The Test You Cannot Buy

The objection assumes there’s an alternative in which nobody is measured. There’s no such alternative. A selective institution is not choosing between sorting and not sorting. It’s choosing an instrument. Every instrument measures something, and the only question worth arguing about is what.

Ask that question and the usual ranking inverts. The instruments that feel humane, that look at the whole person and the full context and the story behind the application, are the ones that most reliably measure the environment a candidate was raised in. The instrument that feels cold measures the candidate.

Children from families in the top 1 percent are more than twice as likely to attend an Ivy-Plus college as middle-class students with comparable SAT or ACT scores. Two thirds of that gap is not application behavior or self-selection. It’s the admissions decision itself, made between candidates whose measured academic credentials are the same.

Three things drive it: preferences for the children of alumni, weight placed on non-academic credentials, and athletic recruitment.

Now the part that should end the argument. Compare students who ended up at institutions of equal quality, and those same three factors are uncorrelated or negatively correlated with what happens to them afterwards. Academic credentials, SAT and ACT scores among them, are highly predictive of it. The components of the application that carry the wealth advantage are the components that carry no information about the candidate.

This is worth sitting with, because it’s not a claim about what holistic admissions might theoretically do. It’s a measurement of what it did, at the most selective institutions in the world, over decades of files. The essay, the recommendation, the portfolio of extracurriculars, the sport: these are the surfaces on which money operates. A family can buy a coach, a consultant, a summer program, a sport that requires a boat. Nobody has ever bought a higher score by buying a boat.

The mechanism is not a scandal. It’s a design feature. An instrument that rewards a rich context will reward the people with the richest context, and it will do so while feeling generous, because looking at context is what generosity looks like.

It’s also, empirically, close to a cognitive test. Meredith Frey and Douglas Detterman correlated SAT scores with a measure of general cognitive ability drawn from the Armed Services Vocational Aptitude Battery in a national sample, and found a correlation of .82. Against Raven's Advanced Progressive Matrices, a pure figural reasoning test with no words in it, the correlation was .72 once corrected for the narrow range of ability among test takers.

The SAT descends from a group intelligence test, and it still behaves like one. That is not a criticism. It’s the reason the SAT works at all, and it’s the reason the score is the most portable and least purchasable line in the application.

But the SAT is not only that, and the interesting question is what happens in the gap. The SAT is a curriculum test as well as a reasoning test. It is tied to one country's school content, charged at a registration fee, taken in a scheduled sitting, with sections built out of vocabulary and taught mathematics.

Every one of those departures from a pure reasoning test is a place where advantage gets in.

Coaching is the obvious one. Claudia Buchmann, Dennis Condron and Vincent Roscigno measured what test preparation buys: roughly 30 points from a commercial course, roughly 37 from private tutoring, after controls. More telling than the size is the distribution. Around 70 percent of the most privileged seniors use some form of test preparation. Fewer than half of low-income students use any.

The rest of the departures are quieter and worse. A test built on one country's syllabus is a test of how much of that country's schooling you have had. A transcript is worse again, because it records a particular school, a particular curriculum and a particular set of parents, and it’s close to unreadable outside the country that produced it. The seventeen-year-old with an unfamiliar national qualification is not evaluated as a weaker candidate. They are evaluated as an unreadable one, which is worse.

So the honest summary is this. The SAT's reasoning content is the part that travels, resists coaching and predicts outcomes. Its curriculum content is the part that can be bought, that requires a particular schooling, and that stops at the border. The instrument would be fairer if it were more like an intelligence test, not less.

That is the whole argument for testing cognitive ability directly. We’re not replacing a respectable instrument with something exotic. We’re keeping the part of the respectable instrument that works, and dropping the part that makes it less fair.

The first is figural. It’s built from abstract patterns, and answering it requires no reading, no vocabulary and no arithmetic. Raven's Progressive Matrices, the archetype of this format, has been in cross-cultural use since 1936 for exactly that reason.

The second is verbal reasoning, and the distinction that matters is what kind. A vocabulary test asks whether you have met a rare word before, which is a question about the school you attended and the books that were in the house. A verbal reasoning test asks whether you can hold a relation in your head, follow an inference, and find the point where an argument breaks. It uses ordinary words to do it. The first measures acquisition. The second measures operation, and only the second is worth anything to us.

Neither half tests a syllabus. There is no history on it, no literature, no taught mathematics, nothing that rewards having sat in a particular classroom in a particular country.

The claim to make about a test like this is that it’s culture-reduced, not culture-free. The literature is clear that even visuo-spatial reasoning is not perfectly invariant across contexts, and a serious review of that question exists. Anyone who tells you a test is culture-free is selling something. The verbal half is the more loaded of the two, and pretending otherwise would be the kind of overclaiming this article exists to argue against. What we can say is what it is built to exclude: rare vocabulary, literary register, and anything that rewards a particular schooling.

The right claim is comparative. Reasoning items of either kind carry far less cultural loading than a test made of vocabulary, and far less again than an essay about your summer.

One constraint belongs in the open rather than in a footnote. The verbal half runs in English, because FrontierU runs in English. The teaching is in English, the work is in English, and a member who cannot follow an argument in English cannot use the thing we are selling. That is a requirement doing real work, which is the standard this article has spent several thousand words holding everyone else to, and we do not get an exemption from it. It is still a requirement, and in much of the world English fluency tracks the kind of schooling a family pays for. We would rather say that than have it discovered. The figural half carries no such requirement, which is the half that travels furthest, and it is why the test has one.

