Comparing Effects in Dichotomous Logistic Regression: A Variety of Standardized Coefficients

Objective. The purpose of this study is to present and develop a variety of "standardized" coefficients for use with dichotomous logistic regression. They can be used to determine the relative size of the effects of different independent variables and to make judgments about the absolute strength of the relationship. Methods. The development of these coefficients draws on the idea from OLS regression of standardizing coefficients by using a "comparable" metric for each variable; i.e., comparing predicted changes in the dependent variable that correspond to a standard amount of change in the independent variables. These coefficients are semi-standardized because only the independent variables' scales have been standardized; one of these measures change in the natural logarithm of the odds of the dependent variable whereas the other two measure changes in the probability of the dependent variable. The use and interpretation of these coefficients is illustrated using data from the 1987 NORC General Social Survey to predict opinion (approval versus disapproval) about laws forbidding interracial marriage. Conclusions. Even though all the coefficients have interpretations that should be accessible to a general audience, the use of the semi-standardized coefficient measuring the change in the predicted probability of the outcome is preferable because it is intuitively appealing and is bounded in the interval [— 1, + 1]. Its bounded nature makes interpretation of its magnitude easier.