Predicting Individual Effects in Fixed Effects Panel Probit Models

Abstract Many applied settings in empirical economics require estimation of a large number individual effects, like teacher effects or location effects; health economics, prominent examples include patient doctor hospital effects. Increasingly, these are the object interest estimation, and predicted often used for further descriptive regression analyses. To avoid imposing distributional assumptions on they typically estimated via fixed methods. In short panels, conventional maximum likelihood estimator binary response models provides poor estimates since finite sample bias is substantial. We present bias-reduced that better by removing first-order asymptotic bias. An additional, practical advantage it predictions all sample, including those which corresponding dependent variable has identical outcomes time periods over (either zeros ones); these, prediction infinite. illustrate approach simulation experiments an application to care utilization.