CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0905
Topic: Title: Estimating logit models with many fixed effects Authors:  Amrei Luise Stammann - Heinrich-Heine University Duesseldorf (Germany) [presenting]
Daniel McFadden - University of California (United States)
Florian Heiss - Heinrich-Heine University Duesseldorf (Germany)
Abstract: For the panel data analysis of binary dependent variables, the fixed effect logit model is a popular specification. The conditional fixed effects logit (CFL) estimator has the drawback that it does not deliver estimates of the fixed effects or marginal effects. It is also computationally costly if the number of observations per individual $T$ is large. The dummy variable logit (DVL) estimator is a simple logit estimator with a dummy variable for each individual. It suffers from the incidental parameters problem which causes severe biases for small $T$. Another problem is that with a large number of individuals $N$, the computational costs can be prohibitive. We suggest a pseudo-demeaning algorithm that delivers the same results as the DVL estimator without its computational burden for large $N$. It uses the sparsity of the Hessian and the special features of the logit model. We also discuss how to correct for the incidental parameters bias of parameters and marginal effects. Monte Carlo evidence suggests that the bias-corrected estimator has similar properties as the CFL estimator in terms of parameter estimation. Its computational burden is much lower than the CFL or the DVL estimators, especially with large $N$ and/or $T$.