A modified model for test validation and selection research

Abstract

It is argued that the clasisc prediction model is grossly oversimplified and has resulted in corresponding oversimplifications in the design of most validation studies. A modified and more complex prediction model is presented. Implications for future validation research are discussed in the context of the kinds of behaviors to be predicted, the necessity for investigating heterocedastic and nonlinear relationships, and the important advantages in prediction which may be realized by discovering homogeneous subsets of jobs, tests, people, and behavior within which prediction equations may be developed and crossvalidated. (32 ref.) (PsycInfo Database Record (c) 2025 APA, all rights reserved)