Average Partial Effects in Nonseparable Panel Data Models: Identification and Testing

• Departmental Seminar

Date and Location

Friday, January 18, 2013, 10:30 AM - 12:00 PM
3001 Plant and Environmental Sciences

Abstract

In many empirical settings involving microeconomic panel data, the researcher's objective is to identify the average partial effect of a variable on an outcome of interest. Thisi paper examines nonparametric identification of average partial effects in this setting and proposes tests of identifying assumptions that do not rely on functional-form restrictions. The paper first studies the nonparametric identification problem starting from a data-generating process that exhibits both individual and time heterogeneity. The trade-off between identifying assumptions that restrict individual and/or time heterogeneity is formally characterized. The paper then proposes a menu of identifying assumptions that the empirical researcher may choose from. To test the identifying assumptions, bootstrap-adjusted Kolmogorov-Smirnov and Cramer-von-Mises statistics are proposed and are shown to be asymptotically valid. The tests include a nonparametric test of the fixed effects assumption, which is applied to the human capital earnings function using a subsample of the national longitudinal survey of youth, previously used in Angrist and Newey(1991).