ABSTRACT
This chapter describes methods to assess mediation when the dependent variable is categorical. It also describes logistic and probit regression, the appropriate regression methods for categorical dependent variables. Logistic regression coefficients are in the metric of the natural logarithm of the odds ratio, that is, the logit. In fact, the probit regression coefficients are approximately equal to 0.625 times the logistic regression coefficients. There are several aspects of probit regression that make it preferable for mediation analysis. The SAS statements and output were used to obtain the logistic regression coefficient estimates used to compute the mediated effect and its standard error. Standardizing coefficients and standard errors can be used to appropriately model the different scales across the two logistic regressions. Models for continuous variables can also be evaluated when the dependent variable is categorical. The probit is used to model the relationship between the observed categorical variable and the latent normally disturbed variable.
