Partial Effects in Probit and Logit Models with a Triple Dummy-Variable Interaction Term

Abstract

In nonlinear regression models, such as probit or logit models, coefficients cannot be interpreted as partial effects. The partial effects are usually nonlinear combinations of all regressors and regression coefficients of the model. We derive the partial effects in such models with a triple dummy-variable interaction term. The formulas derived here are implemented in the Stata inteff3 command. The command also applies the delta method to compute the standard errors of the partial effects. We illustrate the use of the command with an empirical application, analyzing how the gender gap in labor-market participation is affected by the presence of children and a university degree. We find that the presence of children increases the gender gap in labor-market participation but that this increase is smaller for more highly educated individuals.