Doing Discriminant Analysis: Some Problems and Solutions.

Recent years have witnessed the growing use by political scientists of multlvarlate techniques designed expressly for the analysis of non-Interval dependent variables. One of these techniques, discriminant analysis, has grown In popu larity because: (1) It Is Ideally suited to a data config uration commonly encountered In political science; (2) It Is similar In many respects to the familiar multiple regression model; (3) It has been made accessible to political scientists by a concise, user-oriented monograph (Klecka, 1980a); and (4) It Is available In standard statistical software packages. As Its use has grown more widespread, discriminant anal ysis has become the source of a certain amount of confusion and uncertainty, typically centering on what choices one should make In setting up a discriminant analysis and how one should Interpret discriminant results once the analysis has been completed. This paper presents some workable solu tlons to these problems. A brief overview of dlsclmlnant analysis will facilitate discussion of these problems and solutIons.