Logistic regression - effect of unobserved heterogeneity on estimators bias variance

Abstract

In this paper we study unobserved heterogeneity in logistic regression, which occurs as a result of omitted variables. Unlike linear regression, logistic regression estimates are a ected by model misspeci cation even if omitted variables are not correlated to the explanatory variables in the model. As a result, interpretation of log-odds ratios and odds ratios is not straight forward and similar models with di erent independent variables can not be compared. We study average marginal e ect as a possible measure of overcoming the unobserved heterogeneity problem.