A Method for Choosing Weights to Predict College Grades for Admission Decisions and to Assess their Fairness by Race/Ethnicity

Abstract

Previous research suggests that equal weights tend to outperform statistically optimal weights in cross-validation studies. This paper argues that the findings from the equal weights literature are relevant for researchers that predict college grades and/or assess differential prediction of college grades by student characteristics. An application of the criterion profile methodology (CPM) is presented to demonstrate how to examine individual criterion profiles. This study showed how to use the CPM to determine the extent to which equal and statistically optimal coefficients differentially predicted college grades for minority and majority students. The results support previous findings, in that, 92.5% of the explained variance in college grades was attributed to equal weights, where standardized test scores and high school rank were weighted equally, and 7.5% of the explained variance was accounted for by statistically optimal coefficients that weighted ACT Math scores less than ACT English and high school rank. Additionally, equally weighting admission information was more accurate for predicting Asian Americans’ future academic performance than European Americans.