Optimum cutting scores to discriminate groups of unequal size and variance

Abstract

The accuracy with which a test classifies people, objects, or events as belonging to 1 of 2 groups depends upon distance between the means, relative variability, and relative size of the 2 groups. An analytical method is presented for determining the optimal cutting score when estimates of these parameters are available and when it can be assumed that the test scores are normally distributed for each of the 2 groups. It is shown that many situations exist in which "valid" tests cannot improve upon base-rate predictions. Tables are provided for a rapid determination of the optimal cutting score for a given condition; these tables also indicate the conditions under which base-rate predictions should be made and the proportion of erroneous decisions to be expected when the optimum strategy is used. (PsycInfo Database Record (c) 2025 APA, all rights reserved)