Heterogeneity and Selection in Dynamic Panel Data

This paper shows nonparametric identification of dynamic panel data models with nonseparable heterogeneity and dynamic selection by nonparametrically differencing out these two sources of bias. For T = 3, the model is identified by using a proxy variable. For T = 6, the three additional periods construct the proxy to establish identification. As a consequence of these identification results, a constrained maximum likelihood criterion follows, which corrects for selection and allows for one-step estimation. Applying this method, I investigate whether SES affects adult mortality. The method circumvents the survivorship bias and accounts for unobserved heterogeneity. I find that employment has protective effects on survival for the male adults in the NLS Original Cohorts: Older Men.