Discriminant Analysis

It has been increasingly realized that (1) multivariate methods are essential in most quantitative studies (Fish, 1988; Thompson, 1992), and (2) all conventional parametric analytic methods are correlational and invoke least squares weights (e.g., the beta weights in regression) (Knapp, 1978; Thompson, 1991). The present paper reviews one very popular multivariate analytic method that explicitly invokes weighting to optimize one criterion: the analytic method that researchers have come to call predictive discriminant analysis (Huberty and Barton, 1989; Huberty and Wisenbaker, 1992). Predictive discriminant analysis (PDA) is differentiated from descriptive discriminant analysis (DDA) (Dolenz, 1993) by a focus on predicting membership in intact groups. The paper is intended as a primer introducing researchers to the distinction between PDA and DDA and explaining analytic issues related to the PDA application (Van Epps, 1987). For example, sdding predictors can actually result in worse prediction in PDA, though in no other analytic methods can adding predictors result in worse effects. Methods for evaluating predictor variable importance using leave-one-out (L-0-0) strategies are also explored. Included are two tables and one figure. (Contains 17 references.) (Author/SLD) * Reproductions supplied by EDRS are the best that can be made from the original document. kr