can be done and at what cost? Lancet 382, 452–477 20. Chakraborty, T. & Jayaraman, R. School feeding and learning achievement: (2013). evidence from India’s Midday Meal Program. IZA Discussion Paper 10086 50. Rieger, M. & Trommlerová, S. K. Age-speciﬁc correlates of child growth. (2016). Demography 53, 241–267 (2016). 21. Afridi, F. Child welfare programs and child nutrition: evidence from a 51. Jayachandran, S. & Pande, R. Why are Indian children so short? The mandated school meal program in India. J. Dev. Econ. 92, 152–165 (2010). role of birth order and son preference. Am. Econ. Rev. 107, 2600–2629 22. Singh, A., Park, A. & Dercon, S. School Meals as a Safety Net: An evaluation of (2017). the Midday Meal Scheme in India. Econ. Dev. Cult. Change 62, 275–306 52. Özaltin, E., Hill, K. & Subramanian, S. V. Association of maternal stature with (2014). offspring mortality, underweight, and stunting in low- to middle-income 23. Nabwera, H. M., Fulford, A. J., Moore, S. E. & Prentice, A. M. Growth faltering countries. JAMA 303, 1507–1516 (2010). in rural Gambian children after four decades of interventions: a retrospective 53. Desai, S., Vanneman, R. & National Council of Applied Economic Research. cohort study. Lancet Glob. Health 5, e208–e216 (2017). India Human Development Survey (IHDS), 2005 (ICPSR 22626). Inter- 24. Addo, O. Y. et al. Maternal height and child growth patterns. J. Pediatr. 163, University Consortium for Political and Social Research. https://doi.org/ 549–554. e1 (2013). 10.3886/ICPSR22626.v12 (2018). 25. Hambidge, M., Mazariegos, M., Kindem, M., Wright, L. & Cristobal-Pereza, C. 54. Desai, S., Vanneman, R. & National Council of Applied Economic Research. Infant stunting is associated with short maternal stature. J. Pediatr. India Human Development Survey-II (IHDS-II), 2011-12 (ICPSR 36151). Gastroenterol. Nutr. 54, 117–119 (2012). Inter-University Consortium for Political and Social Research. https://doi.org/ 26. Stein, A. D. et al. Comparison of linear growth patterns in the ﬁrst three 10.3886/ICPSR36151.v6 (2018). years of life across two generations in Guatemala. Pediatrics 113, e270–e275 55. Deaton, A. Panel data from time series of cross-sections. J. Econ. https://doi. (2004). org/10.1016/0304-4076(85)90134-4 (1985). 27. Bundy, D. A., Drake, L. J. & Burbano, C. School food, politics and child health. 56. Cotlear, D. Going Universal. World Bank https://openknowledge.worldbank. Public Health Nutr. 16, 1012–1019 (2013). org/bitstream/handle/10986/22011/9781464806100.pdf (2015). 28. Ruel, M. T. & Alderman, H. Nutrition-sensitive interventions and 57. D’Silva, J. Can India pull off its ambitious National Health Mission? BMJ programmes: how can they help to accelerate progress in improving maternal https://doi.org/10.1136/bmj.f2134 (2013). and child nutrition? Lancet 382, 536–551 (2013). 58. Bosworth, B. & Collins, S. M. Accounting for growth: comparing China and 29. García, S. & Saavedra, J. E. Educational impacts and cost-effectiveness of India. J. Econ. Perspect. https://doi.org/10.1257/jep.22.1.45 (2008). conditional cash transfer programs in developing countries: a meta-analysis. 59. Chakrabarti, S., Raghunathan, K., Alderman, H., Menon, P. & Nguyen, P. Rev. Educ. Res. 87, 921–965 (2017). India’s integrated child development services programme; equity and extent of 30. Leroy, J. L., Ruel, M. & Verhofstadt, E. The impact of conditional cash transfer coverage in 2006 and 2016. Bull. World Health Organ. https://doi.org/10.2471/ programmes on child nutrition: a review of evidence using a programme BLT.18.221135 (2019). theory framework. J. Dev. Effect. 1, 103–129 (2009). 60. Chakrabarti, S., Kishore, A. & Roy, D. Effectiveness of food subsidies in raising 31. Hoynes, H., Schanzenbach, D. W. & Almond, D. Long-run impacts of healthy food consumption: public distribution of pulses in India. Am. J. Agric. childhood access to the Safety Net. Am. Econ. Rev. 106, 903–934 (2016). Econ. https://doi.org/10.1093/ajae/aay022 (2018). 32. Cohodes, S. R., Grossman, D. S., Kleiner, S. A. & Lovenheim, M. F. The effect 61. Kim, R., Mohanty, S. K. & Subramanian, S. V. Multilevel geographies of of child health insurance access on schooling: evidence from public insurance poverty in India. World Dev. 87, 349–359 (2016). expansions. J. Hum. Resour. 51, 727–759 (2016). 62. Muralidharan, K. & Kremer, M. Public-private schools in rural India. School 33. Almond, D., Currie, J. & Duque, V. Childhood circumstances and adult Choice International. https://doi.org/10.7551/mitpress/9780262033763.003.0005 outcomes: Act II. J. Econ. Lit. 56, 1360–1446 (2018). (2013). 34. Aizer, A., Eli, S., Ferrie, J. & Lleras-Muney, A. The long-run impact of cash 63. Bernal, J. L., Cummins, S. & Gasparrini, A. The use of controls in interrupted transfers to poor families. Am. Econ. Rev. 106, 935–971 (2016). time series studies of public health interventions. Int. J. Epidemiol. https://doi. 35. Wherry, L. R. & Meyer, B. D. Saving teens: using a policy discontinuity to org/10.1093/ije/dyy135 (2018). estimate the effects of medicaid eligibility. J. Hum. Resour. 51, 556–588 (2016). 64. Shadish, W., Cook, T. & Campbell, D. Quasi-experimental designs that use 36. Cheng, T. L., Johnson, S. B. & Goodman, E. Breaking the intergenerational both control groups and pretests. Experimental and Quasi-Experimental cycle of disadvantage: the three generation approach. Pediatrics https://doi. Designs (2002). org/10.1542/peds.2015-2467 (2016). 65. Bernal, J. L., Cummins, S. & Gasparrini, A. Interrupted time series regression 37. Drèze, J. & Khera, R. Recent social security initiatives in India. World Dev. 98, for the evaluation of public health interventions: a tutorial. Int. J. Epidemiol. 555–572 (2017). https://doi.org/10.1093/ije/dyw098 (2017). 38. Barrett, C. B. & Carter, M. R. The power and pitfalls of experiments in 66. Headey, D., Hoddinott, J. & Park, S. Accounting for nutritional changes in six development economics: some non-random reﬂections. Appl. Econ. Perspect. success stories: a regression-decomposition approach. Glob. Food Security 13, Policy 32, 515–548 (2010). 12–20 (2017). 39. Verbeek, M. & Vella, F. Estimating dynamic models from repeated cross- 67. International Food Policy Research Institute (IFPRI) & University of sections. J. Econ. https://doi.org/10.1016/j.jeconom.2004.06.004 (2005). Washington. Intergenerational Nutrition Beneﬁts of India’s National School 40. Kone, Z. L., Liu, M. Y., Mattoo, A., Ozden, C. & Sharma, S. Internal borders Feeding Program. https://doi.org/10.7910/DVN/JTN87W (2021). and migration in India. J. Econ. Geogr. https://doi.org/10.1093/jeg/lbx045 (2018). 41. Chen, Q., Galfalvy, H. & Duan, N. Effects of disease misclassiﬁcation on Acknowledgements exposure-disease association. Am. J. Public Health. https://doi.org/10.2105/ We acknowledge feedback from the participants of the following conferences where AJPH.2012.300995 (2013). drafts of the paper were presented: Nutrition 2018 (organized by the American Society 42. UNICEF. UNICEF Data Base on Primary Education (2018). for Nutrition), North East Universities Development Consortium (NEUDC) 2018, and 43. Alderman, H., Behrman, J. & Tasneem, A. The contribution of increased the National Institute of Nutrition (NIN) 2018 Centenary conference. Bill & Melinda equity to the estimated social beneﬁts from a transfer program: an illustration Gates Foundation through Partnerships and Opportunities to Strengthen and Harmonize from PROGRESA/Oportunidades. World Bank Econ. Rev. https://doi.org/ Actions Against Malnutrition in India (POSHAN), led by the International Food Policy 10.1093/wber/lhx006 (2019). Research Institute (IFPRI). 44. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, July 1993–June 1994, NSS 50th Round. National Data Archive. DDI-IND- MOSPI-NSSO-50Rnd-Sch1.0-1993-94 (2019). 45. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, Author contributions July 1999–June 2000, NSS 55th Round. National Data Archive. DDI-IND- S.C. conceived the idea for the research. S.C. and S.P.S. contributed to all other aspects MOSPI-NSSO-55Rnd-Sch1-July1999-June2000 (2019). including data analysis, interpretation, and manuscript preparation and writing. H.A., 46. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, P.M., and D.O.G. provided inputs to analysis and contributed to manuscript writing. All July 2004–June 2005, NSS 61st Round. National Data Archive. DDI-IND- co-authors read and approved the ﬁnal version of the manuscript. MOSPI-NSSO-61Rnd-Sch1-July2004-June2005 (2019). 47. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, Type 1: July 2011–June 2012, NSS 68th Round. National Data Archive. DDI- IND-MOSPI-NSSO-68Rnd-Sch1.0-July2011-June2012 (2019). Competing interests 48. DHS Program. India: Standard DHS, 2015–2016 (2016). The authors declare no competing interests. NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w Additional information Open Access This article is licensed under a Creative Commons Supplementary information The online version contains supplementary material Attribution 4.0 International License, which permits use, sharing, available at https://doi.org/10.1038/s41467-021-24433-w. adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Correspondence and requests for materials should be addressed to S.P.S. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless Peer review information Nature Communications thanks the anonymous reviewer(s) for indicated otherwise in a credit line to the material. If material is not included in the their contribution to the peer review of this work. Peer reviewer reports are available. article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from Reprints and permission information is available at http://www.nature.com/reprints the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afﬁliations. © The Author(s) 2021 10 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Nature Communications Springer Journals http://www.deepdyve.com/lp/springer-journals/intergenerational-nutrition-benefits-of-india-s-national-school-K8DuLe0FYR

Loading next page...

References (92)

