ABSTRACT

In recent chapters we examined various regression models-these allow us to examine the relationship between one or more predictors when the outcome is continuous (ordinary least squares regression) or categorical (logistic regression). This was followed by a discussion of multivariate analysis of variance procedures that allow for multivariate examination of mean differences. In this chapter, we are introduced to discriminant analysis, a procedure developed by Fisher (1936) that provides for classification of groups when the outcome is categorical, the same goal as that of logistic regression but, as we’ll see, goes a bit beyond what logistic regression allows us to do. This procedure can be helpful in situations where classification into groups for purposes of intervention, training, receipt of something (e.g., monetary loan), and similar scenarios is needed or wanted. Cluster analysis, discussed in the next chapter, is similar in that it allows cases to be classified based on a set of variables. However, with discriminant analysis, the researcher knows a priori the mutually exclusive groups within which he or she wants to classify (i.e., the categorical dependent variable). In comparison, with cluster analysis, there are no such a priori groups into which the independent variables will be classified.