Casey W. Miller
"Factors that can influence performance on the GRE general test 2006-2007"
The development of the APS Bridge Program (APS-BP) provides our community with a great opportunity to help physics shed the notoriety of being the least diverse of the sciences. We can all play a role in this effort, whether or not our individual programs are or will become affiliated with the APS-BP. In this article, I hope to raise awareness of some relatively simple but impactful means to enhance racial and gender diversity. What follows is hardly comprehensive, but hopefully suggests pragmatic next steps that can be widely and rapidly implemented.
There are about 180 physics programs listed in the AIP Graduate Programs book. The General GRE is required by 96%; a quarter of these have an explicitly stated minimum Quantitative GRE (QGRE) score for admission, with the median stated cut-off being 700 (64th–70th percentile, depending on year). As educators, we naturally expect exams to be meaningful. Most people believe this is the case for the GRE exams, and may thus prefer high scores. But analysis of the data often finds no significant correlation between long-term success and GRE scores (the stated predictive power is limited to first-year grades). In my time on the Graduate Council of the University of South Florida, I have seen few programs that have taken the time to correlate "success" with admissions materials. Of these, none, including my program, have concluded the GRE predicts success in research, which no one argues is the aim of the PhD. Our analysis finds that the QGRE correlates with only one metric, the graduate GPA (but it is such a weak correlation the scientist in me rebels when fitting it to a line). That said, we find undergraduate GPA to be a better predictor of graduate GPA. We also find that undergraduate GPA is correlated with all three sections of the General GRE.
So why use the GRE at all? One certain answer: national rankings. Consider US News, whose rankings of graduate programs are widely influential among both prospective graduate students and administrators. In the US News formula, the weight given to the mean GRE score is 12% (6% each to Verbal and Quantitative). The acceptance rate weighting of 6% is added to give "student selectivity" a total weight of 18%. This exceeds student/faculty ratio (4.5%), percent of faculty with awards (2.5%), and doctoral degrees granted (5%), and is far too close in my opinion to tangible productivity measures such as research expenditures (15% each for dollars per program and dollars per faculty member). This weighting can lead administrators to the simple conclusion that their rankings will go up if they only admit students with perfect GRE scores.
Justifying using the GRE becomes significantly more complicated, however, when the test results are dissected by race and gender. The figure plots QGRE scores by race/ethnicity and gender for US citizens whose intended graduate major was "physical sciences". The top and bottom of the lines are the 75th and 25th percentiles of the score distributions, respectively; the tick is the mean. This pattern is qualitatively unchanged when controlling for undergraduate GPA. Note the implications for diversity of using 700 as a minimum acceptable score: nearly three quarters of Hispanics would be rejected, and significantly more than this for American Indians, African Americans, and Puerto Ricans; similarly, women are filtered out at a higher rate than men. Mixing cut-off scores with these racial and gender disparities sets the foundation of a glass ceiling erected by the lopsided treatment of minorities and women before they even set foot in grad school.
The Asian > White > Hispanic > Black pattern permeates standardized testing: it is the same for the SAT, and is reflected in the recent race-based levels set by Florida and Virginia for grade schoolers' performance on state-wide standardized tests.
The GRE pattern, however, must be more complex in origin than a property tax distribution problem–these are students who have done so well in college that they consider themselves serious enough candidates for graduate studies to spend hundreds of dollars on the tests and score reporting. As for gender, there is not a single section of the GRE general test in which women outscore men, regardless of how the data are diced. Isn't the ubiquity of these patterns odd?
The results of Fig. 1 are not new, but are not trumpeted loudly or often enough to be well known among physicists. When a skeptical colleague asserts, "something is wrong with those data!" I point out that they come from ETS, the company that administers the GRE (nevertheless, even though I accept their validity, I wholeheartedly support the assertion that something is wrong). Others have noted the impact of the GRE on diversity in physics, hopefully not limited to:
Casey W. Miller is Associate Professor of Physics and Associate Director of Physics Graduate Programs at the University of South Florida. He is a recipient of the AFOSR Young Investigator and NSF CAREER Awards.
©1995 - 2023, AMERICAN PHYSICAL SOCIETY
APS encourages the redistribution of the materials included in this newspaper provided that attribution to the source is noted and the materials are not truncated or changed.