Early Childhood Nutrition and Adult Outcomes in India

The investigators plan to use the 2010-2012 wave of the APCAPS data. They will access these publicly available data by applying for collaboration with the APCAPS study group.

The 2010-2012 follow-up survey invited all surviving children born during the original trial period of 1987-1990 (hereafter called index children) to participate. These index children were 20-25 years old at the time of this survey. Of the invited index children, 1,360 (715 in intervention villages and 645 in control villages) participated. The APCAPS surveys have collected a wide variety of data on socio-demographics, lifestyle, medical history, mental and reproductive health, anthropometry, cardiovascular physiology, spirometry, and biomarkers.

Using these data, the investigators will examine the association between birth in an intervention village and adult educational, marriage and labor market outcomes. For each study participant, a survey question collected information on the highest level of education attained by the participant (self-reported). From this variable, the investigators will create two indicators of whether the adult: (1) completed at least secondary school, i.e. class X/XII, intermediate, vocational course, or equivalent; (2) completed at least graduate level, i.e. Bachelor's degree, diploma, or equivalent.

The third outcome variable of the analysis will be whether the adult was ever-married (currently married, widowed, or divorced) at the time of the survey. Finally, the investigators will also examine whether the adult was either employed in paid or unpaid (e.g. household agriculture) work, or enrolled in higher education (degree or training course).

The investigators will use at least two different analytical methods. First, they will conduct a multivariate regression analysis (logit or probit) of each of the four outcome indicators. Along with the main independent variable of interest - birth in an intervention village - the model will include the following covariates: age, sex (whether female), and birth order of the index child, indicators of parental schooling attainment, religion, caste, and household standard of living as measured by quintiles of a composite asset index. All standard errors of regression will be heteoskedasticity-robust and clustered at the village level. The analysis will be conducted (and results reported) for the full sample and separately for the male and female subsamples.

In order to mitigate any bias arising from systematic differences in the characteristics of the intervention and control village adults (e.g. due to non-random attrition or other reasons), the investigators will also use quasi-experimental matching methods. One widely used methodology is matching based on propensity scores. The investigators will first conduct a selection analysis in which the likelihood of birth in an intervention village is regressed (probit or logit model) on the set of participant, parental, and household covariates mentioned above. Then, based on the predicted probability of intervention assignment (called propensity score) obtained from this model, participants in the intervention group will be matched with similar participant(s) in the control group. The investigators will use a variety of matching algorithms in order to test the sensitivity of the results, e.g. one-to-one nearest neighbor, Kernel, and multiple nearest neighbors matching. Common metrics for the quality of matching exercise and covariate balancing tests will be reported.

After matching, the investigators will report the average difference in outcome indicators between the intervention and matched control groups (technically named as the average treatment effect on the treated) as the estimated association between birth in an intervention village and adult outcomes. Findings from the full sample as well as male and female subsamples will be reported.