(Cheng, T. L., Johnson, S. B. & Goodman, E. Breaking the intergenerational cycle of disadvantage: the three generation approach. Pediatrics10.1542/peds.2015-2467 (2016).)
Cheng, T. L., Johnson, S. B. & Goodman, E. Breaking the intergenerational cycle of disadvantage: the three generation approach. Pediatrics10.1542/peds.2015-2467 (2016).
Cheng, T. L., Johnson, S. B. & Goodman, E. Breaking the intergenerational cycle of disadvantage: the three generation approach. Pediatrics10.1542/peds.2015-2467 (2016)., Cheng, T. L., Johnson, S. B. & Goodman, E. Breaking the intergenerational cycle of disadvantage: the three generation approach. Pediatrics10.1542/peds.2015-2467 (2016).
(Muralidharan, K. & Kremer, M. Public-private schools in rural India. School Choice International. 10.7551/mitpress/9780262033763.003.0005 (2013).)
Muralidharan, K. & Kremer, M. Public-private schools in rural India. School Choice International. 10.7551/mitpress/9780262033763.003.0005 (2013).
Muralidharan, K. & Kremer, M. Public-private schools in rural India. School Choice International. 10.7551/mitpress/9780262033763.003.0005 (2013)., Muralidharan, K. & Kremer, M. Public-private schools in rural India. School Choice International. 10.7551/mitpress/9780262033763.003.0005 (2013).
K. Hambidge, M. Mazariegos, Mark Kindem, L. Wright, Christina Cristobal-Perez, Lucrecia Juárez-García, J. Westcott, N. Goco, N. Krebs (2012)
Infant Stunting Is Associated With Short Maternal Stature
Journal of Pediatric Gastroenterology and Nutrition, 54
Karthink Muralidharan, M. Kremer (2009)
Public-Private Schools in Rural India
(Bernal, J. L., Cummins, S. & Gasparrini, A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int. J. Epidemiol. 10.1093/ije/dyw098 (2017).)
Bernal, J. L., Cummins, S. & Gasparrini, A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int. J. Epidemiol. 10.1093/ije/dyw098 (2017).
Bernal, J. L., Cummins, S. & Gasparrini, A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int. J. Epidemiol. 10.1093/ije/dyw098 (2017)., Bernal, J. L., Cummins, S. & Gasparrini, A. Interrupted time series regression for the evaluation of public health interventions: a tutorial. Int. J. Epidemiol. 10.1093/ije/dyw098 (2017).
H. Alderman, D. Headey (2017)
How Important is Parental Education for Child Nutrition?
World Development, 94
Jef Leroy, M. Ruel, Ellen Verhofstadt (2009)
The impact of conditional cash transfer programmes on child nutrition: a review of evidence using a programme theory framework
Journal of Development Effectiveness, 1
(Deaton, A. Panel data from time series of cross-sections. J. Econ. 10.1016/0304-4076(85)90134-4 (1985).)
Deaton, A. Panel data from time series of cross-sections. J. Econ. 10.1016/0304-4076(85)90134-4 (1985).
Deaton, A. Panel data from time series of cross-sections. J. Econ. 10.1016/0304-4076(85)90134-4 (1985)., Deaton, A. Panel data from time series of cross-sections. J. Econ. 10.1016/0304-4076(85)90134-4 (1985).
J. Drèze, R. Khera (2016)
Recent Social Security Initiatives in India
ERPN: Social Policy (Sub-Topic)
Marno Verbeek, F. Vella (2002)
Estimating Dynamic Models from Repeated Cross-Sections
Econometrics eJournal
Angus Deaton (1985)
Panel data from time series of cross-sections
Journal of Econometrics, 30
B. Bosworth, Susan Collins (2007)
NBER WORKING PAPER SERIES ACCOUNTING FOR GROWTH: COMPARING CHINA AND INDIA
D. Bundy, N. Silva, S. Horton, G. Patton, Linda Schultz, D. Jamison, Amina Abubakara, A. Ahuja, H. Alderman, Nicolas Allen, Laura Appleby, E. Aurino, P. Azzopardi, Sarah Baird, L. Banham, Jere Behrman, H. Benzian, S. Bhalotra, Z. Bhutta, M. Black, P. Bloem, C. Bonell, M. Bradley, S. Brinkman, S. Brooker, C. Burbano, Nicolas Burnett, Tania Cernuschi, S. Clarke, C. Coffey, Peter Colenso, K. Croke, Amy Daniels, Elia Cruz, D. Walque, Anil Deolaikar, L. Drake, L. Fernald, Meena Fernandes, Deepika Fernando, G. Fink, R. Galloway, A. Gelli, A. Georgiadis, Caroline Gitonga, Boitshepo Giyosa, P. Glewwe, J. Nzovu, Amber Gove, N. Graham, B. Greenwood, E. Grigorenko, Caitlin Heath, Joan Hamory, Melissa Hidrobo, K. Hill, Tara Hill, T. Hollingsworth, Elissa Kennedy, Imran Khan, Josephine Kiamba, Jane Kim, M. Kremer, D. Lamontagne, Z. Lassi, R. Laxminarayan, J. Mahon, Lu Mai, Sebastian Martinez, S. Meresman, K. Merseth, E. Miguel, Arlene Mitchell, Sophie Mitra, Anoosh Moin, A. Mokdad, D. Mont, Arindam Nandi, J. Nankabirwa, Daniel Plaut, Elina Pradhan, R. Pullan, N. Reavley, J. Santelli, B. Sarr, S. Sawyer (2017)
Investment in child and adolescent health and development: key messages from Disease Control Priorities, 3rd Edition
The Lancet, 391
(Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey-II (IHDS-II), 2011-12 (ICPSR 36151). Inter-University Consortium for Political and Social Research.10.3886/ICPSR36151.v6 (2018).)
Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey-II (IHDS-II), 2011-12 (ICPSR 36151). Inter-University Consortium for Political and Social Research.10.3886/ICPSR36151.v6 (2018).
Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey-II (IHDS-II), 2011-12 (ICPSR 36151). Inter-University Consortium for Political and Social Research.10.3886/ICPSR36151.v6 (2018)., Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey-II (IHDS-II), 2011-12 (ICPSR 36151). Inter-University Consortium for Political and Social Research.10.3886/ICPSR36151.v6 (2018).
J. Drèze, G. Kingdon (1999)
School Participation in Rural India
Labor: Human Capital
(International Food Policy Research Institute (IFPRI) & University of Washington. Intergenerational Nutrition Benefits of India’s National School Feeding Program. 10.7910/DVN/JTN87W (2021).)
International Food Policy Research Institute (IFPRI) & University of Washington. Intergenerational Nutrition Benefits of India’s National School Feeding Program. 10.7910/DVN/JTN87W (2021).
International Food Policy Research Institute (IFPRI) & University of Washington. Intergenerational Nutrition Benefits of India’s National School Feeding Program. 10.7910/DVN/JTN87W (2021)., International Food Policy Research Institute (IFPRI) & University of Washington. Intergenerational Nutrition Benefits of India’s National School Feeding Program. 10.7910/DVN/JTN87W (2021).
H. Nabwera, A. Fulford, S. Moore, A. Prentice (2017)
Growth faltering in rural Gambian children after four decades of interventions: a retrospective cohort study
The Lancet. Global Health, 5
Matthias Rieger, Sofia Trommlerová (2016)
Age-Specific Correlates of Child Growth
Demography, 53
E. Cavatorta, B. Shankar, Artemisa Flores-Martinez (2015)
Explaining Cross-State Disparities in Child Nutrition in Rural India
World Development, 76
(Chakrabarti, S., Kishore, A. & Roy, D. Effectiveness of food subsidies in raising healthy food consumption: public distribution of pulses in India. Am. J. Agric. Econ. 10.1093/ajae/aay022 (2018).)
Chakrabarti, S., Kishore, A. & Roy, D. Effectiveness of food subsidies in raising healthy food consumption: public distribution of pulses in India. Am. J. Agric. Econ. 10.1093/ajae/aay022 (2018).
Chakrabarti, S., Kishore, A. & Roy, D. Effectiveness of food subsidies in raising healthy food consumption: public distribution of pulses in India. Am. J. Agric. Econ. 10.1093/ajae/aay022 (2018)., Chakrabarti, S., Kishore, A. & Roy, D. Effectiveness of food subsidies in raising healthy food consumption: public distribution of pulses in India. Am. J. Agric. Econ. 10.1093/ajae/aay022 (2018).
(2017)
Deploying the power of social protection to improve nutrition what will it take
(DHS Program. India: Standard DHS, 2015–2016 (2016).)
DHS Program. India: Standard DHS, 2015–2016 (2016).
DHS Program. India: Standard DHS, 2015–2016 (2016)., DHS Program. India: Standard DHS, 2015–2016 (2016).
(Bundy, D. A. P. et al. Investment in child and adolescent health and development: key messages from Disease Control Priorities, 3rd Edition. Lancet10.1016/S0140-6736(17)32417-0 (2017).)
Bundy, D. A. P. et al. Investment in child and adolescent health and development: key messages from Disease Control Priorities, 3rd Edition. Lancet10.1016/S0140-6736(17)32417-0 (2017).
Bundy, D. A. P. et al. Investment in child and adolescent health and development: key messages from Disease Control Priorities, 3rd Edition. Lancet10.1016/S0140-6736(17)32417-0 (2017)., Bundy, D. A. P. et al. Investment in child and adolescent health and development: key messages from Disease Control Priorities, 3rd Edition. Lancet10.1016/S0140-6736(17)32417-0 (2017).
H. Hoynes, D. Schanzenbach, D. Almond (2012)
Long Run Impacts of Childhood Access to the Safety Net
NBER Working Paper Series
(Ministry of Health and Family Welfare Government of India. National Family Health Survey-4, 2015-2016, India Fact Sheet. http://rchiips.org/nfhs/pdf/NFHS4/India.pdf (2017).)
Ministry of Health and Family Welfare Government of India. National Family Health Survey-4, 2015-2016, India Fact Sheet. http://rchiips.org/nfhs/pdf/NFHS4/India.pdf (2017).
Ministry of Health and Family Welfare Government of India. National Family Health Survey-4, 2015-2016, India Fact Sheet. http://rchiips.org/nfhs/pdf/NFHS4/India.pdf (2017)., Ministry of Health and Family Welfare Government of India. National Family Health Survey-4, 2015-2016, India Fact Sheet. http://rchiips.org/nfhs/pdf/NFHS4/India.pdf (2017).
(Ministry of Human Resource Development. Mid Day Meal Scheme. Department of School Education & Literacy, Government of India. http://mdm.nic.in/ (2017).)
Ministry of Human Resource Development. Mid Day Meal Scheme. Department of School Education & Literacy, Government of India. http://mdm.nic.in/ (2017).
Ministry of Human Resource Development. Mid Day Meal Scheme. Department of School Education & Literacy, Government of India. http://mdm.nic.in/ (2017)., Ministry of Human Resource Development. Mid Day Meal Scheme. Department of School Education & Literacy, Government of India. http://mdm.nic.in/ (2017).
Suman Chakrabarti, A. Kishore, D. Roy (2018)
Effectiveness of Food Subsidies in Raising Healthy Food Consumption: Public Distribution of Pulses in India
American Journal of Agricultural Economics, 100
Sarah Cohodes, D. Grossman, S. Kleiner, Michael Lovenheim (2014)
The Effect of Child Health Insurance Access on Schooling: Evidence from Public Insurance Expansions
The Journal of Human Resources, 51
H. Sachdev, Clive Osmond, A. Kurpad, T. Thomas (2022)
Intergenerational nutrition benefits of India’s national school feeding program: Reality or a bridge too far?
Nature Communications, 13
(National Sample Survey Office. India—Household Consumer Expenditure, July 1993–June 1994, NSS 50th Round. National Data Archive. DDI-IND-MOSPI-NSSO-50Rnd-Sch1.0-1993-94 (2019).)
National Sample Survey Office. India—Household Consumer Expenditure, July 1993–June 1994, NSS 50th Round. National Data Archive. DDI-IND-MOSPI-NSSO-50Rnd-Sch1.0-1993-94 (2019).
National Sample Survey Office. India—Household Consumer Expenditure, July 1993–June 1994, NSS 50th Round. National Data Archive. DDI-IND-MOSPI-NSSO-50Rnd-Sch1.0-1993-94 (2019)., National Sample Survey Office. India—Household Consumer Expenditure, July 1993–June 1994, NSS 50th Round. National Data Archive. DDI-IND-MOSPI-NSSO-50Rnd-Sch1.0-1993-94 (2019).
(Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey (IHDS), 2005 (ICPSR 22626). Inter-University Consortium for Political and Social Research. 10.3886/ICPSR22626.v12 (2018).)
Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey (IHDS), 2005 (ICPSR 22626). Inter-University Consortium for Political and Social Research. 10.3886/ICPSR22626.v12 (2018).
Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey (IHDS), 2005 (ICPSR 22626). Inter-University Consortium for Political and Social Research. 10.3886/ICPSR22626.v12 (2018)., Desai, S., Vanneman, R. & National Council of Applied Economic Research. India Human Development Survey (IHDS), 2005 (ICPSR 22626). Inter-University Consortium for Political and Social Research. 10.3886/ICPSR22626.v12 (2018).
(Drèze, J. & Kingdon, G. School participation in rural India. London School of Economics and Political Science (1999).)
Drèze, J. & Kingdon, G. School participation in rural India. London School of Economics and Political Science (1999).
Drèze, J. & Kingdon, G. School participation in rural India. London School of Economics and Political Science (1999)., Drèze, J. & Kingdon, G. School participation in rural India. London School of Economics and Political Science (1999).
Sandra García, J. Saavedra (2013)
Educational Impacts and Cost-Effectiveness of Conditional Cash Transfer Programs in Developing Countries: A Meta-Analysis
Review of Educational Research, 87
A. Panagariya (2014)
NATIONAL COUNCIL OF APPLIED ECONOMIC RESEARCH
(National Sample Survey Office. India—Household Consumer Expenditure, Type 1: July 2011–June 2012, NSS 68th Round. National Data Archive. DDI-IND-MOSPI-NSSO-68Rnd-Sch1.0-July2011-June2012 (2019).)
National Sample Survey Office. India—Household Consumer Expenditure, Type 1: July 2011–June 2012, NSS 68th Round. National Data Archive. DDI-IND-MOSPI-NSSO-68Rnd-Sch1.0-July2011-June2012 (2019).
National Sample Survey Office. India—Household Consumer Expenditure, Type 1: July 2011–June 2012, NSS 68th Round. National Data Archive. DDI-IND-MOSPI-NSSO-68Rnd-Sch1.0-July2011-June2012 (2019)., National Sample Survey Office. India—Household Consumer Expenditure, Type 1: July 2011–June 2012, NSS 68th Round. National Data Archive. DDI-IND-MOSPI-NSSO-68Rnd-Sch1.0-July2011-June2012 (2019).
Tanika Chakraborty, Rajshri Jayaraman (2018)
Forthcoming: Journal of Development Economics SCHOOL FEEDING AND LEARNING ACHIEVEMENT: EVIDENCE FROM INDIA’S MIDDAY MEAL PROGRAM
(Chen, Q., Galfalvy, H. & Duan, N. Effects of disease misclassification on exposure-disease association. Am. J. Public Health. 10.2105/AJPH.2012.300995 (2013).)
Chen, Q., Galfalvy, H. & Duan, N. Effects of disease misclassification on exposure-disease association. Am. J. Public Health. 10.2105/AJPH.2012.300995 (2013).
Chen, Q., Galfalvy, H. & Duan, N. Effects of disease misclassification on exposure-disease association. Am. J. Public Health. 10.2105/AJPH.2012.300995 (2013)., Chen, Q., Galfalvy, H. & Duan, N. Effects of disease misclassification on exposure-disease association. Am. J. Public Health. 10.2105/AJPH.2012.300995 (2013).
(Chakrabarti, S., Raghunathan, K., Alderman, H., Menon, P. & Nguyen, P. India’s integrated child development services programme; equity and extent of coverage in 2006 and 2016. Bull. World Health Organ. 10.2471/BLT.18.221135 (2019).)
Chakrabarti, S., Raghunathan, K., Alderman, H., Menon, P. & Nguyen, P. India’s integrated child development services programme; equity and extent of coverage in 2006 and 2016. Bull. World Health Organ. 10.2471/BLT.18.221135 (2019).
Chakrabarti, S., Raghunathan, K., Alderman, H., Menon, P. & Nguyen, P. India’s integrated child development services programme; equity and extent of coverage in 2006 and 2016. Bull. World Health Organ. 10.2471/BLT.18.221135 (2019)., Chakrabarti, S., Raghunathan, K., Alderman, H., Menon, P. & Nguyen, P. India’s integrated child development services programme; equity and extent of coverage in 2006 and 2016. Bull. World Health Organ. 10.2471/BLT.18.221135 (2019).
