The Utility of Homogeneous Subgroups and Individual Information in Prediction

Multivariate Behav Res. 1975 Oct 1;10(4):449-61. doi: 10.1207/s15327906mbr1004_4.

Abstract

A number of behavioral research studies have employed subgroups of subjects for prediction. In using subgroups for predictive purposes, a key assumption is often overlooked, i.e., at least one of the grouping or predictor dimensions affects the relationships between the other grouping dimensions and the criteria. The purposes of the present study were (a) to emphasize this commonly overlooked assumption and (b) to examine empirically the utility of subgroup versus individual information in prediction. Data collected on 509 freshmen in 32 life history subgroups (19 male and 13 female subgroups) were used to evaluate the utility of subgroup information in prediction. Of 24 criteria (college experience data, e.g., academic achievement) predicted by subgroup information, predictive efficiency of four criterion measures was enhanced when subgroup information was added to individual information. However, for the vast majority of the criterion measures, predictive efficiency would have been lost had only subgroup information been used in prediction. It is recommended that researchers carefully evaluate the effects on predictive efficiency when using subgroups rather than individual information in prediction.