Use of Linear Models vs. Multiple Discriminant Analysis with Three Groups

Abstract

It is well known that the three-group discriminant function cannot be expressed as a specialized case of the general linear model or multiple linear regression. However, researchers should be alert to the possibility that the set of three-group membership vectors might be adequately represented by an unidimensional bipolar variable of three points thus permitting the use of regression techniques.
A research example is presented in which data were examined both on the basis of the three-group multiple discriminant function as well as by regression procedures. Results of the comparative analysis were such that regression techniques furnished an accurate picture of the findings. The obvious implication is to suggest that researchers consider using a three-point dependent variable and regression techniques when it makes theoretical or logical sense to conceptualize the three-group membership vectors as a single variable.