The unequal group size problem in discriminant analysis

The technique of multiple discriminant analysis has seen widespread use in marketing studies. In assessing the predictive efficiency of the derived discriminant functions, the proportional chance criterion is frequently used. This criterion is adequate except in situations when the classification groups are of greatly unequal size. The purpose of this article is to illustrate a procedure which adjusts the groups' sizes so as to allow effective use of the proportional chance criterion under a wider range of circumstances. The article presents an actual problem situation experienced by the author and the procedure used to overcome it.