What that buys is reach. Nothing on the test requires the schooling of any particular country, so it can be sat by someone whose national qualification no admissions officer in Austin has ever seen. It costs us almost nothing to administer, which means it can cost the candidate nothing to take, which means the filter runs before money enters the process rather than after.

That is the population the incumbents structurally cannot see. Not the underrated applicant at a good school. The one who was never on anybody's list, in a country nobody was recruiting from, holding a transcript nobody can read. For that person a two-hour test they can sit from anywhere is not a barrier. It’s the only door in the building.

Cognitive ability is among the most replicated constructs in psychology, and it’s also routinely oversold. The most honest thing we can do is cite the correction rather than the hype. In 2022 Paul Sackett and colleagues showed that decades of validity estimates had been systematically overcorrected for range restriction. The familiar figure of around .5 for cognitive ability predicting job performance came down to roughly .3.

We think that number should be quoted by us, not at us. It’s a real relationship and a modest one. It’s not destiny, it doesn’t describe a person, and anyone building an institution on the premise that one number tells you who someone is will build a bad institution.

But notice what the pseudoscience objection costs the person making it. The SAT correlates .82 with the construct in question. If general cognitive ability is astrology, then the SAT is astrology, the LSAT and the GRE and the GMAT are astrology, and every selective university on earth has spent a century sorting applicants on a horoscope. The objection doesn’t save the incumbent system. It condemns it more thoroughly than we ever would.

You cannot hold that the SAT is a legitimate academic instrument and that the thing it correlates .82 with is meaningless. Everyone is already sorting on this. We’re doing it in the open, and with the version that cannot be tutored.

MIT reinstated its testing requirement in 2022. Dartmouth followed in February 2024, and did something unusual first: it asked four of its own economists to study the question and published what they found.

Two findings matter here. Grade inflation had degraded the high school transcript to near-uselessness: students arriving with a perfect 4.0 earned college grades just 0.1 points higher than students who arrived with a 3.2. And under the optional policy, lower-income applicants were withholding scores that would have helped them, because they judged their scores against a published average rather than against the applicants they were actually competing with. Scores below Dartmouth's average would have confirmed those candidates were qualified. Withheld, they left an admissions officer with nothing but the context of the school they came from.

That is the general result, and it’s the one to remember. Removing a measurement doesn’t remove the sorting. It moves the sorting onto the instruments that remain, and those instruments are the ones wealth already controls. Take away the number and you have not made admissions kinder. You have made it illegible, and illegibility has always favored whoever can afford to be explained.

The Court struck down both, in the same sentence, for the same reason. Neither the diploma nor the test, Chief Justice Burger wrote, was shown to bear a demonstrable relationship to successful performance of the jobs. The ruling was not anti-testing. It was anti-unvalidated-screening, and it was right. The touchstone, the Court said, is business necessity. If a screen keeps people out, it has to be doing real work.

Employers heard half of the ruling.

Direct cognitive testing became legally hazardous and expensive to defend, so it receded. The credential requirement, condemned by the same Court in the same breath, did the opposite. It spread until it became the default gate on American economic life. Joseph Fuller and Manjari Raman at Harvard Business School analyzed 26 million job postings and found, for one example among many, that 67 percent of postings for production supervisors demanded a college degree while only 16 percent of the people already doing that job had one. They put more than six million middle-skill jobs at risk from the same pattern.

That is an unvalidated screen with a documented disparate impact, which is precisely the thing Griggs prohibited. It survived because nobody litigates a norm.

So the two-hour test was replaced by a four-year, six-figure credential, and this was understood as progress. Whatever else that is, it’s not inclusion. It’s the most expensive filter ever built, sitting in front of the most consequential door in the economy, and its cost falls hardest on exactly the people the original ruling was meant to protect.

We would rather run the two-hour version.

Above the floor we look for proof of agency and obsession: evidence that a candidate starts things, finishes things, and has gone unreasonably deep on something without being asked. No legacy preference, no allocation, no recruited athletes. Every seat awarded on a basis other than ability and drive is noise pushed into a signal that employers are paying to read.

Then the part that a fair reader should press us on, because a cognitive filter is only the more inclusive instrument if what sits behind it is reachable.

Membership online is priced at a small fraction of what tuition costs, and the campuses are priced as membership rather than tuition, with no four-year lock-in and no debt. We’re establishing a scholarship fund for members who clear the bar and cannot pay. We would rather be held publicly to that sentence than write a softer one.

Selection and funding are separate problems and should be solved separately. Confusing them is how institutions end up doing both badly: lowering the standard to look accessible, while the actual barrier, the money, goes untouched.

Widening access means finding potential and funding it. It doesn’t mean lowering the bar and calling the result inclusion.

If you believe FrontierU's admissions test is elitist, you must explain why it’s more elitist than an instrument that costs a registration fee, tests one country's syllabus, rewards 37 points to whoever can afford a private tutor, and sits inside a file where the essay, the sport and the surname are doing measurable work.

We did not choose a cognitive test because we wanted a harder gate. We chose it because it was the widest one we could find. It’s the only instrument in the entire apparatus of selective admissions that a seventeen-year-old cannot be outbid on.

That is not a compromise with meritocracy. It’s the only version of it anyone has ever managed to build.