Using Orthogonal Regression Analysis to Investigate Differential Prediction: New Insights on Predicting College Performance

• In Event: Topics in Validity Evaluation

Abstract

The standard approach for investigating differential validity is to examine the magnitude of correlation (validity) coefficients, while ordinary least-squares regression analysis is used to study differential prediction. In our previous study, orthogonal regression was used to investigate differential prediction for African American college students. Our findings indicated that the true-score relationships between the predictors, SATV and SATM, and the criterion of first-year college GPA are, in fact, highly similar for African American and White students. In this study, we expanded the use of orthogonal regression in three ways: (1) We conducted analyses of Hispanic students; (2) We expanded our outcome measures to include a 5-year cumulative GPA; (3) We will cross-validate our results using data from four other large universities.

Authors

• John W. Young, International Baccalaureate

• Michael T. Kane, ETS

• Lora F. Monfils, Educational Testing Service

• Chen Li, ETS

• Chelsea Ezzo, ETS