Evaluation of Small-sample Compromised Randomization: Long-term Effects of Early Childhood Intervention on Health and Addictive Behavior

Barros contribution to the literature of policy evaluation stems from advances in statistical methods applied to rigorous empirical analysis. Another central question of Barros research is the analysis of eficient policies for reducing economic inequality. In the pursue this question, Barros line of work has recently shifted towards the study of early childhood investment as a tool for promoting economic growth and improving the odds of children born in disadvantage families (Barros et al., 2011; Barros and Olinto, 2008). We follow Barros steps by developing a formal statistical evaluation of the Perry pre-school program, the oldest and most cited early childhood experiment in US. Evaluating the Perry program poses three statistical challenges. First, the small sample size cast doubts on standard inference methods that are based on asymptotic assumptions. Second, randomization compromises call into question the validity of the simple statistical procedures often applied to social experiments. Third, the large number of outcome variables gives rise to the danger of selectively reported signi cant estimates. We develop a statistical method that accounts for all these problems and is tailored to the problems we face in the Perry intervention. We focus on long-term impact on health and addictive behavior variables. We use new data at age 40 never analyzed before. We find that treated females have less negative effects of drug/alcohool usage in a range of later-life activities. For males, we find that treated participants use less hard drugs, such as heroine and hashish.