Three essays on the economics of child development

This dissertation focuses on three topics on the economics of child development that have not yet been deeply studied. The first two essays address the issues of attrition and selection biases that may affect estimates of the long-term effect of high-quality preschool programs in studies using non-experimental, longitudinal data. Both essays are based on the Chicago Longitudinal Study of the federally-funded Child-Parent Center early education program. The third essay uses the Peruvian sample of the Young Lives Study to estimate the effect of a mother's education on her child's development of cognitive skills in a developing country.;The second essay, "Long-Term Effects of Early Education in Chicago: Instrumental and Non-Instrumental Variables Approaches for Modeling Selection", investigates the strength of the Chicago Longitudinal Study quasi-experimental design by using several econometric methods to control for selection on observed and unobserved characteristics that could conceivably bias the estimates of the effects of the Child-Parent Center preschools.;The third essay of my dissertation exploits an exogenous change in policy to evaluate the effect of maternal education on child's cognitive skills in Peru. Theoretically, while there is a general agreement that maternal education leads to improvements in children's achievement and health, (Becker, 1981), the empirical findings are inconsistent. Some authors found a significant effect of maternal education on child achievement for developed countries (James-Burdumy, 2005; Leibowitz, 1977), while others found little or no evidence of a causal relationship for a developing country (Paxson & Schady, 2007). Similarly, some authors found a significant effect of maternal education on child health (Glewwe, 2001), while others found little or no evidence of a causal relationship (Desai & Alva, 1998).;The first essay of my dissertation addresses the issue of attrition in longitudinal studies. As valuable as longitudinal studies are to public policy decisions, nonrandom attrition presents a threat to accurate estimation of program effects regardless of research design. In program evaluation, one approach for dealing with attrition involves using instrumental variables. As has been discussed frequently in the topic of estimation of treatment effects, the lack of an obvious instrument or suitable exclusion restriction has led to uncertainty about the real effects of a program. Also, even when a potential instrument is found, it is difficult to know or demonstrate whether it is really exogenous. To control for nonrandom attrition, this study uses two parametric techniques and one semi-parametric technique. These methods of propensity score matching, propensity weighting and bounding methods, have proved to reduce biases associated with different selection processes and thus strengthened causal inference. While these frameworks have been used in selection correction of nonrandomized samples, they have not been used in nonrandom attrition correction.