The SAT Predicts Success at Harvard. Does It at Berkeley?

20 Sep, 2026

The SAT predicts differently across institutions, and a third study suggests better prediction may change admissions less than the debate assumes.

Minesweeper, Sather Gate at the University of California, Berkeley, 2003. Wikimedia Commons, CC BY SA 3.0.

Two studies, a year apart, asked closely related questions about the SAT and got opposite answers. In March 2025, researchers publishing through the National Bureau of Economic Research looked at several Ivy Plus colleges and found that test scores predicted first year grades far better than high school GPA. A perfect 1600 predicted a first year GPA about 0.43 points higher than a 1200. A perfect 4.0 in high school predicted less than a tenth of a point more than a 3.2. A year later, a different team working with a decade of cohorts from a large urban public university system found almost the reverse: there, a one standard deviation increase in high school GPA was four to six times more predictive of six year graduation than the same increase in SAT scores. I don't think either is wrong. They measured different things, and together they point to an answer sharper than "it depends": the SAT helps where high school grades have stopped telling students apart. A third study, using a different test in graduate admissions, raises a separate question: how much better prediction actually changes admissions decisions.

Look at what changed between the studies. Not the test. The outcome, for one: first year grades at the Ivy Plus schools, six year graduation in the public system, and a measure can predict one far better than the other. The population, for another. The Ivy Plus authors say as much themselves. Tests seem to do better in more selective settings, possibly because grade inflation has pushed top applicants against the 4.0 ceiling, where GPA stops telling anyone apart. (That 4.0 versus 3.2 result is what a ceiling looks like.)

Then, in July 2026, a third paper made things more interesting. Johann Gaebler and three coauthors analyzed more than 13,000 applications to a large public policy master's program. GRE scores substantially improved predictions of first year grades. But when the authors used those predictions to pick an admitted class, the class got better by about 0.03 grade points. Better predictions mostly led to the same decisions, and where they didn't, the choice was usually between two similarly qualified applicants. Prediction is not decision. The GRE is not the SAT, and graduate admissions are not undergraduate admissions, but the study isolates a distinction the SAT debate often misses: a test can contain information, improve prediction, and still change surprisingly few decisions.

Berkeley's unfinished experiment

Berkeley makes this personal for me, because I went to college there. The University of California also runs one of the largest experiments in test free admissions, and it approached the question more carefully than the slogans on either side suggest. In 2018 UC's president asked the Academic Senate to review admissions testing, and an 18 member Standardized Testing Task Force formed in early 2019. Its January 2020 report did not call the SAT worthless. Quite the opposite. The task force found that at UC, scores predicted first year grades better than high school GPA did and were about as good at predicting retention and graduation. They added information beyond grades, in part, the report noted, because grade inflation had eroded the predictive power of high school GPA. The task force recommended keeping the tests for a period as UC worked on an assessment better aligned with what California students were expected to learn. Six members filed an additional statement favoring a faster exit because of unequal effects on applicants, even as they acknowledged the predictive value.

In May 2020 the Regents unanimously went another way: test optional admissions for 2021 and 2022, test blind for 2023 and 2024, and a new UC assessment, with the SAT and ACT gone for good if no acceptable replacement emerged. None did, and current Regents policy bars SAT and ACT scores from comprehensive review. So the faculty found the test informative and wanted something better, and the Regents chose to pursue that better thing. When no replacement proved workable, UC went forward without scores.

In a September 16 New York Times guest essay, "This California Educational Juggernaut Doesn't Need the SAT," Miriam Pawel argues that UC's record since dropping the test undercuts calls to restore it. She points to rising applications at all nine undergraduate campuses, a jump in applications from high achieving Black and Latino students at UCLA, growing selectivity at the most prestigious campuses, and record retention and graduation rates. She also argues that weak mathematics preparation is a K through 12 problem that reinstating the SAT would not fix.

Her figures show that UC can admit and graduate strong classes without the SAT, not whether scores would tell admissions readers something the application misses, and neither settles whether that information would produce better decisions. If students arrive weak in algebra, diagnostic testing after admission, placement exams, summer bridge programs and better high school preparation all go after that deficit, and a student can deserve admission and still need an honest diagnostic.

So does it help?

Where applicants' grades crowd the ceiling, scores carry information that grades no longer can. That describes the Ivy Plus colleges, and Harvard, which reinstated its requirement in 2024, says it has found that scores are the best predictors of Harvard grades. By UC's own analysis it described UC as well. Berkeley and UCLA, among the most selective public universities in the country, are where that pattern should be strongest, and where it most deserves testing. Where GPA still varies widely, as in the broad access system, grades already do the work and scores add little. Yes at Harvard, then, likely at UC by its own faculty's analysis, and mostly no at a broad access public university.

The second half of the answer is size. A better prediction is not a better class, and the GRE study found that knowing the scores barely changed who got in. The 0.03 figure does not transfer to undergraduate admissions, but the mechanism may: when strong applicants far outnumber seats, a sharper prediction mostly reorders people who were already close, and the class that results looks much the same. The SAT helps selective institutions predict. The GRE study suggests that better prediction may change admissions decisions much less than the size of the predictive improvement would lead us to expect. None of that makes UC wrong to go without it, and UC's own results suggest it has not paid much of a price. UC is also unusually well placed to settle the question, because its admissions records from the years it still collected scores could help estimate whether knowing them would change the admitted class. If the answer is small, the better use of the test is the one the math problem already points to: placement after admission rather than selection before it.

So the answer to the title is probably yes, though it may matter less than you think. The SAT predicts success at Harvard, and UC's own faculty found that it predicted success across UC. What neither its defenders nor its critics have shown is that knowing a student's score would change who walks through Sather Gate.

Further Reading

From this blog

• I Weigh How Much? - on the criteria an engineer chooses before any number reaches the display. Admissions tests raise the same question at institutional scale.

Sources

• Friedman, Sacerdote, Staiger and Tine, Standardized Test Scores and Academic Performance at Ivy-Plus Colleges, NBER Working Paper 33570 (2025).
• Joyce, Afrouzi Khosroshahi, Truelsch, Gentsch and Du, Standardized Test Scores and Academic Performance at a Public University System, NBER Working Paper 34975 (2026).
• Gaebler, Isley, Avery and Goel, Reassessing the Role of Standardized Tests in University Admissions, NBER Working Paper 35407 (2026). Source of the 0.03 grade point figure.
• Harvard College, Frequently Asked Questions. Source of the statement that SAT and ACT scores are the best predictors of Harvard grades.
• UC Academic Senate, Standardized Testing Task Force reports, including the January 2020 report and the additional statement from six members.
• Miriam Pawel, This California Educational Juggernaut Doesn't Need the SAT, New York Times, September 16, 2026.

#HigherEd