This paper explores the use of dynamic panel data models and sibling-based estimation procedures for evaluating the effects of educational interventions aimed at children or youth. It specifies statistical models of the schooling decision that are consistent with dynamic behavioral models and that allow for unobservable heterogeneity. The frameworks presented provide a way of directly assessing short-term program impacts and of simulating longer term impacts that may extend beyond the program exposure times observed in the data. This paper also proposes a new estimation approach for the model, which combines matching, differencing and instrumental variables in a way that minimizes the need for parametric modeling assumptions. The approach uses retrospective data on older siblings’ schooling histories to identify how an educational intervention affects younger siblings. We apply the proposed method as well as other methods to study the effects of the Mexican Oportunidades conditional cash transfer program on schooling outcomes of children and youth living in urban areas. The analysis samples come from the 2002-2004 ENCELURB evaluation datasets. The empirical results show significant short-term impacts of the program on school enrollment. Simulations based on the dynamic model indicate that long-term exposure to the program increases educational attainment by about half a year.
Available at: http://works.bepress.com/susan_parker/4/