Differential Prediction for Non-Random Subgroups

Abstract

A variety of models for the establishment of a prediction system (Roudabush, 1966; Bashaw, 1965, 1963; Lindquist, 1963; Tucker, 1963; Willingham, 1963; Bloom and Peters, 1961; Clemans, 1960; Toops, 1933), for the treatment of missing data (Horst, 1967; Johnson, 1959; Meredith, 1958), and for the selection and weighting of test predictor batteries for multiple criteria (Horst, 1955; Horst, 1954) have been reported in the psychological and educational research literature. These models have been established to cope with specific problems. Since, in practice, the investigator typically encounters a constellation of such problems, an integration of models is needed. To date, however, such a synthesis is lacking. Several problems pertaining to test selection and weighting were encountered in the development of the SRA Vocational Planning Inventory Prediction Programs (VPI), a prediction system for use in vocational/technical education.