MIT used to admit Black students at thirteen percent of every class. Two years after the Supreme Court ended race-conscious admissions, that share has been cut nearly in half. A second thumb on the scale — one that boosts women to roughly twice the male acceptance rate — never moved at all.
For a quarter-century before June 2023, MIT did what most elite American universities did. It practiced race-conscious holistic admissions. The Institute never published racial acceptance rates — few schools do — but the shape of the entering class spoke clearly.
Across the four classes from 2024 to 2027, roughly 13 percent of MIT undergraduates were Black, 15 percent Hispanic, 41 percent Asian American, and 38 percent white. Those proportions over-represented Black and Hispanic students relative to their share of the high-achieving applicant pool, and under-represented Asian American students relative to theirs.
The numbers, on their own, leave open how race entered the decision. The next chapter is where they were quantified.
The most detailed public look inside an elite admissions office came not from MIT but from a federal courtroom in Boston. In SFFA v. Harvard, Duke economist Peter Arcidiacono ran logistic regressions on years of full Harvard reader data — race, scores, ratings, hooks, and outcomes.
His baseline applicant was a middle-class Asian American male with a roughly 25 percent chance of admission. Hold every other variable constant and change only race, and the predicted odds moved like this:
The simulation literature found something equivalent at the score level: an implicit boost worth about 250 SAT points for Black applicants. While these numbers come from Harvard, the ruling and the underlying math applied across elite institutions including MIT.
In June 2023, the Supreme Court read the same numbers and called them unconstitutional.
On June 29, 2023, the Supreme Court ruled 6-2 in Students for Fair Admissions v. Harvard that race-based affirmative action in college admissions violates the Equal Protection Clause. Race could no longer be a factor in selection.
MIT responded faster and more publicly than most peers. The Institute stopped soliciting race or ethnicity from applicants. It reinstated the SAT and ACT requirement — Dean Stu Schmill noted, pointedly, that the year MIT had restored testing already showed increased diversity. And it leaned hard on socioeconomic levers.
Free tuition was extended to families earning under $75,000, then under $200,000. The Institute’s QuestBridge match for high-achieving low-income students was quintupled.
A year later, the first post-SFFA class arrived. So did the first numbers.
The Class of 2028 walked through the dome in August 2024. The demographic snapshot, when MIT released it, was the sharpest break of any HYPSM peer.
Black enrollment fell from 13 percent to 5 percent — a 60 percent drop. Hispanic enrollment fell from 15 to 11. Asian American enrollment rose from 41 to 47. White enrollment was essentially flat.
The Class of 2029, reported a year later under the IPEDS methodology federal regulators now require, told a similar story under different accounting. Black enrollment ticked up modestly to 6 percent. It remained well below the pre-SFFA 13.
MIT’s drop was unusual. Among HYPSM peers, only one institution moved in the same direction at anywhere near the same magnitude.
The post-SFFA shock did not land evenly. Princeton and Yale reported almost no change in their Black or Hispanic shares. Harvard’s numbers shifted on paper but with a methodology change underneath. Stanford moved meaningfully. MIT moved the most.
Why MIT? Two clues sit in plain sight. First, MIT was more diverse than its peers heading into the ruling, with a 13-15 percent Black-Hispanic share that had more room to compress. Second, MIT does not give a legacy preference — an indirect lever some peers retained that correlates with race in opaque ways.
Whatever the cause, the cross-school pattern is clean. The institutions that maintained their pre-SFFA share are the ones that found compliant proxies the fastest. The institution that shed the most diversity is the one with the fewest backdoors.
Race is one thumb on the scale. The other one — gender — never came up at the Supreme Court.
MIT receives roughly twice as many male applicants as female. It enrolls a class that is almost exactly half women. The arithmetic of those two facts produces a single, durable result.
In the Class of 2027 cycle, an estimated 21,700 men applied and roughly 3 percent were admitted. About 11,600 women applied and roughly 6 percent were admitted. Holding everything else constant, a female applicant had about a 94 percent better chance of admission than a male applicant.
This is not a single-year fluke. NCES enrollment data shows the bias ratio — the implicit gender boost required to get a 65/35 male applicant pool to a 50/50 enrolled class — has stayed mathematically consistent for over twenty years.
MIT does not deny the pattern. It explains it.
MIT’s admissions office frames the gender gap not as preference, but as team construction. The Institute, in its own language, seeks “a richly varied team of capable people” rather than ranking individuals on a single scale.
The pool, MIT argues, is heavily male-skewed at 65-70 percent male, so reaching parity requires a higher female acceptance rate by simple arithmetic. The female applicant pool is also more self-selected, the office maintains — and therefore comparably or more qualified on average.
A second wrinkle pushes the gap wider: female yield — the share of admitted students who actually enroll — runs lower than male yield at MIT. To land 50/50 in the dorms, the office must admit more than 50/50 in the letters.
Whether that math constitutes a preference or simply an aggregation effect is exactly the question SFFA v. Harvard answered for race. For gender, the question has not been seriously litigated.
If race can no longer be a thumb, what replaces it?
With race off the formal table, MIT — and every elite institution — is left with proxies. The leading candidates are first-generation status, Pell eligibility, and rural geography. Each correlates with race, but none is race.
For modeling purposes the simulation framework here treats the post-SFFA admissions floor as a flat 1.0× across racial categories, with proxies layered on top. First-generation applicants get a 1.4× logit-space boost. Pell-eligible applicants get 1.3×. Rural applicants get 1.2×.
Gender, meanwhile, stays where it was. At STEM-heavy institutions like MIT and Caltech, women receive an estimated 1.8× boost. At engineering schools more broadly, the multiplier is around 1.5×. At LACs, the thumb tips back the other way: men get a slight boost at Williams and Amherst.
The two thumbs on the scale, side by side. One was just removed by the Supreme Court. The other was never on the docket.