Breastfeeding and child development outcomes across early childhood and adolescence: Doubly robust estimation with machine learning

We estimate the impact of breastfeeding initiation and duration on a range of cognitive, health, and behavioral outcomes spanning early childhood through adolescence, employing a doubly robust estimation method to mitigate potential bias from misspecification in either the treatment or outcome models while adjusting for selection effects. Novel to the breastfeeding literature, our approach incorporates several supervised machine learning (ML) algorithms to improve propensity score estimates. We demonstrate that the gradient boosting machine algorithm minimizes prediction errors more effectively compared to logit, probit, and other ML algorithms. We find a robust link between having been breastfed and several improved cognitive outcomes during early childhood. In contrast, evidence of effects on non-cognitive outcomes is more limited. Our heterogeneity analyses provide further policy-relevant insights into the differential effects of breastfeeding, showing greater benefits for minorities and girls, and minimal marginal benefits from breastfeeding duration beyond 12 months.