-Discriminant analysis is known to be a more robust and useful procedure than our friend Dr Cadoret and his associate seem to appreciate. We welcome the opportunity to correct their misapprehension. Unfortunately, the literature comparing logistic regression and discriminant analysis seems to be written in a form inaccessible to many clinical scientists, so it is useful to clarify here the purposes and situations for which the methods are appropriate.
First, Drs Cadoret and Brier mistakenly claim that logistic regression is superior to discriminant analysis for determining which factors affect the probability of a response. Discriminant analysis is derived strictly if all predictor variables have a multivariate normal distribution, but this requirement need not be strongly satisfied for purposes of factor selection or classification. More specifically, significance tests for selection of relevant risk factors by discriminant analysis are not sensitive to violations of the normality assumption and in practice are equivalent