Accession Number : AD0673967

Title :   A COMPARISON OF FOUR MODELS FOR MAKING PREDICTIONS ACROSS INSTITUTIONS,

Corporate Author : WASHINGTON UNIV SEATTLE DEPT OF PSYCHOLOGY

Personal Author(s) : Cory,Charles Henry

Report Date : AUG 1968

Pagination or Media Count : 98

Abstract : The usefulness of prediction equations has been severely attenuated in the past whenever measurement, either of the predictors or of the criterion performance itself, has been subject to systematic variation related not to differences among individuals for whom predictions are made but rather to differences among sources of those measures. The present research has adapted three models for correcting source variations to an empirical test, the first attempt to make such evaluative comparisons. The empirical test of the models involved criterion data from six different community colleges and predictor data classified according to the time at which data were collected, either in high school or at college enrollment. The empirical study replicated across seven criterion subject matter areas. Data for one-half the subjects were used to develop prediction equations under the several models and data for the remaining half to cross-validate each of the prediction equations. On cross-validation, only one model, the within subgroups model, yielded improved predictions across the range of criterion activities. The finding that the model most sensitive to these differences is the one simplest to apply both in the development and utilization of the prediction equation is an important one. (Author)

Descriptors :   (*ACHIEVEMENT TESTS, *CORRELATION TECHNIQUES), LEARNING, PERFORMANCE(HUMAN), UNIVERSITIES, MATHEMATICAL PREDICTION, MATHEMATICAL MODELS, REGRESSION ANALYSIS, SAMPLING, STUDENTS, THESES

Subject Categories : Psychology

Distribution Statement : APPROVED FOR PUBLIC RELEASE