Unbiased

Both Sides of the SAT Debate Claim the Same Disadvantaged Student

Since 2022 a wave of selective universities has reinstated the SAT on evidence that scores predict who will thrive, while equity scholars counter that the same scores mostly proxy the wealth of the family that produced them.

9 min read2026-07-15

A four-year reversal that settled nothing

In March 2022 the Massachusetts Institute of Technology became the first highly selective college to abandon the pandemic-era test-optional experiment, under which an applicant chooses whether to submit an SAT or ACT score, and to require the exam again.¹ Two years later the reversal reached the Ivy League: in February 2024 Dartmouth became the first of its peers to reinstate a requirement for the class of 2029, and days later Yale announced a "test-flexible" rule that would accept an SAT, ACT, Advanced Placement, or International Baccalaureate score.²³ Brown, Harvard, and others followed. Each school reinstated on its own internal data, not on a shared brief.

The schools bringing the test back and the scholars who study those same tests describe their quarrel using the same protagonist. Both sides say they are acting for the talented teenager from a low-income family or an under-resourced high school, the applicant most likely to be missed. The reinstating colleges argue a required score is the signal that finally reveals that student. The measurement scholars argue a required score is the filter that buries her. The disagreement is not really about who deserves protection. It is an empirical dispute that lives in two precise places: what a test score predicts about college once family income is held constant, and what four years of test-optional policy actually did to who applied and who enrolled.

Dartmouth, MIT, Yale, and the Opportunity Insights economists

A required score is a common yardstick that lets admissions identify high-achieving, less-advantaged applicants who otherwise go unseen, and dropping it withholds information that helped exactly those students most.

Dartmouth's case is built on its own applicant records. A faculty working group reported that under test-optional roughly 46 percent of applicants, and 31 percent of enrolled students, submitted no score at all, so the committee was reading many files blind.⁴ The concrete harm they found was that low-scoring silence hid genuine talent. Among less-advantaged applicants scoring between 1450 and 1490, revealing the score raised the admission probability by a factor of 3.7, from 2.0 percent to 7.4 percent, because a strong score from a modest background is precisely what flags a student to the committee. By the group's estimate, about 1,000 less-advantaged applicants in each pool held scores above 1400 that they never submitted.⁴

The predictive claim behind the policy comes from the Opportunity Insights economists, whose evidence is the study of predictive validity, the degree to which a measure forecasts later performance. Comparing students of the same race, gender, income, and high-school grades, they find test scores predict first-year college grades with a normalized slope four times that of the high-school GPA.⁵ A student scoring near 1600 earns a first-year GPA about 0.43 points above one scoring 1200 with the same background, and applicants who withheld scores under test-optional went on to perform like submitters who had scored roughly 1300, an ACT of about 28.⁶ In the full technical version, lower scorers were 42 percentage points likelier to earn at least one grade of C+ or below in the first year, and the authors report no calibration bias: at a given score, less-advantaged students do not underperform their more advantaged peers.⁶

Because the Dartmouth report and that economics paper share three of four authors, the camp's claim to independence rests on the other schools reaching the same conclusion from separate data. MIT had already done so: its dean of admissions wrote that not having scores "tends to raise socioeconomic barriers to demonstrating readiness," and that a requirement is "more equitable and transparent" than optional testing because it removes the advantage held by students with better advising about strategic score disclosure.¹ Yale's dean reported the same pattern in its files, that applicants without scores "have been less likely to be admitted" and that this was "especially true for applicants from lower-income backgrounds," because even a below-median score from a weak high school gives the committee confidence a transcript alone cannot.³

The camp's sharpest equity argument turns the meritocracy complaint around. In work on admission to selective private colleges, Raj Chetty, David Deming, and John Friedman find that the factors giving rich applicants their real edge, legacy preference, athletic recruitment, and non-academic ratings, are uncorrelated or negatively correlated with how graduates later do, whereas test scores are highly predictive of top earnings and elite graduate admission.⁷ On this reading the score is the part of the application that rewards ability rather than lineage, so removing it protects the very advantages it is blamed for.

The Opportunity Insights economists (Friedman, Sacerdote & Tine), standing here because they hold the linked microdata connecting applicants' scores to their college and career outcomes across a dozen selective colleges, the record the reinstating schools rest their predictive case on.⁶

Saul Geiser, Christopher Bennett, and FairTest

Scores track family income, parental education, and race far more tightly than grades do, so ranking applicants by score entrenches the advantage it claims to measure, while dropping the requirement widened and modestly diversified applicant pools without lowering quality.

