ruschenpohler/worms-doubleml

Project examining heterogeneous treatment effects of school-based deworming using double/debiased ML on the ~10-year follow-up data on the Primary School Deworming Project in rural Kenya (Miguel & Kremer, 2004) # Who Benefits From Childhood Deworming? Heterogeneous Treatment Effects by Age and Gender on Long-Run Human Capital and Labor Market Outcomes Note: This project was implemented using the ML Platform for Trustworthy Human-Agent Research Production. --- ## Abstract School-based deworming programs have reached hundreds of millions of children across sub-Saharan Africa and South Asia, scaling up from a landmark randomized trial in Kenya. Whether benefits from these programs are concentrated among specific groups (younger children who were treated during more critical developmental windows, or girls who face steeper labor market barriers) has direct implications for program design. We apply double/debiased machine learning (DoubleML) and causal forest estimation to the Kenya Life Panel Survey follow-up of the Miguel and Kremer (2004) deworming trial, estimating long-run effects of childhood deworming on adult BMI, underweight prevalence, educational attainment, and employment, and characterizing heterogeneity by age at treatment and gender. The developmental-window hypothesis is not confirmed by formal inference: no estimated heterogeneity survives Romano-Wolf multiple testing correction (minimum adjusted p = 0.29). We show that the null is unlikely to be an artifact of attrition or methodological choices. The design is, however, underpowered for plausible nutritional effect sizes at the 20-year horizon. Against this primary null, we document three directionally consistent but fragile findings that motivate further investigation at scale. --- ## Background and motivation The Primary School Deworming Project (PSDP), conducted in western Kenya between 1998 and 2001, randomly assigned 75 primary schools to receive deworming at different times. Miguel and Kremer (2004) found substantial reductions in school absenteeism. Hamory et al. (2021) followed the same cohort two decades later and found significant gains in earnings and consumption. Evidence from this trial inf …