Which Colleges Need Standardized Tests the Most?

Concrete Evidence (Monday, July 13, 2026)

Courtesy Marlena Sloss for The Washington Post via Getty Images

Standardized tests for college admissions are enjoying a vibe shift these days.

After abandoning test requirements during the pandemic, all eight Ivy League schools have reinstated them, with Columbia’s change set to take effect next summer. Powerful voices within the University of California system, including large groups of STEM and humanities professors, have launched attacks on its “test-blind” regime. Even the New York Times editorial board recently urged the UC colleges to reinstate standardized testing.

Elite schools say they can’t identify bright students from substandard high schools without the consistent metric of a test score, and that poorer students often fail to submit scores that would help them in test-optional regimes. The UC professors report a shocking influx of kids unprepared for college-level math.

I’m a testing fan and cheer these trends. Standardized tests measure skills using the same yardstick for everyone, they do a good job of predicting college success, and they can help distinguish true geniuses from reasonably smart kids with inflated 4.0 high-school transcripts.

We don’t like to get swept up in fads here at Concrete Evidence, though. So let’s give a fair hearing to two recent working papers, both released through the National Bureau of Economic Research, that raise contrary points worth grappling with.

The first, from a group of coauthors affiliated with the City University of New York, analyzes data from an anonymous “large urban public university system” with 11 four-year colleges. Contrary to findings at elite schools, the authors conclude that high-school GPA is a much better predictor of college success than test scores are—and that models combining the two gain little accuracy over a GPA-only approach.

In technical terms, when both are included in the same model (with a few other controls), each one-standard-deviation increase in high-school GPA predicts a college GPA about 0.36 points higher, while a similar boost to test scores increases college GPA just 0.14 points. There are similar disparities in predicting graduation rates. When the model also considers which high school each GPA came from (as some grade easier than others), GPA leads by even larger margins.

The authors note that some previous studies, conducted in similar contexts, reached similar conclusions. And they highlight several reasons why the benefits of testing might differ between public universities and elite private ones. Elite schools are more concerned about differentiating among students who got similarly high (often perfect) grades in high school, for example, and less concerned about graduation as an outcome (since nearly all of their students graduate).

Meanwhile, the other new NBER study, from four authors at Harvard, relies on “more than 13,000 applications over roughly a decade to a large public policy master’s program in the United States.” In this context, they concede, “GRE [testing] scores substantially improve predictions of first-year grades relative to predictions based on [undergrad] GPA alone.” And yet, they add, a lot of that predictive power isn’t very useful. For example, if two students will both get in regardless of whether their test scores are considered, it doesn’t really help to predict which one will get higher grades—the decision at the margin of being accepted or rejected is all that truly matters in the admissions context.

“Even with improved predictions, the downstream admissions decisions are often the same,” the authors report, and “where there are differences, they often involve selecting between similarly qualified applicants.” On top of that, they point out that transcripts and other application materials can improve the power of raw GPA, further making up for the loss of tests.

In their simulations, about two-thirds to three-quarters of the admitted students are the same whether the system considers GRE test scores or not. Classes admitted using the GRE have average GPAs only 0.03 to 0.05 points higher (though the students who were actually swapped must, by definition, differ a bit more than that).

Both studies still show a benefit from standardized tests, just a modest one. They drive home that, rather than simply following the crowd, universities should seize the present moment to hard-headedly evaluate the costs and benefits of testing to their own situations. Not every single program needs to use tests the same way. More elite schools and STEM-focused programs seem to benefit the most.

At the same time, I’d warn against making too much of these mathematical exercises. A lot of colleges drop tests as a response to political pressure and concerns about racial demographics, and a lot of admissions departments hardly need convincing to pursue non-meritocratic ends. There’s no reason to assume that, with test scores out of the way, colleges will shift the weight previously given to tests into other objective measures of academic preparation with similar predictive power.

That’s one way to square the circle between the new studies’ results and the complaints of the professors actually living with a test-blind system at the University of California. As the Times put it, “California’s top public universities have essentially randomized aspects of the admissions process.”

I’d also worry that fewer kids will take tests if public university systems no longer require them, which could prevent the discovery of bright disadvantaged kids whose high scores could prompt them to aim higher.

There’s something to be said for having kids take the test, and then making admissions officers look at the numbers, even if, in theory, much of the same information could be gleaned another way.

Other Work of Note

• A new study finding overdose-death increases after Washington and Oregon decriminalized drugs. And don’t miss Charles Lehman’s take, published in this space last week.

• Big new report on fertility trends out of the Institute for Family Studies, as well as a new study from Claudia Goldin on the topic. And do fertility declines spur developments of labor-saving technology?

• Very few cash-welfare recipients work today, even among those considered work-eligible, per an issue brief out of the Department of Health and Human Services. There’s a target of 50 percent work participation, but most states manage to drop that guideline all the way to zero by reducing their caseloads instead.

• This study finds a link between Stand Your Ground laws and killings of police officers. I’m skeptical, and indeed I take this result as a reason to be skeptical of other studies of this policy as well. (I rather doubt people are killing cops because the law says they can stand their ground when attacked in a public place. So, other things must be happening in states that pass these laws that aren’t controlled for.)

• Trends in deaths by law enforcement. A big caution is that this relies on data from death certificates, which undercount such deaths and have been subject to media coverage and other pressures to improve reporting. It’s hard to say how much, if any, of the reported increase is real vs. a change in reporting practices.

• How speeding tickets affect reoffending.

• Do cops slow response times in districts led by folks who want to cut their budget?

• The Congressional Research Service rounds up the many marriage penalties and bonuses in the tax code. Shameless plugs: Check out my report from a couple years ago on the same topic, as well as this book chapter.

• Third-party reporting reduces tax evasion. This happens, for example, when you get a 1099 or W-2, copies of which are also provided to the government.

• Foreign direct investment helps rural areas.

• How working hours change as countries develop economically.

• The European Union’s quest for climate neutrality could cost the low-income the most, because the “relative energy content of one’s consumption basket falls significantly as a function of one’s relative income.”

• How the Congressional Budget Office estimates the economic effects of federally funded R&D.

• The “mis-measurement in U.S. GDP due to missing [the] value of data has been as high as 6% in 2018.”

• Interesting analysis: The impact of disability on earnings depends on whether disability is measured by a one-time self-report, or at multiple points.

• “Consistent with prior studies, we find persistent negative effects of the pandemic on test scores, as well as substantial disparities in test scores along racial and ethnic lines.”

• A law-review article on the tension between the First Amendment and efforts to regulate the addiction-enhancing design practices of social media and video games.