Partial effects estimation for fixed-effects logit panel data models

We propose a multiple step procedure to estimate Average Partial Effects (APE) in fixed-effects panel logit models. Because the incidental parameters problem plagues the APEs via both the inconsistent estimates of the slope and individual parameters, we reduce the bias by evaluating the APEs at a fixed-T consistent estimator for the slope coefficients and at a bias corrected estimator for the unobserved heterogeneity. The proposed estimator has bias of order O(T −2 ) as n → ∞ and performs well in finite sample, even when n is much larger than T . We provide a real data application based on the labor supply of married women.