Earlier this year, a study* was published in Nature on the impacts of affirmative action policies on Asian American applicants. The researchers’ analysis of admissions data for 11 elite institutions, including Ivy Leagues, found that Asian American students generally had much lower odds of attending than white students with similar academic credentials. While they found that this gap can partly be attributed to geography and legacy status, these factors did not account for all of the discrepancies.
Yale College alumnus and former AAAYA board officer Thomas Powers, who also holds a PhD in Business Economics from Harvard Business School, conducted his own review of this research, and shares his insight on this specific study with additional context on affirmative action policies as a whole—see below to read more.
Affirmative Action in College Admissions: Is there an “Asian Penalty”?
A perspective from 2024 by an Asian-American Yale alum.
Thomas Y. Powers (YC ’09)
A recent study in Nature suggested that affirmative action policies at top U.S. universities make it much harder for Asians to be admitted. The authors estimate that Asians have 28% lower odds of admission, and South Asians, in particular, 49% lower odds of admission. These are big numbers! Are they credible?
AAAYA President Austin Baik asked me to write a few comments on the Nature article based on my background as an economist. For context, I’ll also bring in some thoughts based on the 1996 ban on affirmative action in University of California schools and events around the Students for Fair Admissions (SFFA) v. Harvard case.
I’ve tried to approach this topic from the perspective of a fellow Asian-American Yale alum interested in how my future kids might do when applying to my own alma mater.
My reading of the data is that across top U.S. universities, Asian applicants have, in recent years, been admitted at somewhat lower rates than their full “resumes” (academic work, extracurriculars, and recommendation letters) would indicate. A 10-20% “penalty” (reduction in probability of admission) seems consistent with the evidence; the Nature study’s results, despite large headline numbers, end up being in the same ballpark after controlling for geographic and legacy preferences. Additionally, given the 2023 decision in SFFA v. Harvard, such a penalty may no longer be in force.
Concerns about affirmative action are salient given the re-election of President Trump. At the end of Trump’s first term, the DOJ launched an investigation of Yale’s admissions practices, but little work was completed. Trump’s latest announcements about education seem to indicate that he will crack down on what he views as improper practices in higher education. For better or worse, with Yale’s share of Asian-American admits surprisingly falling from 30% to 24% in the wake of the SFFA decision, our alma mater may well be in the crosshairs of both SFFA and new DOJ leadership.
The Nature Study
Grossman, Tomkins, Page, and Goel (2024) argue in Nature that Asian applicants face a 28% penalty to their odds of enrollment in top U.S. universities (~20% for East and Southeast Asians and ~50% for South Asians). They base this conclusion on data from over 685,000 college applications, of which around 293,000 were from Asian students, and the rest were White students. (Because of this setup, all the statistics here are relative to the probability of admission for a White student.)
First of all, while these headline numbers seem big, they are not adjusted for two important factors: legacy and geography. Universities generally want to have a class with geographic diversity, but Asians tend to live in locations that have stronger students and more competition (think: the Bay Area). Asians are also less likely to be legacies than Whites. Once you control for these factors, the “Asian penalty” drops from 28% to 15%, a still-nonzero, but a less shocking number.
Moving on to the content of the paper, I do think the research makes a few important contributions:
First, the paper actually has data on applications to top universities! These data are not generally public, so we are usually left to guess at what universities are doing.
Second, the paper successfully, in my view, documents that Asians do have a lower probability of admission controlling for certain observable factors like academics and extracurriculars.
Third, I was glad to see the paper break out difference between East Asians, Southeast Asians, and South Asians – who are often lumped together in admissions statistics.
The main drawback I see in this work is that some important factors are not recorded in the dataset. For example, the authors only observe enrollment, not admissions, so, technically, the results could be driven by student enrollment choices rather than admissions decisions (e.g., maybe Asians just don’t like going to, or can’t afford, top schools). Other important factors that are missing include things like essays, recommendation letters, alumni interviews, intended major, athletic recruitment (they can see some information on sports, but not on whether the student was recruited), or donor connections.
Whatever one thinks of such inputs into the admissions process, a failure to control for them could potentially lead to overestimating racial bias. For example, Asians might be less likely to be interested in majoring in English Literature, and thus fail to benefit from a university’s desire to have some diversity across majors. That result would appear to be racial bias in this study because it cannot control for intended major. Thus, my suspicion is that the article somewhat overestimates the impact of race on admissions, despite making important contributions.
A Case Study: Admissions at Berkeley
It’s helpful to put the Nature study in context by looking at alternative data sources that might confirm or deny their results. The banning of affirmative action at University of California schools in 1996 due to Proposition 209 has often been used as a way of measuring the impact of affirmative action policies. The chart below shows that the “Asian share” of freshmen rose from 41% in 1997 to 42% in 1998, the first year that the proposition took effect, leading some to conclude that affirmative action had had little impact on Asians after all.
However, I think this analysis is a bit misleading, and that the “true” Asian share actually rose significantly after Prop 209. In the same year that affirmative action was banned, the number of students who “decline to state” a race jumped sharply from 5% to 14%. If we instead compute the Asian enrollment as a share of students for whom the race is known, the Asian share rises from 43% to 49% from 1997 to 1998.
Alternatively, we could assume that demographics of the “decline to state” group are similar to the overall freshman class; in that case, roughly 40% of them would be Asian, and the impact of Prop 209 would have been to increase the Asian share from 43% to 48%. Either way, the increase in the Asian share was much larger than is observed in the raw data.
My main takeaway: affirmative action at Berkeley prior to 1998 probably led to at least a ~13% decline in enrollment for Asians relative to not having affirmative action; i.e., for every 8 Asians that could have enrolled without affirmative action, only 7 enrolled under the affirmative action policy. So the policies probably had some effect, in the ballpark of what was implied by the Nature study.¹
Students for Fair Admissions v. Harvard
A recent Supreme Court case provides further context. In 2023, SCOTUS held in Students for Fair Admissions (SFFA) v. Harvard that university admissions policies could not consider race except in very specific circumstances. In the lead-up to the decision, both sides produced expert briefs analyzing a Harvard admissions data for the Classes of 2014 - 2019. Peter Arcidiacono, an economist at Duke, wrote the brief for the plaintiffs, and David Card, an economist at Berkeley, wrote the brief defending Harvard.
Arcidiacono alleged that Harvard’s affirmative action policies led to a decrease in the probability of admission for Asian students of around 14% compared to Whites (consistent with the Nature study), while Card found that there was no negative impact of being Asian on admissions probabilities.
The difference in the two analyses came down to three key factors:
Arcidiacono does not allow Harvard’s “personal rating” to be an input to his model of admissions. This rating, which is determined by the Harvard admissions office, is lower for Asians than Whites. Card treats this variable as a legitimate input that measures the strength of a candidate’s application, and thus tends to find less bias against Asians after controlling for it. Arcidiacono, who focuses on “harder” (more quantifiable) metrics, considers this variable to be subjective and a source of racial bias.
Arcidiacono excludes ALDCs (athletes, legacies, development candidates, and children of faculty and staff) from his model, because he believes that the admissions process for these applicants is different. Card argues that one needs to include all applicants.
Card estimates five separate models, one for each year, because he believes that the admissions process is different every year. Arcidiacono sees this as a way of reducing the number of data points used in each model, which would tend to make the results noisier and lower the probability of finding a significant bias.
While each of these choices could be debated, I wasn’t totally sold that the “personal rating” was as bias-free as Harvard was claiming. (And I’m not just saying that because I went to Yale….) Asians seemed to do well on Harvard’s alumni interviews and recommendation letters, which seems like it should indicate that these external reviewers didn’t have any problem with the “personal” characteristics of Asian applicants.
We can also look at what happened to the Asian share at various universities in the wake of the SFFA v. Harvard decision to see whether banning affirmative action led to an increase in the Asian share. The graph below from Education Reform Now shows some numbers to this effect. If we take these numbers at face value, they imply a small change: a +3% average increase in the Asian share (as a fraction of the original share), or a 0% median increase.
However, as with the Berkeley example, methodological issues can reverse the results. In the case of Harvard, the share of students who did not report a race doubled from 4% to 8% after the SFFA decision. According to an analysis by The Crimson, the university incorrectly reported that the Asian share stayed the same at 37%, but this was due to 1) an error in reporting the correct statistic for the prior year and 2) a change in methodology, applied only to the post-SFFA Class of 2028 number, that excluded students who do not report their race. The Crimson found that the Asian share, calculated consistently as a share of all students who reported their race, rose from 31% to 37% in the wake of the SFFA decision.² This corrected statistic would imply a roughly 16% lower chance of admission for Asians prior to the decision, consistent with Professor Arcidiacono’s estimate.
Unfortunately, Yale does not appear to have released the share of students in the class of 2028 who did not report their race, so it’s hard to do a similar analysis. What we can say is that between this non-reporting bias, the potential for universities to change their methodology over time, and the fact that international students and students of multiple races may or may not be excluded from these numbers, it can be very difficult to accurately create a diagram like the one above that properly compares universities across one another and across time. A careful analysis of each data point is necessary.
Conclusions
If affirmative action policies are to have any impact, they must improve the admissions chances of some students, which consequently reduces the admissions chances of other students. A limited form of affirmative action was legal between the Supreme Court’s Bakke (1976) and SFFA v. Harvard decisions which allowed universities to include race as a factor in admissions for the promotion of diversity, although universities could not use direct quotas or points systems. The data presented above seem to suggest that under this regime, Asian students were admitted at a somewhat lower probability at top U.S. universities, although the headline 20%-50% numbers suggested in the Grossman et al. (2024) study may overestimate the “Asian penalty.” While it remains to be seen whether such penalties will continue to exist in the wake of SFFA v. Harvard and (potential) DOJ investigations under the Trump II administration, initial data suggest somewhat higher admissions rates for Asians for the class of 2028.
Footnotes
¹ This 13% is not exactly comparable to the numbers from the Nature study because the Nature study compared admissions rates for Asians against a baseline for Whites, while the Berkeley data compare admission rates for Asians under affirmative action against a baseline of Asians under no affirmative action. However, for simplicity, I’ve conflated these concepts somewhat. Additionally, the Nature study quotes its results in terms of odds ratios rather than probabilities; however, the two numbers aren’t too different in this specific case.
² The graph from Education Reform Now has different numbers for Harvard because it makes the opposite adjustment: it divides the number of Asian students by the total size of the freshman class in both 2027 and 2028. While this is at least consistent across years, it leads to the same issue identified with the Berkeley statistics above – it tends to bias downward the Asian share in years in which the number of students who decline to identify their race increases sharply.
*Grossman, J., Tomkins, S., Page, L. et al. The disparate impacts of college admissions policies on Asian American applicants. Sci Rep 14, 4449 (2024). https://doi.org/10.1038/s41598-024-55119-0
Comments