Prediction of college performance for freshmen at Virginia Polytechnic Institute

Abstract

The thesis reports an investigation of 1060 freshman admissions to Virginia Polytechnic Institute in fall, 1961. Multiple regression methods were used to produce equations linking college performance with high school rank and College Board verbal and mathematical scores. Analyses were done for males and females, civilians and cadets. The three predictors accounted for only 34 percent of the variation in first year QCA among males. High school rank contributed most to the prediction; verbal score was found to be a very poor predictor. Predictions were more reliable for accumulative year performance than for first quarter only. Prediction error is discussed and expectancy tables constructed. Actual and predicted quality credit averages were used to investigate subgroups of the sample. Differences were found between civilians and cadets, Virginians and Out Of-Staters. Cooperative, drop-out, and trial summer school groups all had distinct characteristics. Curriculum groups differed widely in actual performance, but in most cases, corresponded to prediction. Exceptions are discussed. The same data was subjected to discriminant analysis, using two extra variables, civilian or cadet status, and Virginia or Out Of-State school. The function divided students by a predicted QCA of 1.000, with 28 percent probability of misclassification. Rank alone was found to be more effective in prediction than verbal and mathematical scores combined. Results of investigation into subgroups using discriminant predictions agreed with regression findings, with different exceptions to curriculum non-significance. Prediction is advised by regression rather than by discriminant analysis, but the present results are not recommended for future application, for reasons discussed.