Predicting College Performance From Test Scores: Investigating Differential Prediction for Disadvantaged Students and Schools

• In Event: Roundtable Session 3
  In Roundtable Session: Understanding High School Effects on Higher Education Access and Success

Abstract

Little is known about the value of high-stakes high school tests in predicting performance in college. Using data from the City University of New York, I examine how well Regents exam scores predict freshman year grade point average for low-income and minority students, and how this compares to SAT scores. Using OLS regression, I find that the Regents tests are of similar predictive value as the SAT for low-income students, Black students and Hispanic students, with comparable patterns of differential prediction and validity for Black and Hispanic students. I then use two-level mixed models to identify whether the differential prediction is within or between schools, and I find significant differential prediction for Black and Hispanic students at the school level.

Author

• Preeya Pandya Mbekeani, Harvard University