(Verbeek, M. & Vella, F. Estimating dynamic models from repeated cross-sections. J. Econ. 10.1016/j.jeconom.2004.06.004 (2005).)
Verbeek, M. & Vella, F. Estimating dynamic models from repeated cross-sections. J. Econ. 10.1016/j.jeconom.2004.06.004 (2005).
Verbeek, M. & Vella, F. Estimating dynamic models from repeated cross-sections. J. Econ. 10.1016/j.jeconom.2004.06.004 (2005)., Verbeek, M. & Vella, F. Estimating dynamic models from repeated cross-sections. J. Econ. 10.1016/j.jeconom.2004.06.004 (2005).
Farzana Afridi (2010)
Child welfare programs and child nutrition: Evidence from a mandated school meal program in India
Journal of Development Economics, 92
Laura Wherry, Bruce Meyer (2012)
Saving Teens: Using a Policy Discontinuity to Estimate the Effects of Medicaid Eligibility
The Journal of Human Resources, 51
A. Aizer, Shari Eli, J. Ferrie, A. Lleras-Muney (2016)
The Long-Run Impact of Cash Transfers to Poor Families.
The American economic review, 106 4
H. Alderman, Jere Behrman (2015)
The Contribution of Increased Equity to the Estimated Social Benefits from a Transfer Program: An Illustration from Progresa/Oportunidades
World Bank Policy Research Working Paper Series
(Chakraborty, T. & Jayaraman, R. School feeding and learning achievement: evidence from India’s Midday Meal Program. IZA Discussion Paper 10086 (2016).)
Chakraborty, T. & Jayaraman, R. School feeding and learning achievement: evidence from India’s Midday Meal Program. IZA Discussion Paper 10086 (2016).
Chakraborty, T. & Jayaraman, R. School feeding and learning achievement: evidence from India’s Midday Meal Program. IZA Discussion Paper 10086 (2016)., Chakraborty, T. & Jayaraman, R. School feeding and learning achievement: evidence from India’s Midday Meal Program. IZA Discussion Paper 10086 (2016).
Suman Chakrabarti, K. Raghunathan, H. Alderman, P. Menon, P. Nguyen (2019)
India’s Integrated Child Development Services programme; equity and extent of coverage in 2006 and 2016
Bulletin of the World Health Organization, 97
J. Drèze, R. Khera (2017)
Development ReviewRecent Social Security Initiatives in India
World Development, 98
Yuyu Chen, Hongbin Li (2009)
Mother's education and child health: is there a nurturing effect?
Journal of health economics, 28 2
Farzana Afridi (2011)
The Impact of School Meals on School Participation: Evidence from Rural India
The Journal of Development Studies, 47
(Raghunathan, K., Chakrabarti, S., Menon, P. & Alderman, H. Deploying the power of social protection to improve nutrition what will it take? Econ. Polit. Wkly52, (2017).)
Raghunathan, K., Chakrabarti, S., Menon, P. & Alderman, H. Deploying the power of social protection to improve nutrition what will it take? Econ. Polit. Wkly52, (2017).
Raghunathan, K., Chakrabarti, S., Menon, P. & Alderman, H. Deploying the power of social protection to improve nutrition what will it take? Econ. Polit. Wkly52, (2017)., Raghunathan, K., Chakrabarti, S., Menon, P. & Alderman, H. Deploying the power of social protection to improve nutrition what will it take? Econ. Polit. Wkly52, (2017).
(2005)
ICPSR 22626)
S. Jayachandran, R. Pandi (2017)
Why Are Indian Children So Short? The Role of Birth Order and Son Preference.
The American economic review, 107 9
J. Bernal, S. Cummins, A. Gasparrini (2016)
Interrupted time series regression for the evaluation of public health interventions: a tutorial
International Journal of Epidemiology, 46
M. Black, S. Walker, L. Fernald, C. Andersen, A. DiGirolamo, Chunling Lu, D. McCoy, G. Fink, Y. Shawar, J. Shiffman, A. Devercelli, Q. Wodon, E. Vargas-Barón, S. Grantham‐McGregor (2017)
Early childhood development coming of age: science through the life course
The Lancet, 389
R. Black, C. Victora, S. Walker, Z. Bhutta, P. Christian, M. Onis, M. Ezzati, S. Grantham‐McGregor, J. Katz, R. Martorell, R. Uauy (2013)
Maternal and child undernutrition and overweight in low-income and middle-income countries
The Lancet, 382
(UNICEF. UNICEF Data Base on Primary Education (2018).)
UNICEF. UNICEF Data Base on Primary Education (2018).
UNICEF. UNICEF Data Base on Primary Education (2018)., UNICEF. UNICEF Data Base on Primary Education (2018).
A. Prendergast, J. Humphrey (2014)
The stunting syndrome in developing countries
Paediatrics and International Child Health, 34
Z. Bhutta, Jai Das, A. Rizvi, M. Gaffey, N. Walker, S. Horton, P. Webb, A. Lartey, R. Black (2013)
Evidence-based interventions for improvement of maternal and child nutrition: what can be done and at what cost?
The Lancet, 382
(2015)
Going Universal
(Development Initiatives. Global Nutrition Report 2017: Nourishing the SDGs (2017).)
Development Initiatives. Global Nutrition Report 2017: Nourishing the SDGs (2017).
Development Initiatives. Global Nutrition Report 2017: Nourishing the SDGs (2017)., Development Initiatives. Global Nutrition Report 2017: Nourishing the SDGs (2017).
(National Sample Survey Office. India—Household Consumer Expenditure, July 2004–June 2005, NSS 61st Round. National Data Archive. DDI-IND-MOSPI-NSSO-61Rnd-Sch1-July2004-June2005 (2019).)
National Sample Survey Office. India—Household Consumer Expenditure, July 2004–June 2005, NSS 61st Round. National Data Archive. DDI-IND-MOSPI-NSSO-61Rnd-Sch1-July2004-June2005 (2019).
National Sample Survey Office. India—Household Consumer Expenditure, July 2004–June 2005, NSS 61st Round. National Data Archive. DDI-IND-MOSPI-NSSO-61Rnd-Sch1-July2004-June2005 (2019)., National Sample Survey Office. India—Household Consumer Expenditure, July 2004–June 2005, NSS 61st Round. National Data Archive. DDI-IND-MOSPI-NSSO-61Rnd-Sch1-July2004-June2005 (2019).
M. Ruel, H. Alderman (2013)
Nutrition-sensitive interventions and programmes: how can they help to accelerate progress in improving maternal and child nutrition?
The Lancet, 382
Abhijeet Singh, A. Park, S. Dercon (2012)
School Meals as a Safety Net: An Evaluation of the Midday Meal Scheme in India
Economic Development and Cultural Change, 62
Z. Kone, Maggie Liu, A. Mattoo, C. Ozden, Siddharth Sharma, World Group
Smith ScholarWorks Smith ScholarWorks
(National Sample Survey Office. India—Household Consumer Expenditure, July 1999–June 2000, NSS 55th Round. National Data Archive. DDI-IND-MOSPI-NSSO-55Rnd-Sch1-July1999-June2000 (2019).)
National Sample Survey Office. India—Household Consumer Expenditure, July 1999–June 2000, NSS 55th Round. National Data Archive. DDI-IND-MOSPI-NSSO-55Rnd-Sch1-July1999-June2000 (2019).
National Sample Survey Office. India—Household Consumer Expenditure, July 1999–June 2000, NSS 55th Round. National Data Archive. DDI-IND-MOSPI-NSSO-55Rnd-Sch1-July1999-June2000 (2019)., National Sample Survey Office. India—Household Consumer Expenditure, July 1999–June 2000, NSS 55th Round. National Data Archive. DDI-IND-MOSPI-NSSO-55Rnd-Sch1-July1999-June2000 (2019).
(Kone, Z. L., Liu, M. Y., Mattoo, A., Ozden, C. & Sharma, S. Internal borders and migration in India. J. Econ. Geogr. 10.1093/jeg/lbx045 (2018).)
Kone, Z. L., Liu, M. Y., Mattoo, A., Ozden, C. & Sharma, S. Internal borders and migration in India. J. Econ. Geogr. 10.1093/jeg/lbx045 (2018).
Kone, Z. L., Liu, M. Y., Mattoo, A., Ozden, C. & Sharma, S. Internal borders and migration in India. J. Econ. Geogr. 10.1093/jeg/lbx045 (2018)., Kone, Z. L., Liu, M. Y., Mattoo, A., Ozden, C. & Sharma, S. Internal borders and migration in India. J. Econ. Geogr. 10.1093/jeg/lbx045 (2018).
C. Barrett, M. Carter (2010)
The Power and Pitfalls of Experiments in Development Economics: Some Non‐random Reflections
Applied Economic Perspectives and Policy, 32
(2017)
Mid Day Meal Scheme. Department of School Education & Literacy, Government of India
(Bosworth, B. & Collins, S. M. Accounting for growth: comparing China and India. J. Econ. Perspect. 10.1257/jep.22.1.45 (2008).)
Bosworth, B. & Collins, S. M. Accounting for growth: comparing China and India. J. Econ. Perspect. 10.1257/jep.22.1.45 (2008).
Bosworth, B. & Collins, S. M. Accounting for growth: comparing China and India. J. Econ. Perspect. 10.1257/jep.22.1.45 (2008)., Bosworth, B. & Collins, S. M. Accounting for growth: comparing China and India. J. Econ. Perspect. 10.1257/jep.22.1.45 (2008).
(Bernal, J. L., Cummins, S. & Gasparrini, A. The use of controls in interrupted time series studies of public health interventions. Int. J. Epidemiol. 10.1093/ije/dyy135 (2018).)
Bernal, J. L., Cummins, S. & Gasparrini, A. The use of controls in interrupted time series studies of public health interventions. Int. J. Epidemiol. 10.1093/ije/dyy135 (2018).
Bernal, J. L., Cummins, S. & Gasparrini, A. The use of controls in interrupted time series studies of public health interventions. Int. J. Epidemiol. 10.1093/ije/dyy135 (2018)., Bernal, J. L., Cummins, S. & Gasparrini, A. The use of controls in interrupted time series studies of public health interventions. Int. J. Epidemiol. 10.1093/ije/dyy135 (2018).
M. Onís, F. Branca (2016)
Childhood stunting: a global perspective
Maternal & Child Nutrition, 12
D. Bundy, L. Drake, C. Burbano (2012)
School food, politics and child health
Public Health Nutrition, 16
D. Headey, J. Hoddinott, Seollee Park (2016)
Drivers of nutritional change in four South Asian countries: a dynamic observational analysis
Maternal & Child Nutrition, 12
A. Dhawan, R. Rao, A. Ambekar, Amal Pusp, R. Ray (2017)
Treatment of substance use disorders through the government health facilities: Developments in the “Drug De-addiction Programme” of Ministry of Health and Family Welfare, Government of India
Indian Journal of Psychiatry, 59
D. Almond, J. Currie, Valentina Duque (2017)
Childhood Circumstances and Adult Outcomes: Act Ii
CSN: Development (Topic)
(2002)
Quasi-experimental designs that use both control groups and pretests
(Cotlear, D. Going Universal. World Bankhttps://openknowledge.worldbank.org/bitstream/handle/10986/22011/9781464806100.pdf (2015).)
Cotlear, D. Going Universal. World Bankhttps://openknowledge.worldbank.org/bitstream/handle/10986/22011/9781464806100.pdf (2015).
Cotlear, D. Going Universal. World Bankhttps://openknowledge.worldbank.org/bitstream/handle/10986/22011/9781464806100.pdf (2015)., Cotlear, D. Going Universal. World Bankhttps://openknowledge.worldbank.org/bitstream/handle/10986/22011/9781464806100.pdf (2015).
M. Onís, K. Dewey, E. Borghi, A. Onyango, M. Blössner, B. Daelmans, E. Piwoz, F. Branca (2013)
The World Health Organization's global target for reducing childhood stunting by 2025: rationale and proposed actions.
Maternal & child nutrition, 9 Suppl 2
(D’Silva, J. Can India pull off its ambitious National Health Mission? BMJ10.1136/bmj.f2134 (2013).)
D’Silva, J. Can India pull off its ambitious National Health Mission? BMJ10.1136/bmj.f2134 (2013).
D’Silva, J. Can India pull off its ambitious National Health Mission? BMJ10.1136/bmj.f2134 (2013)., D’Silva, J. Can India pull off its ambitious National Health Mission? BMJ10.1136/bmj.f2134 (2013).
Rockli Kim, I. Mejía-Guevara, D. Corsi, V. Aguayo, S. Subramanian (2017)
Relative importance of 13 correlates of child stunting in South Asia: Insights from nationally representative data from Afghanistan, Bangladesh, India, Nepal, and Pakistan.
Social science & medicine, 187
O. Addo, Aryeh Stein, Caroline Fall, D. Gigante, A. Guntupalli, Bernardo Horta, C. Kuzawa, N. Lee, S. Norris, P. Prabhakaran, L. Richter, H. Sachdev, R. Martorell (2013)
Maternal Height and Child Growth Patterns
The Journal of Pediatrics, 163
D. Headey, J. Hoddinott, Seollee Park (2017)
Accounting for nutritional changes in six success stories: A regression-decomposition approach
Global Food Security, 13
(Alderman, H., Behrman, J. & Tasneem, A. The contribution of increased equity to the estimated social benefits from a transfer program: an illustration from PROGRESA/Oportunidades. World Bank Econ. Rev. 10.1093/wber/lhx006 (2019).)
Alderman, H., Behrman, J. & Tasneem, A. The contribution of increased equity to the estimated social benefits from a transfer program: an illustration from PROGRESA/Oportunidades. World Bank Econ. Rev. 10.1093/wber/lhx006 (2019).
Alderman, H., Behrman, J. & Tasneem, A. The contribution of increased equity to the estimated social benefits from a transfer program: an illustration from PROGRESA/Oportunidades. World Bank Econ. Rev. 10.1093/wber/lhx006 (2019)., Alderman, H., Behrman, J. & Tasneem, A. The contribution of increased equity to the estimated social benefits from a transfer program: an illustration from PROGRESA/Oportunidades. World Bank Econ. Rev. 10.1093/wber/lhx006 (2019).
A. Stein, H. Barnhart, Meng Wang, M. Hoshen, Karen Ologoudou, U. Ramakrishnan, R. Grajeda, M. Ramírez-Zea, R. Martorell (2004)
Comparison of linear growth patterns in the first three years of life across two generations in Guatemala.
Pediatrics, 113 3 Pt 1
Rockli Kim, S. Mohanty, S. Subramanian (2016)
Multilevel Geographies of Poverty in India
World Development, 87
Qixuan Chen, H. Galfalvy, N. Duan (2013)
Effects of disease misclassification on exposure-disease association.
American journal of public health, 103 5
T. Cheng, Sara Johnson, E. Goodman (2016)
Breaking the Intergenerational Cycle of Disadvantage: The Three Generation Approach
Pediatrics, 137
D. Corsi, I. Mejía-Guevara, S. Subramanian (2016)
Risk factors for chronic undernutrition among children in India: Estimating relative importance, population attributable risk and fractions.
Social science & medicine, 157
(Shadish, W., Cook, T. & Campbell, D. Quasi-experimental designs that use both control groups and pretests. Experimental and Quasi-Experimental Designs (2002).)
Shadish, W., Cook, T. & Campbell, D. Quasi-experimental designs that use both control groups and pretests. Experimental and Quasi-Experimental Designs (2002).
Shadish, W., Cook, T. & Campbell, D. Quasi-experimental designs that use both control groups and pretests. Experimental and Quasi-Experimental Designs (2002)., Shadish, W., Cook, T. & Campbell, D. Quasi-experimental designs that use both control groups and pretests. Experimental and Quasi-Experimental Designs (2002).
Emre Ozaltin, K. Hill, S. Subramanian (2010)
Association of maternal stature with offspring mortality, underweight, and stunting in low- to middle-income countries.
JAMA, 303 15
J. D’Silva (2013)
Can India pull off its ambitious National Health Mission?
BMJ : British Medical Journal, 346
J. Bernal, S. Cummins, A. Gasparrini (2018)
The use of controls in interrupted time series studies of public health interventions.
International journal of epidemiology, 47 6
(Development Initiatives. 2020 Global Nutrition Report—Global Nutrition Report. https://globalnutritionreport.org/reports/2020-global-nutrition-report/ (2020).)
Development Initiatives. 2020 Global Nutrition Report—Global Nutrition Report. https://globalnutritionreport.org/reports/2020-global-nutrition-report/ (2020).
Development Initiatives. 2020 Global Nutrition Report—Global Nutrition Report. https://globalnutritionreport.org/reports/2020-global-nutrition-report/ (2020)., Development Initiatives. 2020 Global Nutrition Report—Global Nutrition Report. https://globalnutritionreport.org/reports/2020-global-nutrition-report/ (2020).

