Using logit coefficients to obtain the effects of independent variables on changes in probabilities

Abstract

Unnecessarily awkward procedures for interpreting logit coefficients in terms of odds-ratios or ratios of odds-ratios appear in sociological research. Because odds-ratios are not linearly related to probabilities, statements about relative odds-ratios do not identify the effects of group membership as measured by dummy independent variables on a dichotomous dependent variable. Procedures from econometrics allow calculating the probabilities of being in the category of interest of a dichotomous dependent variable. The effects of dummy independent variables equal the differences in probabilities among the different groups. For continuous independent variables, the maximum possible effect on changes in probability is always one-quarter of the value of the logit coefficient. © 1991 The University of North Carolina Press.