Initial vs. retest GRE performance: A study of one million Graduate Record Examination test-retest observations

A large test preparation industry has arisen around the popular idea that students can dramatically improve their scores. However, can students greatly improve their scores when they test more than once? Understanding this variation in retesting helps testing reliability of measures, expectations of validity and prediction and also providing bounds and realistic expectations within which test preparation is potentially impactful. Using a sample (N = 689,063) of Graduate Record Examination test takers with at least two GRE general test observations between 2015 and 2020, we identified 978,334 test-retest pairs and analyzed the relationship between initial and retest scores within the three subtests. Initial and retest scores were highly correlated (0.91 for Quantitative, 0.82 for Verbal, and 0.78 for Analytical Writing), and retest scores were 1.43 points higher for GRE-Q and 1.49 points higher for GRE-V on average but with wide variation. We did not find evidence that test score changes varied meaningfully by length of time between test observations. Finally, we found evidence of regression to the mean, with test takers with initial scores in the top or bottom bracket scoring closer to the mean on retest. These results provide evidence of the reliability of the GRE over time as a test of academic skill.