Publisher: Springer Journals
Copyright: Copyright © The Author(s) 2021
eISSN: 2041-1723
DOI: 10.1038/s41467-021-24433-w
Publisher site: See Article on Publisher Site

Abstract

ARTICLE https://doi.org/10.1038/s41467-021-24433-w OPEN Intergenerational nutrition beneﬁts of India’s national school feeding program 1 1 1 1 1 Suman Chakrabarti , Samuel P. Scott , Harold Alderman , Purnima Menon & Daniel O. Gilligan India has the world’s highest number of undernourished children and the largest school feeding program, the Mid-Day Meal (MDM) scheme. As school feeding programs target children outside the highest-return “ﬁrst 1000-days” window, they have not been included in the global agenda to address stunting. School meals beneﬁt education and nutrition in par- ticipants, but no studies have examined whether beneﬁts carry over to their children. Using nationally representative data on mothers and their children spanning 1993 to 2016, we assess whether MDM supports intergenerational improvements in child linear growth. Here we report that height-for-age z-score (HAZ) among children born to mothers with full MDM exposure was greater (+0.40 SD) than that in children born to non-exposed mothers. Associations were stronger in low socioeconomic strata and likely work through women’s education, fertility, and health service utilization. MDM was associated with 13–32% of the HAZ improvement in India from 2006 to 2016. 1 ✉ Poverty Health and Nutrition Division, International Food Policy Research Institute, Washington, DC, USA. email: [email protected] NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 1 1234567890():,; ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w lobally, 149 million children are too short for their age disparities advocate for a multi-generation approach that and over half of these children live in Asia . Within India, addresses parental socioeconomic status (SES), child and ado- G38% of children were stunted in 2015–2016 (ref. ). Linear lescent health and development, and young adult’s capacity for growth failure is a marker of chronic undernutrition and multiple planning and future parenting . However, since interventions to pathological changes which, together, have been termed the improve maternal height and education must be implemented ‘stunting syndrome’ . Stunted children are at risk of not reaching years before those girls and young women become mothers, their developmental potential, thus stunting has large implica- empirical assessment of the effectiveness of such programs for tions for human capital and the economic productivity of entire reducing undernutrition among future offspring is challenging. 4–6 societies . The World Health Assembly set the ambitious target This paper studies the intergenerational nutrition beneﬁts of of reducing childhood stunting by 40% from 2010 to 2025 (ref. ), India’s MDM scheme. We use seven population level datasets a target that likely will not be met . Thus, it is imperative to spanning 1993 to 2016, including multiple rounds of National understand how countries can accelerate progress toward stunt- Sample Surveys of Consumer Expenditure (NSS-CES), National ing reduction. Family Household Surveys (NFHS), and India Human Develop- Though much focus has been placed on nutrition-speciﬁc ment Surveys (IHDS). We match cohorts of mothers by state, interventions during the 1000-day period from conception to the birth year, and SES with data on MDM coverage measured as the child’s second birthday, investments across multiple life periods proportion of primary-school-age girls receiving MDM using and which address underlying determinants are also important to data from the NSS-CES. Birth cohort ﬁxed effects and controlled 8,9 achieve stunting reductions . Interventions may work directly interrupted time series models are used to estimate the associa- through maternal–child biological pathways or indirectly through tion of mother’s exposure to the MDM scheme with the nutri- socioeconomic mechanisms. In India, women’s height and edu- tional status of her future children. We ﬁnd that maternal cohorts cational attainment are among the strongest predictors of child living in areas with higher coverage of the MDM scheme are less 10–15 stunting . likely to have stunted children than cohorts living in low coverage In the Indian context, a candidate intervention which poten- areas. This effect is robust to the inclusion of a broad set of tially improves both women’s height and education—and which, controls at multiple levels and ﬁxed effects. Controlled inter- therefore, may lead to reductions in stunting among children rupted time series models conﬁrm that the 14 states which rolled born to these women—is the national school feeding program, out MDM in the late-1990s experienced improvements in child the Mid-Day Meal (MDM) scheme . Launched in 1995 by the height earlier than the rest of the nation, which scaled up MDM Government of India, the MDM scheme provides a free cooked in the 2000s after the Supreme Court mandate. Plausibility is meal to children in government and government-assisted primary supported by our ﬁndings of MDM association with participants’ schools (classes I–V; ages 6–10 years). The mandated minimum education, age at birth, number of children, use of antenatal care, meal energy content is 450 kcal and the meal must contain 12 g of and delivery in a medical facility. protein. In 2016–2017, 97.8 million children received a free cooked meal through the scheme every day, making the MDM Results scheme the largest school feeding program in the world . Program description and motivation. The MDM scheme, Econometric evaluations of India’s MDM scheme have shown a initiated by the central government in 1995, was intended to 18,19 positive association with beneﬁciaries’ school attendance , cover all government schools under the National Programme of 20 21 learning achievement , hunger and protein-energy malnutrition , Nutritional Support for Primary Education . Due to institu- and resilience to health shocks such as drought —all of which tional challenges, only a few states scaled up the program may have carryover beneﬁts to children born to mothers who immediately. NSS-CES data from 1999 show that only 6% of all participated in the program. We are not aware of studies that have girls aged 6–10 years received mid-day meals in school (Fig. 1). explored whether program beneﬁts for the MDM or similar pro- Between 1999 and 2004, program coverage increased in many grams in other countries extend to the next generation. Filling this states, largely due to an order from the Supreme Court of India research gap is critical, as (1) stunting carries over from one gen- directing state governments to provide cooked mid-day meals eration to the next and is therefore optimally studied on a multi- in primary schools . In 2004, 32% of Indian girls aged 6–10 23–26 generational time horizon , (2) school feeding programs are years were covered by the program. Finally, following a sub- implemented in almost every country , and (3) social safety nets stantial increase in the budget allocation for the program in such as India’s MDM scheme have the potential for population- 2006, by 2011, 46% of girls aged 6–10 years beneﬁted from the level stunting reduction as they are implemented at scale and target program. Coverage among boys was similar throughout this multiple underlying determinants in vulnerable groups . period. NSS-CES data show that substantial state variability in At a broader level, a substantial literature documents effects of MDM rollout existed even ten years after the central mandate. cash transfer programs on education of girls in low- and lower- A complete listing of state heterogeneity in program roll-out middle-income countries . While transfer programs clearly can be found in Supplementary Table 1. address food security, their track record on improving anthro- Our empirical exploration of the intergenerational beneﬁts of pometry is mixed at best, possibly because evaluations focus on the MDM scheme was motivated by the observation that stunting 28,30 relatively short-term impacts . However, even in the United prevalence was lower among children aged 0–5 years in 2016 in States, a timely transfer—for example, the Supplemental Nutri- states where MDM coverage was higher in 2005 (Fig. 2). The tion Assistance Program—has been shown to have health beneﬁts ability of historical MDM coverage to predict the prevalence of over time . Other studies document effects of cash transfers, stunting in 2016 suggests that a mother’s exposure to the program health insurance, and other programs for children in beneﬁciary during primary school may have future returns for her children. households on future adult outcomes such as incomes, achieved However, the observed association may be biased because policy 32 33,34 35 schooling , nutritional status , and mortality . variables in observational data are unlikely to be independent of The described literature suggests a potential pathway through latent individual and institutional characteristics . which school feeding programs and other cash transfer or in-kind safety nets focused on education may have intergenerational effects on child nutrition outcomes. Current frameworks for Birth cohort ﬁxed effects analyses. To inform the birth cohort understanding the intergenerational transmission of health ﬁxed effects analysis, we examined coverage and scale-up of the 2 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE Mid-day meal (MDM) program participation among girls aged 6-10y Stunting among children <5y No data 0-9.9% 10-19.9% 20-29.9% 30-39.9% 40-49.9% ≥50% 38% 46% 32% Coverage: 6% Year: 1999 2004 2011 Pathway to impact: MDM is offered at school Adolescent girl (now 12-17y) who Children (<5y) of women (6-10y); school enrolment was exposed to MDM in 2004 has exposed to MDM in 2004 and attendance increases increased years of education and (now 17-22y) are less likely height to be stunted Fig. 1 Overview of study design and proposed pathway. Coverage refers to the proportion of girls aged 6–10 years who received a MDM in school. Source for MDM program coverage data (green maps): NSS-CES 55 (2000), 61 (2005) and 68 (2012). Source for child stunting data (red map): NFHS4 (2016). MDM, mid-day meal. Source data are provided as a Source Data ﬁle. ab 60 -1 Low SES Middle SES High SES -1.2 30 -1.4 -1.6 -1.8 1980 1986 1992 1998 1980 1986 1992 1998 Mother's birth year Mother's birth year Fig. 3 MDM coverage and child HAZ by mother’s birth year and socioeconomic status. Bottom 3 deciles are the poorest households in the 0 10 20 30 40 50 60 70 sample and top 4 deciles are non-poor. MDM exposure of women born State-level MDM coverage in 2005, % between 1980 and 1998 (a) and HAZ of children under 5 years old in 2016 Fig. 2 Association between stunting prevalence among children under 5 of mothers born between 1980 and 1998 (b). Source of MDM coverage years old in 2016 and MDM coverage among girls 6–10 years old in data: NSS-CES 50 (1994), 55 (2000), and 61 (2005). Source of HAZ data: 2005. Each circle represents an individual state in India, with the size NFHS 4 (2016). HAZ height-for-age z-score, MDM mid-day meal. Source representing the state population size. Fit line and shaded 95% conﬁdence data are provided as a Source Data ﬁle. interval are also weighted by state population size. Sources: NFHS 4 (2016) for stunting data and NSS-CES 61 (2005) for MDM coverage data. MDM maternal birth year, wealth, state, and state-speciﬁc-birth-year mid-day meal. Source data are provided as a Source Data ﬁle. ﬁxed effects, as well as a set of child-speciﬁc controls, HAZ in children born to mothers who lived in areas with 100% MDM coverage was 0.40 SD higher than HAZ in children born to MDM scheme and HAZ of children by mother’s birth year and mothers living in areas without the MDM (p < 0.05). The SES. The rate of MDM scale-up across SES deciles moved in inclusion of ICDS and PDS access variables did not attenuate tandem with child HAZ along the mother’s birth year axis this association. The effect of the program varied by SES; children (Fig. 3). Later-born mothers from poor households were more from poor households had the largest effect (0.5 SD, p < 0.05) likely to be exposed to the program than either earlier-born followed by children from middle SES strata (0.33, p < 0.05), mothers or mothers from non-poor households (Fig. 3a). HAZ in relative to children from the wealthiest SES strata. In robustness children also increased with later mother’s birth year and was checks, program access coefﬁcients were slightly attenuated but higher in non-poor households compared to poor households remained signiﬁcant when adding birth year speciﬁc SES ﬁxed (Fig. 3b). The observed trends provide motivation for using effects but were not signiﬁcant after adding birth year and state- MDM rollout by mother’s birth year as a source of variation that speciﬁc SES ﬁxed effects. Further, regressions on subsamples of is time varying and cohort speciﬁc . stunted children showed higher precision but smaller coefﬁcients In the birth cohort model, maternal MDM coverage was for the beneﬁts of MDM coverage on HAZ compared to children associated with future child HAZ (Fig. 4a). After adjusting for who were not stunted (Supplementary Fig. 5). NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 3 Stunting prevalence in 2016, % MDM coverage, % Height-for-age-z-score, SD ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w a b Intervention Control Intervention Control -1 -1.1 MDM coverage 20 -1.2 10 -1.3 0 -1.4 -4 -3 -2 -1 0 1 2 3 4 -4 -3 -2 -1 0 1 2 3 4 Event time (0=birth year 1992) Event time (0=birth year 1992) -.6 -.4 -.2 0 .2 .4 .6 .8 c d -.018 -.03 .5 Parallel trends -.021 Parallel trends .00093 -.019 -.028 MDM x poor .039 .046 MDM x middle DID of slopes .042 DID of slopes .055 .045 .029 -.209 -.1 -.05 0 .05 .1 .15 -.1 -.05 0 .05 .1 .15 MDM coverage Height for Age Z-score (SD) per year Height for Age Z-score (SD) per year -.6 -.4 -.2 0 .2 .4 .6 .8 District RE District FE State FE Low SES Middle SES High SES HAZ(SD) difference at full MDM coverage Fig. 5 Relationship between MDM and future child HAZ: controlled Fig. 4 Relationship between MDM coverage and future child HAZ: birth interrupted time series analyses. All models exclude Kerala and Tamil cohort ﬁxed effects model. Panel a shows the relationship between MDM Nadu. Panel a shows MDM coverage by event time across intervention and coverage and future child HAZ in the birth cohort model (Eq. 1) while panel control states. The program begins between event time 0 and 1. Panel b b shows the relative association across wealth strata (Eq. 2). The circles shows the local polynomial of HAZ of children in 2016, born to women represent the point estimates and whiskers are 95% conﬁdence intervals. belonging to birth cohorts, before and after the start of the program in each Point estimates are interpreted as the difference in HAZ due to 100% state. The shaded gray area indicates the 95% conﬁdence interval. Panel c exposure to the MDM scheme during primary school years for the relevant shows the coefﬁcient on γ (parallel trends) and γ (DID) from Eq. (3). 