Developing New Test Selection and Weight Stabilization Techniques for Designing Classification Efficient Composites.

Abstract : The major goal of this research was to specify a classification-efficient methodology for the construction of assignment composites of optimally selected and weighted tests drawn from a single battery of ASVAB and experimental tests and targeting a job family. The experiments examine the effects of the number of tests included in a composite, using different figures of merit as the standard for the selection of tests for components and stabilizing test regression weights. The research approach adopted involves a simulation of the Army selection and classification process using Project A validity data. Comparisons of classification efficiency obtained under each experimental condition are reported in terms of mean predicted performance (MPP). Findings indicate that five-test composites, tailored to operational job families and selected by a predictive validity index to provide positive weights, can provide an acceptable approximation of the maximum obtainable MPP. (MM)