In a timely, comprehensive article in this journal, Joy and Tollefson (J & T hereafter) treated design and interpretation problems for linear multiple discriminant analysis (LMDA). Although the article is generally correct in treating a complex topic, it has two problems:
1. The (widely-used) split-sample validation technique that J & T used is an inappropriate method for establishing the efficacy of the estimated discriminant function.
2. J & T reach erroneous conclusions on the correctness of two methods for assessing the discriminatory power of individual variables.
This comment addresses these two problems.