6 7 sample. Point estimates in panel b for MDM × poor and MDM × middle are Coefﬁcients from three models are speciﬁed as Eq. (3) plus random effects the relative effect of 100% MDM coverage for that SES stratum compared and ﬁxed effects for district and state. Panel d: γ (parallel trends) and γ 6 7 to the average effect of 100% MDM coverage for the wealthiest four (DID) from Eq. (3) with state ﬁxed effects run on a subset of low (SES 1–3), deciles (MDM coverage). MDM coverage is the proportion of girls born middle (SES 4–6), and high (SES 7–10) households. The squares/diamonds between 1980 and 1998, within state-speciﬁc socioeconomic status deciles, represent the point estimate and whiskers are 95% conﬁdence intervals. who reported receiving at least 10 meals free of cost at school in the The DID coefﬁcient can be interpreted as the difference in the average rate previous month. All models control for child age, sex, birth order, maternal of change in HAZ, per-year, before versus after MDM started, in the antenatal care (4+ visits), institutional birth, residence (urban/rural), intervention compared to control states All models control for child age, religion, caste, access to services from the Integrated Child Development sex, birth order, maternal antenatal care (4+ visits), institutional birth, Services (dummies for receiving take home rations, child health check-ups, residence (urban/rural), religion, caste, access to services from the pre-school education, weight measurements, and nutrition counseling) and Integrated Child Development Services (dummies for receiving take home the Public Distribution System (household has a Below Poverty Line card to rations, child health check-ups, pre-school education, weight obtain subsidized food). The models include ﬁxed effects for mother’s birth measurements, and nutrition counseling) and the Public Distribution year, state, household wealth, and for state × mother’s birth year. All System (household has a Below Poverty Line card to obtain subsidized models cluster standard error estimates at the district level. Sources: NFHS food). All models cluster standard error estimates at the state level. 4 (2016) for outcome and covariates. NSS-CES 50 (1994), 55 (2000), and Sources: NFHS 4 (2016) for outcome and covariates. NSS-CES 50 (1994), 61 (2005) for MDM coverage data. HAZ height-for-age z-score, MDM 55 (2000), and 61 (2005) for MDM coverage data. FE ﬁxed effects, MDM mid-day meal, SES socioeconomic status. Source data are provided as a mid-day meal, RE random effects, SES socioeconomic status. Source data Source Data ﬁle. are provided as a Source Data ﬁle. Controlled interrupted time series analyses. The controlled Program pathways. When examining factors that the MDM may interrupted time series model exploits variation in the timing of work through to inﬂuence child HAZ, full MDM coverage during the expansion of the MDM program to estimate program beneﬁts primary school years was a meaningful predictor of all factors relative to a reference period (event time 0 = birth year 1992). examined (Table 1). Full MDM coverage predicted 3.9 years of MDM expansion between event time 0 and 4 (birth years attained maternal education in years, delaying age in years at ﬁrst 1992–1996 capturing the short run impact of the program) dif- birth by 1.6 years, having a fewer (−0.8) children, a higher fered substantially across intervention and control states (Fig. 5a). probability of having at least four antenatal care visits (22%), and Trends in child HAZ were parallel between event time −4 and 0 giving birth in a medical facility (28%) (all p < 0.001). Full MDM across intervention and control states (Fig. 5b). After event time coverage predicted higher adult height among direct beneﬁciaries 0, intervention states saw a larger change in child HAZ compared (0.51 cm) but the association was not statistically signiﬁcant. to control states. In regression models, the coefﬁcient for parallel trends was not signiﬁcant, conﬁrming that trends in child HAZ were statistically similar across intervention and control states Regression decomposition. Our ﬁndings can be put into context before the intervention (Fig. 5c). The estimated association was by considering changes in HAZ among children under 5 years of similar across all three speciﬁcations, 0.038, 0.041, and 0.044 SD age reported in the National Family Health Surveys. HAZ per year (p < 0.05). Relative to wealthier households, the effect improved by 0.4 SDs between 2006 and 2016, on average. Using estimate of the MDM in intervention states was larger among Eq. (4), with an average MDM coverage of 32% in 2004 at the poor and middle-income households at 0.044–0.055 SD per year national level (NSS-CES 61) multiplied by the effect size of 0.166 (p < 0.10) (Fig. 5d). In robustness checks, effect coefﬁcients were SD (raw data model) to 0.401 SD (smoothed data model), we stable when excluding Gujarat, Odisha, and Chhattisgarh (some estimate the MDM explains 0.053–0.128 SD or 13.3–32.1% of districts in these states adopted MDM after Tamil Nadu and average change in HAZ. Using Eq. (5), with an average of 2.6 Kerala) from treatment states (Supplementary Fig. 2). years of exposure multiplied by the effect size of 0.044 SD per 4 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications MDM coverage, % Height-for-age z-score, SD NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE Table 1 Relationship between MDM and direct beneﬁciary education, height, fertility, and health service use in Indian women born between 1980 and 1998. Education, years Height, cm Age at ﬁrst Children, number Antenatal Institutional birth, years care, binary birth, binary Coefﬁcient 3.95 0.51 1.62 −0.80 0.22 0.28 Standard error (0.46) (0.36) (0.20) (0.07) (0.03) (0.02) P value <0.000 0.163 <0.000 <0.000 <0.000 <0.000 R 0.38 0.12 0.32 0.36 0.24 0.15 N 218,810 215,812 218,810 218,810 218,528 218,218 Coefﬁcients are from Eq. (1). Point estimates are interpreted as the difference in the outcome due to 100% exposure to the MDM scheme during primary school years. MDM coverage is the proportion of girls born between 1980 and 1998, within state-speciﬁc socioeconomic status deciles, who reported receiving at least 10 meals free of cost at school in the previous month. All models control for residence (urban/rural), religion, and caste. The models include ﬁxed effects for mother’s birth year, state, household wealth, and for state-speciﬁc mother’s birth year. All models cluster standard error estimates at the district level. Sources: NFHS 4 (2016) for outcome and covariates. NSS-CES 50 (1994), 55 (2000), and 61 (2005) for MDM coverage data. MDM mid-day meal, SES socioeconomic status. year, we estimate the MDM can explain 0.114 SD or 28.6% of expenditure, assets, and parental education. We ﬁnd that girls in average change in HAZ. The range estimated contributions are government schools are, on average, 0.89 cm shorter than those in similar in magnitude and relatively substantial, considering that private schools (p < 0.001). When we add a dummy variable HAZ is dependent on a large set of determinants, of which each indicating the receipt of MDM during primary school for these can individually only explain a small part of total variation in girls (identiﬁed in IHDS-1), we ﬁnd that MDM is associated with South Asian countries . a higher height of 1.3 cm on average (p < 0.001), while govern- ment school attendance is associated with 1 cm lower height (p < 0.001). This suggests that selection effects from program place- Migration. A possible concern for our estimates is susceptibility ment in government schools are likely to bias our estimates to the effects of migration. Since we measure MDM exposure at downward and that MDM is the driver of higher height among the state level in the past and associate it with child nutrition in government school beneﬁciaries. the future, attribution of the estimated role of MDM exposure would be weakened in the presence of substantial migration across states. A recent study allays this concern by providing Testing ﬁxed effects models with raw MDM coverage data. The estimates on migration in India. Although 30% of India’s popu- MDM coverage estimate from a regression model using Eq. (1) lation has ever migrated, two-thirds are intra-district migrants, and the raw coverage data is statistically signiﬁcant but attenuated more than half of whom are women migrating for marriage .In to 0.166 SD as expected (Supplementary Table 5, model 1). We 2001, only 4% of India’s population migrated across state also speciﬁed a second set of regressions using only the 2004 NSS borders . Therefore, migration is not a major concern for mis- data, and matched MDM coverage by district and SES. Again, we classiﬁcation of treatment status in our models. ﬁnd an attenuated but signiﬁcant coefﬁcient of 0.115 SD (Sup- plementary Table 5, model 2) and, as expected, a larger coefﬁcient Discordant SES matching between NSS-CES and NFHS. of 0.189 SD among poor households (p < 0.05) (Supplementary Overall, we ﬁnd a 78% concordance between expenditure-based Table 5, model 3). As the district level exposure does not have SES deciles measured in 2005 and asset-based SES deciles mea- temporal variation by birth year, this model is not directly sured in 2012 at the state level using India’s IHDS (Supplemen- comparable with the birth cohort model. However, it does tary Fig. 3). Given this 22% discordance, we cannot rule out that demonstrate that MDM coverage variation by district and SES is our estimates are somewhat biased due to imperfect classiﬁcation strongly correlated with HAZ of children of mothers born by SES status. However, the degree of bias is likely to be small between 1993 and 1997. because mobility across deciles is limited (the IHDS shows that a The MDM coverage estimate from a regression model using household generally only moves up by one or two SES deciles Eq. (1) and the log-linear smoothed coverage data are statistically over 7 years, if they move at all) and MDM coverage within states signiﬁcant but attenuated to 0.261 SD (Supplementary Table 6, does not ﬂuctuate greatly with small increments of SES classes (in model 1). However, attenuation here is smaller in magnitude 2005, coverage in the IHDS sample ranged between 53 and 62% compared to those using raw data. The model using Eq. (2) shows in the bottom four SES deciles). Moreover, non-differential that children from poor households had the largest effect (0.468 misclassiﬁcation as a form of measurement error generally tends SD, p < 0.01) followed by children from middle SES strata (0.296, to bias estimates towards the null . p < 0.05), relative to children from the wealthiest SES strata (Supplementary Table 6, model 2). Overall, we conclude that both the smoothed and raw data Matching by caste and religion. To test the sensitivity of our models provide evidence of an effect of maternal MDM coverage estimates to demographic measures of socioeconomic position on child anthropometry, though the size of the effect depends on that are less likely to change over time, we matched MDM cov- the preferred model. We have provided evidence that this effect is erage by state of residence, caste, and religion. Similar to SES robust to varying model speciﬁcations, and that the effect of matching, adjusted full maternal MDM coverage using caste and MDM coverage is largest among the poorest households. religion matching was associated with an improvement in HAZ Moreover, the control interrupted time series models do not among children aged 0–59 months (Supplementary Fig. 4). use smoothed coverage but provide qualitatively similar estimates. Selection bias from program placement in government schools. Using 2011 data (IHDS-2) we tested whether girls aged 11–17 years in government schools are shorter than those in private Discussion schools. We ﬁt a model with state ﬁxed effects that controls for We have shown that investments made in school meals in pre- child age, urban residence, occupation, household size, household vious decades were associated with improvements in future child NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w linear growth. The plausibility of this ﬁnding is supported by an received free meals at school from the National Sample Survey of Consumer 44–47 Expenditure (NSS-CES) (1993, 1999, and 2004 rounds) . These data are association between MDM exposure and underlying determi- combined with data on child height-for-age z-scores