Overview

Multiple logistic regression is a direct extension of simple logistic regression. A logistic (S-shaped) function is used to predict a categorical variable from information provided by two or more predictor variables. As is true for multiple linear regression, it is common practice to generate the model (solution) by entering all the variables in a single step; this is sometimes called the standard or simultaneous method. However, other methods call for entering (or entering and then removing) the variables in stages or steps; there are many ways to accomplish this, and SAS Enterprise Guide provides several on a drop-down menu. We will focus here on the standard method predicting a binary dependent variable.

Everything we said in Chapter 18 regarding simple logistic regression is applicable here. One noteworthy feature of the analysis concerns the coding of binary predictor variables.

Coding of binary predictor variables

In Section 18.3.1 of the previous chapter, we discussed coding the dichotomous dependent variable. The default coding scheme used by SAS Enterprise Guide presumes that for the dependent variable the group we wish to use as the reference group is coded as 0 and that the comparison group to which we want to compare the reference group is coded as 1.

The coding scheme for the predictor binary variables has to be just the reverse of the scheme used for the dependent variable.