The quartile difference method of item selection

Abstract

An item selection method for use with categorical data, designed to reduce shrinkage in validity due to error variance as items are added to a selection battery, is outlined and demonstrated. Weights are established for each item and a chi-square is calculated (criterion categories vs. item categories). Using these weights, the items are scored for a hold-out group, and a battery is selected beginning with the item with the largest chi-square value. The subset is selected that gives, for the high criterion hold-out group, the maximum difference between the percents in the top and bottom quarters of the total hold-out group distribution. (PsycInfo Database Record (c) 2025 APA, all rights reserved)