The Causal Effects of Education on Earnings and Health

This paper defines and estimates returns to education for earnings, health, and smoking using a dynamic model of educational choice. We synthesize approaches in the structural dynamic discrete choice literature with approaches used in the reduced form treatment effect literature. We offer an empirically robust middle ground between the two approaches to estimate economically interpretable and policy-relevant treatment effects that recognize the dynamic nature of schooling decisions. Ability bias is a major component of observed educational differentials for most outcomes we study. Both cognitive and non-cognitive abilities play important roles in explaining observed differences in earnings, health, and smoking across all education levels. Nonetheless, after controlling for their influence, there are substantial causal effects of education at all stages of schooling. Continuation values associated with dynamic sequential schooling decisions are empirically important components of estimated returns to schooling for high-ability individuals. There is considerable heterogeneity in the effects of schooling on outcomes at different schooling levels and across persons. The estimated causal effects of education vary with the level of cognitive and non-cognitive endowments. Graduating college is not a wise strategy for all. It has no benefit for low-ability people. Our model allows for the possibility of myopic behavior and time inconsistency in choices. Individuals do not always act on decision-relevant, publicly available information. However, there are strong sorting gains consistent with pursuit of comparative advantage, especially for wage outcomes, but only at higher levels of education. We use our estimated model to evaluate two policy proposals.