Differential Predictability in a Selection Model for Graduate Students: Implications for Validity Testing

Abstract

The purpose of the paper was to present a method for testing differential predictability in a selection model. The method was illustrated on data from 227 graduate students. Regression models for males and females were developed based upon 103 male and 24 female graduates. The models were then tested for equivalence of variance, intercept, and slope. Since the models were not homogeneous in the variance, a method for adjusting for heterogeneity was presented. It was concluded that differential predictability was not present. The resulting common regression model was subsequently cross validated to the remaining 100 subjects.