THE PREDICTIVE VALIDITY OF THE SCHOLASTIC APTITUDE TEST FOR DISABLED STUDENTS

Abstract

ABSTRACTIn two separate rounds of data collection, validity data on disabled students were obtained from 145 institutions with validity data on nonhandicapped students. First year grade point averages (FYAs) were obtained for almost 1,000 disabled students who had taken special test administrations of the SAT with extra time and for more than 650 disabled students who had taken standard test administrations. Empirical Bayes procedures were used in conjunction with the sample of nonhandicapped students to develop separate regression equations for each of the 145 institutions. The focus of this study was whether regression equations based on data from nonhandicapped students predict the performance of handicapped students as well as performance of the nonhandicapped.Consistent with findings from other reports in this series the SAT performance of visually impaired and physically handicapped people was not very different from that of the nonhandicapped students. The SAT scores of learning‐disabled students were considerably lower and those of hearing‐impaired students even lower.A pattern of over‐ and underprediction was evident in FYA predictions based on high school grades alone. Disabled students earning the lowest high school grade point averages tended to be underpredicted–i.e. predicted to earn FYAs lower than their actual FYAs–while disabled students earning the highest high school grades tended to be overpredicted.A second pattern emerged from predictions based on SAT scores alone. Except for hearing‐impaired students, SAT scores from special test administrations have a strong tendency to overpredict the college performance of students with disabilities. The effect is strongest for those with learning disabilities.Using both high school grades and SAT scores to predict the college performance of students from special test administrations results in good overall predictions, but only because overprediction in some areas is offset by underprediction in others. The overprediction of the strongest third of the candidates is balanced by the underprediction of the weakest third.