in 2016 from wave four of nants of child linear growth: women’s education, fertility, and India’s Demographic Health Survey, the National Family Health Survey (NFHS) . health service use. As the analysis covers a large nationally Both are large nationally representative surveys, which make it possible to match representative sample of households, the results reﬂect a program exposure to MDM by cohorts of girls born between 1980 and 1998 at the district level with data on mean child height in the same locations in 2016. The 2016 implemented at scale, with all its ﬂaws, and not a pilot program NFHS4 sample included 217,940 women with 196,310 children under 5 years of designed to provide proof of concept. This, of course, comes at a age. NSS-CES data from 2011 were also used for generating maps for coverage but cost; we could not follow a randomized cohort of girls from not for the primary analyses. Our interest was examining next generation beneﬁts primary school to childbearing. We put the magnitude of the on child stunting and our hypothesis was that intergenerational effects work through ﬁrst generation improvements in education, height, fertility, and access to association into context by using regression decomposition to 14,43,49–52 health services . We expected larger inﬂuence of maternal coverage estimate the share of the actual HAZ improvement explained by compared to paternal coverage given previous evidence showing larger program the predicted MDM effect on HAZ. impacts on girls than on boys . We support our main ﬁndings by using the 2004 While others have examined the effects of school feeding and 2011 rounds of Indian Human Development Surveys for descriptive analyses 53,54 programs on education and nutrition in beneﬁciaries themselves, and robustness checks . IHDS provides a wide array of variables that are not available in the NSS or the NFHS and offers supportive evidence on the main to our knowledge our paper is the ﬁrst to demonstrate an inter- estimates and model assumptions. Our study was a secondary analysis of existing generational transmission of beneﬁts. This ﬁnding provides evi- public survey data; hence, no ethical approval was required for our study. All dence that, when intergenerational effects are considered, the surveys complied with ethical norms with appropriate approvals and consent taken complete beneﬁt of school feeding programs at scale for linear at the time of survey. Summary statistics for the primary and secondary outcomes examined in this paper are shown in Supplementary Table 3. Summary statistics growth is much larger than previously understood. The result that for the covariates from NFHS are shown in Supplementary Table 4. a school feeding program is related to the nutritional status of children in the next generation also has important implications Identiﬁcation strategy. In an ideal experiment, children would be randomly for other transfer programs. The literature generally focuses on assigned access to free lunches from the MDM program in primary school and we investments in nutrition during the 1000-day period to reduce would compare the average HAZ outcomes for the children of the MDM bene- childhood stunting; our ﬁndings suggest that intervening during ﬁciaries and of the MDM non-beneﬁciaries when the original children in the the primary school years can make important contributions to experiment reached adulthood. In the absence of randomized treatment allotment, we chose to use panel data techniques from repeated cross-sections to exploit the reducing future child stunting, particularly given the cumulative strengths of the available data for identiﬁcation—the fact that the data cover birth exposure that is possible through school feeding programs. cohorts over a long period and that MDM coverage varies by state of residence and School meal programs are often motivated by their potential to SES. SES was calculated using a principal component analysis of household assets, increase schooling, particularly that of girls. While enrolment including cooking fuel, ﬂoor and wall materials, land and house ownership; and the possession of assets, including a mattress, pressure cooker, chair, bed, table, fan, parity is within reach in primary schooling –between 2000 and TV, sewing machine, phone, computer, fridge, watch, bicycle, motorbike, car; and 2015, the number of primary school-age children not in school the possession of animals, including cows, goats, and chickens. declined globally from 100 million to 61 million —there is a larger goal of primary and post primary school completion. Very Birth cohort ﬁxed effects analyses. Year of birth, SES decile, and state of resi- little in the literature on school meal programs can quantify dence were used to determine an individual’s exposure to the program. In India, program contribution to total years of schooling completed. children are expected to attend primary school between the ages of 6 and 10 years. Moreover, evidence that the scale-up of school meals is associated The NSS-CES provide data on the age of all household members and whether they received free meals at school in the past 30 days. Of all the girls aged 6–10 years in with increased heights of women—in a population in which the 2005 NSS sample who reported receiving any free meals at school (N = 8873), stunting has been historically linked with maternal under- 95.6% reported receiving at least 10 meals in the previous month. We used a nutrition—provides a new perspective on the contribution of minimum of 10 meals per month to ensure that our coverage estimates were for such programs. This reinforces an increased attention to seeking children who received the program with ﬁdelity. Models were run separately using any MDM access (at least 1 meal) and comparable results were obtained. We use opportunities to improve nutrition in the “next 7000 days” , that this information to calculate the percentage of all girls aged 6–10 years covered by is, to ﬁnd means of addressing undernutrition should efforts in the program for cohorts born between 1980 and 1998. This period gives us an the high priority period prior to a child’s second birthday not be approximately equal number of birth cohorts who were born before and after the fully successful. The results here show that school meals may introduction of the MDM scheme. Since the MDM scheme was introduced in 1995, those born after 1989 would be able to receive free meals in primary school. In contribute to education, nutrition (height), later fertility deci- addition, the NSS-CES provide measures of SES and state of residence, which sions, and access to health care; by doing so, school meals may allowed us to calculate coverage rates for all girls aged 6–10 years, speciﬁc to each reduce the risk of undernutrition in the next generation. In its SES strata in all Indian states. current form, India’s MDM scheme has the potential to address For any cohort, MDM exposure is a function of the number of years an average multiple underlying determinants of undernutrition. Improving child spends in primary school and when the program started in the school they attended. In an ideal data setting, to obtain an accurate coverage estimate for a the quality of meals provided and extending the program beyond birth cohort, we would have data from ﬁve cross-sections surveyed consecutively. primary school might further enhance its beneﬁts , though we For example, to obtain an estimate of MDM coverage for the 1994 cohort, we could not empirically test these hypotheses given the ideally would have coverage data on 6-year-old children measured in 2000, 7-year- available data. olds in 2001, 8-year-olds in 2002, 9-year-olds in 2003, and 10-year-olds in 2004. We would then average these ﬁve coverage estimates into a single estimate, The MDM is mandated by the Supreme Court of India as a representing average MDM exposure assuming a typical ﬁve-year period in social protection program addressing food insecurity. The social primary school for the 1994 cohort. The averaging is necessary because any single protection role of addressing hunger and food insecurity may be a year does not accurately reﬂect exposure for all 5 years in primary school. justiﬁcation by itself for school-based transfers in many settings . Each NSS-CES repeated cross-section, conducted within 5 year intervals, provides MDM coverage by child age as measured in the survey year. We used However, evidence such as presented here depict these programs linear interpolation to estimate a smoothed continuous exposure indicator that as contributing to both food security and to improved outcomes varies by maternal birth year, state, and SES. For example, using coverage estimates in the next generation, thus contribute to the policy framework for 6 year olds in 1999 NSS-CES (birth year 1993) and 6 year olds in the 2004 NSS- for school-based interventions. CES (birth year 1998), we ﬁrst used linear interpolation to estimate the average rate of increase in MDM coverage for 6 year olds for the years (2000, 2001, 2002, and 2003) with no NSS-CES data (these correspond with birth years 1994, 1995, 1996, Methods and 1997). Next, we performed similar interpolation for 7-, 8-, 9-, and 10-year-old Data sources. This paper relies on evidence from seven rounds of three publicly children. This provided smoothed coverage estimates for children born in 1993 for available nationally representative surveys (Supplementary Table 2). The primary the survey years 1999, 2000, 2001, 2002, and 2003—the years the 1993 cohort analysis in this paper uses data on whether children born between 1980 and 1998 would have aged from 6 to 10 years. We take the average coverage for these 5 years 6 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE as the ﬁnal estimate of coverage experience of a speciﬁc birth year (Supplementary households to attend primary school (and to improve nutrition); therefore, the Fig. 1). This process of smoothing (i) estimates the relationship between maternal estimates in Eq. (1) are likely to mask heterogeneity of response to the program. school meals exposure and annual child HAZ outcomes under an assumption of a Masking is anticipated because outcome data from children sampled from non- linear trend in exposure and (ii) reduces probable bias due to measurement error poor households, who would be more likely to opt out of the government school present in the raw data by moving extreme values closer to the center of the system in favor of private schools, would inﬂuence average effect sizes . We expect distribution. that mothers who were enrolled in government schools during their childhood Throughout the paper, we use the term MDM coverage, which refers to an would have worse nutritional outcomes and this might place a downward bias on estimate of state-by-year average program exposure during primary school for the our estimates. To investigate the existence of such heterogeneity, we compared birth cohorts in the sample, under the assumption that coverage increases in a associations across SES groups. We created SES deciles and grouped women in the linear fashion within age groups of children in primary school surveyed in the years bottom three (poor), middle three, and top four (non-poor) deciles to create two 1993, 1999, 2004, and 2011. It is almost certain that exposure in the interval wealth strata. We estimated models for differential associations for poor, middle between 2 years lies between the values in the end points; the assumption that the versus non-poor households by modifying Eq. (1) as follows: expansion is linear is a plausible pattern of program roll out. We assume that Y ¼ γ þ γ MDM Poor þ γ MDM Middle þ γ MDM within 5-year intervals, the duration of primary school, age-speciﬁc trends in iwst wst wst wst wst wst 0 1 2 3 ð2Þ coverage would have increased gradually. Gradual rollout is typical of at-scale þ γ T þ γ S þ γ S T þ γ W þ γ C þ γ P þ ε 4 t 5 s 6 s t 7 w 8 iwst 9 iwst iwst programs in developing countries with numerous implementation, ﬁnancing, and bureaucratic challenges . However, in sensitivity analyses, we subject the where Poor and Middle dummy variables for bottom three and middle (4–6) wst wst assumption of a linear scale-up to an additional robustness check where we smooth SES deciles, respectively, with the top four SES deciles serving as the reference non- MDM coverage using a log-linear process. poor group. γ and γ measure if poor and middle SES households beneﬁtted more 1 2 Next, using birth year, SES deciles and state of residence, we match NFHS data from MDM coverage compared to non-poor households. We expect γ to be larger with NSS-CES data for the percentage of girls covered by the MDM for cohorts than γ , and if these coefﬁcients are statistically signiﬁcant and of a large order, born between 1980 and 1998. NFHS data provide anthropometric measurements then we have evidence that MDM program beneﬁts differ by SES. Note that SES for the last three births for each mother. We use data for all available children with here is current, and mother’s SES may have differed in childhood. To this end, we valid anthropometric measurements. We calculate HAZ using the “zscore06” offer evidence in our sensitivity analyses that SES mobility is likely modest. STATA routine which automatically excludes outlier measurements. We specify the following model: Controlled interrupted time series models using state rollout timing. The birth cohort model exploits variation in treatment measured as the proportion of children Y ¼ γ þ γ MDM þ γ T þ γ S þ γ S T þ γ W þ γ C þ γ P þ ε iwst 0 1 wst 2 t 3 s 4 s t 5 w 6 iwst 7 iwst iwst covered by the program within a birth year, state, and across SES strata. It allows us ð1Þ to express the relationship between MDM and HAZ as a function of coverage. However, it comes at the cost of potential for