Comparing groups in binary regression models using predictions∗

Methods for group comparisons using predicted probabilities and marginal effects on probabilities are developed for the binary regression model. Unlike standard tests that compare regression coefficients across groups, the tests we propose are unaffected by the identification of the coefficients and are expressed in the natural metric of the outcome probability. While we focus on the logit model with two groups, our interpretive framework applies to a broad class of models and can be extended to any number of groups. ∗We thank Long Doan, Trent Mize, and Rich Williams for their comments. †[email protected], Departments of Sociology & Statistics, Indiana University, Bloomington, IN 47401. ‡[email protected]; Department of Sociology, University of Notre Dame, Notre Dame, IN 46556.