Second-order corrected likelihood for nonlinear panel models with fixed effects

We propose a second-order correction for nonlinear fixed-effect panel models. The is made via the log-likelihood function. It removes two leading terms of bias that arises from estimating fixed effects. Maximizing corrected likelihood gives bias-corrected estimator, with O T − 3 , where number time periods. also test statistics. applies to general models independent observations. properties are confirmed in simulations binary-choice