endogeneity because MDM coverage where Y is the height-for-age z-score for child i belonging to SES strata w in iwst could potentially be associated with changes in living conditions that vary within state s in mother’s birth year t. MDM is a continuous indicator coded as the wst cohorts deﬁned by state, birth year, and SES strata. An alternate model exploits the proportion of mothers covered by the MDM as children and ranges between 0 and differential timing of MDM rollout across Indian states as a robustness check on 1. T represents birth-year ﬁxed effects which forces identiﬁcation of within birth- the birth cohort model. This alternative can reveal insights for the short-term year effects and controls for time-varying national level economic changes, cumulative beneﬁts of the program . programs, and policies. Examples of these are national programs such as the States implemented the program at different times; de-facto, the program was National Health Mission introduced in 2005 (ref. ) and changes in national GDP, rolled out in the three phases (Supplementary Table 1). According to the NSS data, which has shown robust growth . MDM coverage patterns by state and birth year show that Tamil Nadu and Kerala, We estimated Eq. (1) using MDM coverage at the state level disaggregated by i.e. “phase 1” states, had average coverage greater than 20% for maternal birth year wealth strata. W represents the wealth-decile ﬁxed effects and provides controls 1988). These states initiated school feeding programs well before the central for all unobserved time-invariant factors associated with household wealth and government funded MDM. Following the central government order, in phase 2, MDM coverage. S is the state ﬁxed effects which controls for all for time-invariant other states—Odisha, Himachal Pradesh, Uttaranchal, Haryana, Rajasthan, Sikkim, differences across states with high and low MDM exposure. S * T or state-birth- s t Tripura, West Bengal, Chhattisgarh, Madhya Pradesh, Gujarat, Maharashtra, year ﬁxed effects controls for unobserved state-speciﬁc time-varying factors that Andhra Pradesh, and Karnataka—implemented the program at scale with coverage could be correlated with the outcome such as the state’s political climate, varying increasing by more than 10% between maternal birth years 1992–1996. In the degrees of implementation of welfare programs, agricultural policies, and remaining states (phase 3), MDM coverage was below 5% and increased by less educational subsidies. A concern for a model estimated without this parameter is than 10% between maternal birth years 1992–1996. that states that introduced free meals in primary school at different times and rates These roll-out patterns lend themselves to analysis using a controlled of coverage expansion could be systematically different. For example, states with 63–65 interrupted time series design (CITS) . Conceptually, the CITS is a residents who had lower education or poorer nutritional status on average may combination of the difference-in-differences and interrupted time series models. It have been more likely to introduce the MDM. Similarly, states with better includes a within group before–after comparison, and a between-group governance may have been better equipped to implement the MDM program at comparison, strengthening the control for potential confounders. The ﬁrst scale. In either case, the correlation between outcomes and MDM implementation difference is the change in the outcome trend within each group, comparing the could be confounded with unobserved state-speciﬁc time-varying factors. period before MDM to the period after (slope change). The second difference is the C represents a vector of individual, household and survey-speciﬁc controls, iwst difference in slope changes in the control group compared to the intervention including child age, sex, birth order, mothers antenatal care status during group (difference-in-differences of slopes). The CITS reduces bias due to other pregnancy, birth in a medical facility, and household characteristics at the time the interventions or events occurring around the same time as the MDM intervention outcome was measured. The vector includes SES, caste, religion, and residence and allows comparison groups to start at different levels of the outcome. Moreover, (urban or rural). P represents a vector of individual and household-speciﬁc iwst the CITS controls for the improvement in HAZ that would be expected without the programmatic controls, including access to services from the Integrated Child MDM and tests for parallel trends within the model. Development Services (dummies for receiving take home rations, child health We exclude Tamil Nadu and Kerala from CITS analysis as they were early check-ups, pre-school education, weight measurements, and nutrition counseling) MDM implementers and both states have better nutrition outcomes compared to and the Public Distribution System (household has a Below Poverty Line card to other states in India. We focus on maternal birth years 1988 to 1996, when we have 59,60 obtain subsidized food) . Controlling for these variables reduces possible a pre intervention period with no MDM across all states, and a post intervention confounding from government interventions that could beneﬁt current child period when some states introduced the program while others did not. Phase nutritional status. All standard error estimates were clustered at the district level. 2 states form the intervention group and phase 3 states serve as the control group. Clustering adjusts standard error estimates after accounting for intra-district We parameterize the CITS model using Eq. (3). correlations and assumes that residuals are independent across districts . The coefﬁcients estimated by Eq. (1) are intent-to-treat (ITT) estimates because Y ¼ γ þ γ Int þ γ T þ γ Post þ γ Post T þ γ Post Int ist 0 1 s 2 t 3 t 4 t t 5 t t the MDM coverage variable measures “potential exposure” to the program on ð3Þ þ γ T Int þ γ T Int Post þ γ C þ γ P þ ε t s t s t ist ist ist entire birth cohorts. Our ITT estimates are a policy-relevant parameter for an ex- 6 7 8 9 post analysis of the effects of a large program on the entire population (birth In state s in mother’s birth year t, Int is a dummy for the intervention states, T s t 21,22 cohorts) . Our models, based on population representative MDM coverage, is the event time, a discrete variable (for maternal birth years 1988–1996) that is estimate the magnitude of improvement in child undernutrition that can be centered at 1992 and ranges between −4 and 4. Post is a dummy for maternal expected if a cohort is potentially treated. birth years 1993–1996. In Eq. (3), γ tests the null hypothesis of parallel pre- intervention trends; if not signiﬁcant, we reject this hypothesis and conclude that pre-interventions differed between intervention and control groups. γ is the Testing for differential beneﬁts for the poor. IHDS data show that 80% of all MDM beneﬁciaries in 2004 attended government schools and that two-thirds of coefﬁcient of interest, and represents a “difference-in-difference of slopes” between children attending government schools were from low-income households (bottom the intervention and control states. If γ is statistically signiﬁcant, the change in six SES deciles), suggesting that MDM was primarily implemented in government HAZ slope for intervention states differs from the change in HAZ slope for control schools rather than in private schools as an incentive for children from poor states. In other words, it tests for faster gains in child linear growth for states with NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w MDM. To account for spatial heterogeneity, we run three speciﬁcations of Eq. (3) use only the 2004 NSS raw coverage data, matched by district and SES. These by adding district random effects, district ﬁxed effects, and state ﬁxed effects. models specify the same level of coverage to SES groups within districts for birth To explore heterogeneity, we investigate differential associations by household cohorts 1993 to 1997. SES by running models within subsamples of poor (SES deciles 1–3), middle (SES deciles 4–6), and non-poor (SES deciles 7–10) households. For robustness, we Testing ﬁxed effects models with MDM coverage smoothed using log-linear check sensitivity of coefﬁcients to exclusion of Gujarat, Odisha, and Chhattisgarh process. To test the sensitivity of our estimates using linearly smoothed coverage, from the intervention group. These states had greater than 10% coverage at event we use an alternative log-linear smoothing process. This process assumes an time 0 and thus could arguably be placed in the phase 1 category. exponential growth in MDM coverage within 5-year intervals. We then ﬁt models with Eqs. (1) and (2). Regression decomposition. To test the plausibility of our results we performed regression-based decomposition with our estimates from Eqs. (1) and (3) (ref. ). Reporting summary. Further information on research design is available in the Nature From Eq. (1), we estimated the population level effect of the program between 2006 Research Reporting Summary linked to this article. and 2016 with Eq. (4). γ MDM2004 ð4Þ Data availability ΔHAZ 44–48,53,54 The conclusions of this article are based on publicly available datasets . Source where γ is the coefﬁcient of MDM from Eq. (1), MDM is the MDM coverage 2004 data are provided with this paper. The cleaned and merged dataset is available on the in 2004 and ΔHAZ change in HAZ between 2006 and 2016. 67 Harvard Dataverse at [https://doi.org/10.7910/DVN/JTN87W] . Source data are From the controlled interrupted time series model, we estimated the effect of provided with this paper. exposure to the program using Eq. (5). γ PostEventTime Code availability ð5Þ ΔHAZ The analysis code that reproduces all the tables and ﬁgures in the manuscript is available where γ is the coefﬁcient from Eq. (3), PostEventTime is the average event time 5 on the Harvard Dataverse at [https://doi.org/10.7910/DVN/JTN87W] . years before and after the start of the program, and ΔHAZ is the change in HAZ between 2006 and 2016. Estimates from Eqs. (4) and (5) are proportions and are Received: 21 February 2021; Accepted: 7 June 2021; expected to be less than 1 because the predicted difference in HAZ explained by MDM must be less than the total change in HAZ observed between 2004 and 2016. Program pathways. We next investigated plausible pathways that might support intergenerational links between the MDM program and child nutrition. We used Eq. (1) to investigate the association of the MDM with six factors that may be References related to the MDM program and which, in turn, correlate with child HAZ: 1. Development Initiatives. 2020 Global Nutrition Report—Global Nutrition mother’s education and height, mother’s age at ﬁrst birth, total number of children Report. https://globalnutritionreport.org/reports/2020-global-nutrition-report/ per mother, number of antenatal care visits attended by the mother during preg- (2020). nancy and if the child was born in a medical facility. We recognize that this is a 2. Ministry of Health and Family Welfare Government of India. National Family plausibility analysis and cannot isolate causality. Health Survey-4, 2015-2016, India Fact Sheet. http://rchiips.org/nfhs/pdf/ NFHS4/India.pdf (2017). Discordant SES matching between NSS-CES and NFHS. We matched MDM 3. Prendergast, A. J. & Humphrey, J. H. The stunting syndrome in developing coverage by state of residence and SES decile between the NSS-CES and NFHS. countries. Paediatr. Int. Child Health 34, 250–265 (2014). This assumes that (1) mobility across SES strata over time is minimal and (2) SES 4. Development Initiatives. Global Nutrition Report 2017: Nourishing the SDGs deciles in NFHS correspond well with those in the NSS. We therefore use panel (2017). data from the IHDS to assess the concordance of expenditure-based SES deciles 5. Black, M. et al. Early childhood development coming of age: science through measured in 2005 and asset-based SES deciles measured in 2012. Since IHDS the life course. Lancet 389,77–90 (2017). follows the same individuals over 7 years, we can track their mobility across SES 6. de Onis, M. & Branca, F. Childhood stunting: a global perspective. Matern. strata over time and then compare their status on both SES measurements. Child Nutr. 12,12–26 (2016). 7. de Onis, M. et al. The World Health Organization’s global target for reducing Matching by caste or religious group. To test the sensitivity of our estimates to childhood stunting by 2025: rationale and proposed actions. Matern. Child demographic measures of socioeconomic position, we matched maternal MDM Nutr. 9,6–26 (2013). coverage by birth year, state of residence and households’ caste/religious groups in 8. Bundy, D. A. P. et al. Investment in child and adolescent health and the NSS-CES and NFHS. The social groups used to match households were development: key messages from Disease Control Priorities, 3rd Edition. scheduled caste (Hindu), scheduled tribe (Hindu), Muslim, Christian, and others. Lancet https://doi.org/10.1016/S0140-6736(17)32417-0 (2017). Similar to the SES model, this model works by assigning a probability of exposure 9. Black, R. E. et al. Maternal and child undernutrition and overweight in low- to MDM for maternal birth cohorts that varies by state, religion, and caste. While income and middle-income countries. Lancet 382, 427–451 (2013). social groups do not follow strict income hierarchies across states, they have the 10. Corsi, D. J., Mejía-Guevara, I. & Subramanian, S. V. Risk factors for chronic advantage of being largely time invariant and thus do not introduce biases that undernutrition among children in India: estimating relative importance, result from income mobility. population attributable risk and fractions. Soc. Sci. Med. 157, 165–185 (2016). 