The empirical core is Saul Geiser's analysis of University of California admissions, and it rests on omitted-variable bias, the distortion that appears when a prediction model leaves out a variable correlated with both the predictor and the outcome. Across 901,905 UC applicants from 1994 to 2011, he finds the SAT/ACT correlates 0.36 with family income, 0.45 with parental education, and negative 0.38 with underrepresented-minority status, while the high-school GPA correlates only 0.11, 0.14, and negative 0.17 with the same three.⁸ Omit those demographics and the test looks like the better predictor of college performance. Put them into the model, he reports, and the ranking reverses: high-school grades become the stronger predictor of UC outcomes.⁸

Geiser addresses the reinstatement wave directly. Responding to the announcements by Harvard, Yale, MIT, and Stanford, he argues that elite private colleges pursue academic excellence while public universities must also advance social mobility, "and for this mission the SAT is ill-suited."⁹ His answer to the reinstating schools uses their own protagonist: ranking applicants by SAT score, he writes, "systematically disadvantages top students (as measured by GPA) from less privileged backgrounds," a cost he judges large enough to swamp any diversity gained through holistic review.⁹

What test-optional did to enrollment is measured most carefully by Christopher Bennett, whose peer-reviewed study is independent of both Geiser and the advocacy groups. Across nearly 100 private institutions that went test-optional between 2005 and 2016, he finds the policy was associated with a 3 to 4 percent rise in Pell Grant recipients, a 10 to 12 percent rise in first-time students from underrepresented racial and ethnic groups, and a 6 to 8 percent rise in enrollment of women, with no clear change in application volume or yield.¹⁰ He is candid that the findings show "the potential, and the limitations," of test-optional policy as an equity tool.

The broadest outcome study on this side is William Hiss's. In "Defining Access," Steven Syverson, Valerie Franks, and Hiss examined a dataset of 955,774 applicant records across 28 test-optional institutions and found that non-submitters "ultimately graduated at rates equivalent to, or marginally higher than, Submitters, the ultimate proof of success."¹¹ In the same records, high-school GPA correlated more strongly with college success for non-submitters than the ACT or SAT did, while "the test scores continued to most strongly correlate with family income."¹¹ The policy those findings defended has since become the norm rather than the exception: FairTest, the advocacy group that has pressed against admissions testing for decades, counts more than 1,825 accredited colleges and universities, over 80 percent of the four-year total, as ACT/SAT-optional or test-free for the fall 2025 cycle, and holds that dropping the requirement broadens access without lowering academic quality.¹²

Saul Geiser (UC Berkeley CSHE), standing here because he built the omitted-variable analysis on nearly a million UC applications and wrote the direct scholarly rebuttal to the 2024 reinstatements.⁹

The fight narrows to a population and a percentage point

One camp says grades barely predict college success and scores predict it fourfold, the other says grades are the stronger predictor once income is controlled. Both numbers are honest, because they were measured on different students. The four-times finding comes from twelve Ivy-Plus colleges, where nearly every admitted applicant already carries close to a 4.0 high-school GPA. When a variable has almost no spread left, it can predict almost nothing, an artifact of the selected pool rather than a fact about grades. Geiser's reversal comes from the whole University of California applicant stream, where high-school GPA still varies widely and retains its predictive power. The first empirical question, then, is one of population: does the Ivy-Plus result travel to a public university with a social-mobility mission? Geiser argues it does not,⁹ and the reinstating schools have not claimed it does.

Calibration bias vs. omitted-variable bias Two "bias" claims sit at the center of this dispute and do not contradict each other. The economists' finding of no calibration bias means students with the same score earn similar college grades regardless of background, so the test does not underpredict for the disadvantaged. Geiser's omitted-variable bias is about model weight: because scores proxy income so heavily, a model that leaves income out over-credits the test. Both can be true at once. The live question is which one should govern an admissions decision.

The second empirical question is what test-optional actually did to the class, and here the sides are closer than their rhetoric. Dartmouth's own report found that going optional did not meaningfully shift the income distribution of its applicants or its class.⁴ Bennett found real diversity gains, but his figures are relative increases against small baselines. Read as a share of the enrolled class, independent analysis puts them at roughly one percentage point more Pell recipients and about one point more underrepresented-minority students.¹³ FairTest reads the same effect as clearly positive.¹² On the numbers, test-optional shifted class composition toward more Pell and underrepresented-minority students but only slightly, which reframes the whole argument as a trade: a diversity gain of about a percentage point set against whatever predictive signal a required score adds.

What would settle it is a study neither camp has run at scale: track the same low-income applicants to a public, mobility-minded university under both regimes, and measure both who gets in and how they do once there. The reinstate camp expects a required score to surface more high-scoring low-income admits and to predict their grades without penalty; the test-optional camp expects score-ranking to push out high-GPA low-income students while adding little that income does not already explain. The first real evidence will come from the reinstating colleges' own admitted classes, beginning with Dartmouth's class of 2029, and from whether any large public system follows them back to a requirement. Placed side by side, the two records leave the central question unsettled: each camp builds its strongest case for the same low-income student, and the evidence in hand does not yet say which measurement should govern her.

A four-year reversal that settled nothing

Dartmouth, MIT, Yale, and the Opportunity Insights economists

Saul Geiser, Christopher Bennett, and FairTest

The fight narrows to a population and a percentage point

Sources