Using Predictions and Marginal Effects to Compare Groups in Regression Models for Binary Outcomes