11. Headey, D., Hoddinott, J. & Park, S. Drivers of nutritional change in four Testing ﬁxed effects models with raw MDM coverage data. To test the sen- South Asian countries: a dynamic observational analysis. Matern. Child Nutr. sitivity of our estimates using smoothed coverage, we offer an additional alternative 12, 210–218 (2016). using raw coverage data from NSS. These coverage estimates are from cross- 12. Alderman, H. & Headey, D. D. How important is parental education for child sections at speciﬁc points in time and are not smoothed using the age proﬁles of nutrition? World Dev. 94, 448–464 (2017). children in the NSS rounds. We ﬁrst created a scatter plot of smoothed coverage 13. Cavatorta, E., Shankar, B. & Flores-Martinez, A. Explaining cross-state estimates against the raw coverage data to gauge the degree and direction of the disparities in child nutrition in rural India. World Dev. 76, 216–237 (2015). smoothing process (Supplementary Fig. 6). 14. Kim, R., Mejía-Guevara, I., Corsi, D. J., Aguayo, V. M. & Subramanian, S. V. The maps in Fig. 1 show a discrete jump in coverage from 6% in 1999 to 32% in Relative importance of 13 correlates of child stunting in South Asia: insights 2004. The smoothed data attempt to ﬁll in data gaps on coverage for the years 2000 from nationally representative data from Afghanistan, Bangladesh, India, to 2003. The scatterplot of the smoothed coverage data against the raw data shows Nepal, and Pakistan. Soc. Sci. Med. 187, 144–154 (2017). that the smoothed data are less extreme than the raw data, which has many 0 and 15. Chen, Y. & Li, H. Mother’s education and child health: is there a nurturing 100% coverage estimates. These extreme values present in the raw data likely reﬂect effect? J. Health Econ. 28, 413–426 (2009). measurement error for cohort-speciﬁc coverage because they do not capture the 16. Raghunathan, K., Chakrabarti, S., Menon, P. & Alderman, H. Deploying the transition of increasing coverage for the initial years of program implementation, power of social protection to improve nutrition what will it take? Econ. Polit. so that coverage for any observation reﬂects only a single year during a time of Wkly 52, (2017). program expansion despite the fact that a student will have spent more than a 17. Ministry of Human Resource Development. Mid Day Meal Scheme. single year in school. To test our hypothesis that measurement error in the raw Department of School Education & Literacy, Government of India. http:// coverage data would attenuate results compared to those from the models using mdm.nic.in/ (2017). smoothed data in keeping with standard expectation with random errors in 18. Afridi, F. The impact of school meals on school participation: evidence from variables, we ran our primary birth cohort model with raw coverage matched by rural India. J. Dev. Stud. 47, 1636–1656 (2011). state, SES, and birth years. We also ran a second test of sensitivity with models that 8 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w ARTICLE 19. Drèze, J. & Kingdon, G. School participation in rural India. London School of 49. Bhutta, Z. A. et al. Evidence-based interventions for improvement of maternal Economics and Political Science (1999). and child nutrition: what can be done and at what cost? Lancet 382, 452–477 20. Chakraborty, T. & Jayaraman, R. School feeding and learning achievement: (2013). evidence from India’s Midday Meal Program. IZA Discussion Paper 10086 50. Rieger, M. & Trommlerová, S. K. Age-speciﬁc correlates of child growth. (2016). Demography 53, 241–267 (2016). 21. Afridi, F. Child welfare programs and child nutrition: evidence from a 51. Jayachandran, S. & Pande, R. Why are Indian children so short? The mandated school meal program in India. J. Dev. Econ. 92, 152–165 (2010). role of birth order and son preference. Am. Econ. Rev. 107, 2600–2629 22. Singh, A., Park, A. & Dercon, S. School Meals as a Safety Net: An evaluation of (2017). the Midday Meal Scheme in India. Econ. Dev. Cult. Change 62, 275–306 52. Özaltin, E., Hill, K. & Subramanian, S. V. Association of maternal stature with (2014). offspring mortality, underweight, and stunting in low- to middle-income 23. Nabwera, H. M., Fulford, A. J., Moore, S. E. & Prentice, A. M. Growth faltering countries. JAMA 303, 1507–1516 (2010). in rural Gambian children after four decades of interventions: a retrospective 53. Desai, S., Vanneman, R. & National Council of Applied Economic Research. cohort study. Lancet Glob. Health 5, e208–e216 (2017). India Human Development Survey (IHDS), 2005 (ICPSR 22626). Inter- 24. Addo, O. Y. et al. Maternal height and child growth patterns. J. Pediatr. 163, University Consortium for Political and Social Research. https://doi.org/ 549–554. e1 (2013). 10.3886/ICPSR22626.v12 (2018). 25. Hambidge, M., Mazariegos, M., Kindem, M., Wright, L. & Cristobal-Pereza, C. 54. Desai, S., Vanneman, R. & National Council of Applied Economic Research. Infant stunting is associated with short maternal stature. J. Pediatr. India Human Development Survey-II (IHDS-II), 2011-12 (ICPSR 36151). Gastroenterol. Nutr. 54, 117–119 (2012). Inter-University Consortium for Political and Social Research. https://doi.org/ 26. Stein, A. D. et al. Comparison of linear growth patterns in the ﬁrst three 10.3886/ICPSR36151.v6 (2018). years of life across two generations in Guatemala. Pediatrics 113, e270–e275 55. Deaton, A. Panel data from time series of cross-sections. J. Econ. https://doi. (2004). org/10.1016/0304-4076(85)90134-4 (1985). 27. Bundy, D. A., Drake, L. J. & Burbano, C. School food, politics and child health. 56. Cotlear, D. Going Universal. World Bank https://openknowledge.worldbank. Public Health Nutr. 16, 1012–1019 (2013). org/bitstream/handle/10986/22011/9781464806100.pdf (2015). 28. Ruel, M. T. & Alderman, H. Nutrition-sensitive interventions and 57. D’Silva, J. Can India pull off its ambitious National Health Mission? BMJ programmes: how can they help to accelerate progress in improving maternal https://doi.org/10.1136/bmj.f2134 (2013). and child nutrition? Lancet 382, 536–551 (2013). 58. Bosworth, B. & Collins, S. M. Accounting for growth: comparing China and 29. García, S. & Saavedra, J. E. Educational impacts and cost-effectiveness of India. J. Econ. Perspect. https://doi.org/10.1257/jep.22.1.45 (2008). conditional cash transfer programs in developing countries: a meta-analysis. 59. Chakrabarti, S., Raghunathan, K., Alderman, H., Menon, P. & Nguyen, P. Rev. Educ. Res. 87, 921–965 (2017). India’s integrated child development services programme; equity and extent of 30. Leroy, J. L., Ruel, M. & Verhofstadt, E. The impact of conditional cash transfer coverage in 2006 and 2016. Bull. World Health Organ. https://doi.org/10.2471/ programmes on child nutrition: a review of evidence using a programme BLT.18.221135 (2019). theory framework. J. Dev. Effect. 1, 103–129 (2009). 60. Chakrabarti, S., Kishore, A. & Roy, D. Effectiveness of food subsidies in raising 31. Hoynes, H., Schanzenbach, D. W. & Almond, D. Long-run impacts of healthy food consumption: public distribution of pulses in India. Am. J. Agric. childhood access to the Safety Net. Am. Econ. Rev. 106, 903–934 (2016). Econ. https://doi.org/10.1093/ajae/aay022 (2018). 32. Cohodes, S. R., Grossman, D. S., Kleiner, S. A. & Lovenheim, M. F. The effect 61. Kim, R., Mohanty, S. K. & Subramanian, S. V. Multilevel geographies of of child health insurance access on schooling: evidence from public insurance poverty in India. World Dev. 87, 349–359 (2016). expansions. J. Hum. Resour. 51, 727–759 (2016). 62. Muralidharan, K. & Kremer, M. Public-private schools in rural India. School 33. Almond, D., Currie, J. & Duque, V. Childhood circumstances and adult Choice International. https://doi.org/10.7551/mitpress/9780262033763.003.0005 outcomes: Act II. J. Econ. Lit. 56, 1360–1446 (2018). (2013). 34. Aizer, A., Eli, S., Ferrie, J. & Lleras-Muney, A. The long-run impact of cash 63. Bernal, J. L., Cummins, S. & Gasparrini, A. The use of controls in interrupted transfers to poor families. Am. Econ. Rev. 106, 935–971 (2016). time series studies of public health interventions. Int. J. Epidemiol. https://doi. 35. Wherry, L. R. & Meyer, B. D. Saving teens: using a policy discontinuity to org/10.1093/ije/dyy135 (2018). estimate the effects of medicaid eligibility. J. Hum. Resour. 51, 556–588 (2016). 64. Shadish, W., Cook, T. & Campbell, D. Quasi-experimental designs that use 36. Cheng, T. L., Johnson, S. B. & Goodman, E. Breaking the intergenerational both control groups and pretests. Experimental and Quasi-Experimental cycle of disadvantage: the three generation approach. Pediatrics https://doi. Designs (2002). org/10.1542/peds.2015-2467 (2016). 65. Bernal, J. L., Cummins, S. & Gasparrini, A. Interrupted time series regression 37. Drèze, J. & Khera, R. Recent social security initiatives in India. World Dev. 98, for the evaluation of public health interventions: a tutorial. Int. J. Epidemiol. 555–572 (2017). https://doi.org/10.1093/ije/dyw098 (2017). 38. Barrett, C. B. & Carter, M. R. The power and pitfalls of experiments in 66. Headey, D., Hoddinott, J. & Park, S. Accounting for nutritional changes in six development economics: some non-random reﬂections. Appl. Econ. Perspect. success stories: a regression-decomposition approach. Glob. Food Security 13, Policy 32, 515–548 (2010). 12–20 (2017). 39. Verbeek, M. & Vella, F. Estimating dynamic models from repeated cross- 67. International Food Policy Research Institute (IFPRI) & University of sections. J. Econ. https://doi.org/10.1016/j.jeconom.2004.06.004 (2005). Washington. Intergenerational Nutrition Beneﬁts of India’s National School 40. Kone, Z. L., Liu, M. Y., Mattoo, A., Ozden, C. & Sharma, S. Internal borders Feeding Program. https://doi.org/10.7910/DVN/JTN87W (2021). and migration in India. J. Econ. Geogr. https://doi.org/10.1093/jeg/lbx045 (2018). 41. Chen, Q., Galfalvy, H. & Duan, N. Effects of disease misclassiﬁcation on Acknowledgements exposure-disease association. Am. J. Public Health. https://doi.org/10.2105/ We acknowledge feedback from the participants of the following conferences where AJPH.2012.300995 (2013). drafts of the paper were presented: Nutrition 2018 (organized by the American Society 42. UNICEF. UNICEF Data Base on Primary Education (2018). for Nutrition), North East Universities Development Consortium (NEUDC) 2018, and 43. Alderman, H., Behrman, J. & Tasneem, A. The contribution of increased the National Institute of Nutrition (NIN) 2018 Centenary conference. Bill & Melinda equity to the estimated social beneﬁts from a transfer program: an illustration Gates Foundation through Partnerships and Opportunities to Strengthen and Harmonize from PROGRESA/Oportunidades. World Bank Econ. Rev. https://doi.org/ Actions Against Malnutrition in India (POSHAN), led by the International Food Policy 10.1093/wber/lhx006 (2019). Research Institute (IFPRI). 44. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, July 1993–June 1994, NSS 50th Round. National Data Archive. DDI-IND- MOSPI-NSSO-50Rnd-Sch1.0-1993-94 (2019). 45. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, Author contributions July 1999–June 2000, NSS 55th Round. National Data Archive. DDI-IND- S.C. conceived the idea for the research. S.C. and S.P.S. contributed to all other aspects MOSPI-NSSO-55Rnd-Sch1-July1999-June2000 (2019). including data analysis, interpretation, and manuscript preparation and writing. H.A., 46. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, P.M., and D.O.G. provided inputs to analysis and contributed to manuscript writing. All July 2004–June 2005, NSS 61st Round. National Data Archive. DDI-IND- co-authors read and approved the ﬁnal version of the manuscript. MOSPI-NSSO-61Rnd-Sch1-July2004-June2005 (2019). 47. National Sample Survey Ofﬁce. India—Household Consumer Expenditure, Type 1: July 2011–June 2012, NSS 68th Round. National Data Archive. DDI- IND-MOSPI-NSSO-68Rnd-Sch1.0-July2011-June2012 (2019). Competing interests 48. DHS Program. India: Standard DHS, 2015–2016 (2016). The authors declare no competing interests. NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-24433-w Additional information Open Access This article is licensed under a Creative Commons Supplementary information The online version contains supplementary material Attribution 4.0 International License, which permits use, sharing, available at https://doi.org/10.1038/s41467-021-24433-w. adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Correspondence and requests for materials should be addressed to S.P.S. Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless Peer review information Nature Communications thanks the anonymous reviewer(s) for indicated otherwise in a credit line to the material. If material is not included in the their contribution to the peer review of this work. Peer reviewer reports are available. article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from Reprints and permission information is available at http://www.nature.com/reprints the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afﬁliations. © The Author(s) 2021 10 NATURE COMMUNICATIONS | (2021) 12:4248 | https://doi.org/10.1038/s41467-021-24433-w | www.nature.com/naturecommunications

Journal

Nature Communications – Springer Journals

Published: Jul 12, 2021

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Intergenerational nutrition benefits of India’s national school feeding program

Intergenerational nutrition benefits of India’s national school feeding program

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Intergenerational nutrition benefits of India’s national school feeding program

Intergenerational nutrition benefits of India’s national school feeding program

References (92)

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies