Predicting Fixed Effects in Panel Probit Models

Abstract

Many applied settings in empirical economics require estimation of a large number of fixed effects, like teacher effects or location effects. In the context of binary response variables, previous studies have been limited to the linear probability model, citing perfect prediction and incidental parameter biases as reasons. We explain why these problems arise and present an appropriate solution for the probit model. In contrast to other estimators, it ensures that predicted fixed effects exist for all units. We illustrate the approach in simulation experiments and an application to health care utilization.

Keywords: Perfect Prediction, Incidental Parameter Bias, Fixed Effects, Panel Data, Binary Response