Despite the popularity of school meals as interventions in education, their effect on learning and health outcomes is not clear. This study uses newly available longitudinal data from the state of Andhra Pradesh in India to estimate these effects in a non-experimental setting. Further, it aims at disaggregating the average program impacts to see if some groups benefited more than the others i.e. whether heterogeneity in program impacts was present. We use changes in WHO anthropometric z scores as the outcome variables to evaluate impact on nutrition, and scores on the Peabody Picture Vocabulary Test (PPVT) to assess the impact on learning skills.  The approach taken employs several different estimation strategies to correct for econometric issues in estimation; our preferred estimate for estimating health gains adopts an Instrumental Variables (IV) approach and for learning gains we use a difference-in-means estimator with propensity score matching methods.

We find the scheme delivers non-trivial gains in both nutrition and learning which are highly significant.  The program acts as a security net for children, cushioning them from negative nutritional factors; in particular, among younger children, there are large and significant gains for children who suffered from the impact of drought. In cognitive skills, we find that school meals boost PPVT scores by over 0.6 s.d. Evidence presented here, combined with previous findings of gains in school participation in other studies, reaffirms the effectiveness of school meal programs in developing countries.