Elimination of variates in linear discrimination problems.

The selection of a subset from a larger list of variates measured on each individual in a study may be an important part of the analysis of medical research data. A number of statistical methods have been used in applications as rules for selecting subsets. The purpose of this study was to compare three of these selection rules with random selection. In order to study variate selection in linear discriminant applications to medical research, sets of data were obtained from five medical studies. Two of the rules: 1) selection of variates using stepwise regression; and 2) selection using Studentized t differences in means, were observed to be better than random.