Standardized Discriminant Coefficients Revisited

Abstract

The reasons and methods for standardizing discriminant function coefficients are reviewed, and the recommendation of Mueller and Cozad (1988) to use the pooled within-groups variance estimate is reexamined. It is demonstrated that using either total or pooled within-groups variance estimates to standardize discriminant function coefficients can be justified on purely mathematical grounds. We argue that using total variance estimates to standardize discriminant function coefficients is a more consistent and parsimonious approach than using pooled within-group estimates.