An investigation of ethnic and gender intercept bias in the SAT's prediction of college freshman academic performance
Abstract
Intercept bias in the SAT’s prediction of college freshman academic performance has long been found in studies evaluating the academic prediction of men and women, and various ethnic groups. Typically the college GPAs of women are underpredicted relative to men by approximately 2/10th of a standard deviation in GPA. In multicultural samples, relative to Whites, Asian American freshman performance is often underpredicted by a similar amount, and the performance of Blacks and Hispanics is overpredicted. The hypothesis of the study presented herein was that this differential prediction can be explained by a set of behavioral and attitudinal variables; i.e., that the inclusion of such variables into the prediction equation alongside SAT would result in the diminishment or elimination of observed differential prediction. It was also hypothesized that use of a GPA adjusted for differences in grading standards that exist across college disciplines and courses would contribute to the reduction of differential prediction. SAT scores, high school performance data, college grades and survey responses were obtained from a diverse sample of over nine hundred students at a large public university in the American Southwest. The data indicated that Asian American performance was not underpredicted by the SAT in this sample, although as anticipated, female performance was underpredicted and MexicanAmerican performance was overpredicted. Inadequate sample size precluded the analysis of African American students. In relation to the hypothesis, results were mixed. Although inclusion of the additional variables and the use of the adjusted GPA criterion did not significantly reduce Mexican-American overprediction, it reduced the underprediction of women’s performance to insignificance.