DEI versus Asians in STEM

One of the main ironies of DEI in STEM is that in the name of fighting white supremacy it mostly restricts… Asians. The proximate reason is the huge contrast in math between Asian outperformance and Black and Hispanic underperformance. This is not a peculiarly American phenomenon. Some Asian groups are outperforming worldwide thanks to a math-supportive culture, talent, and hard work. Africans and Latin Americans are generally underperforming even after adjustment for poverty. Aiming for numerical parity between Asians and Blacks-plus-Hispanics in math-heavy studies or jobs serves no good purpose. It grossly discriminates against Asians, undermines STEM quality, and does nothing to raise the math standards of lagging groups or promote social harmony.

To grasp this intuitively, imagine that pro football insisted that 5% of wide receivers be Asian or that pro basketball insisted that 50% of teams be white. Would that improve performance or heighten respect for under-represented groups? Of course not. Would it be widely denounced as racism against Black basketball stars? Of course it would. Yet math stars are penalized for Asian backgrounds without generating nearly the opprobrium, despite the Supreme Court recently declaring it illegal. This is an anti-productive, immoral disgrace masquerading as social justice.

Approximately 70% of NBA players are Black. As noted in a previous esssay, the chance of this occurring through random selection from American males age 20-24 is less than 10^(-200). Clearly, some combination of genes, social milieu, coaching, motivation, and hard work helped these disproportionately Black athletes shine. At one time, shamefully, American law and custom stifled that. No Americans in their right mind stifle it now. Neither do we obsess over parsing out the causes. Rather, we applaud the stars, marvel at their achievements, and savor the competition that spurs excellence.

At the other extreme, Asian Americans (which in official US categorization excludes Pacific Islanders) are hugely under-represented in top football and basketball, far beyond what we could plausibly consider random. But again, no Americans in their right mind bemoan this as segregation or mistreatment. Asian Americans’ skills, culture, interests, and training incline them to direct their efforts elsewhere.

One area where many Asian Americans shine is math, which we might regard as an engaging sport too, just more focused on mental abstraction. For most of human history, only a small leisured elite could fully play it. New York City was one of the world’s pioneers in cultivating math expertise among poor minority groups. One proud remnant is a tier of eight specialized public high schools offering admission only to the top scorers on competitive exams. Recently Asians won 65% of the 781 spots. This is roughly as unexpected as Black representation in the NBA if all races were equally skilled. Judging from NYC public schools data, a random Asian student was 3 times as likely to gain admission as a random White student, 50 times as likely as a random Hispanic student, and 90 times as likely as a random Black student.

Instead of applauding the winners, the New York Times headlined how few Blacks (8) had gained slots and slandered the schools as “symbols of segregation”. A Google search for “spots in elite NYC schools by race” returns 15.5 million results. Critics, who include mayoral frontrunner Mamdani, want to replace entrance exams with geographic or “holistic” criteria for admission, as magnet schools have done in Fairfax County VA, Philadelphia, and Boston.

In a similar spirit, many of the top universities in the US have de facto capped Asian admissions at 20%-25%, roughly half of what their academic credentials warrant. They don’t do this to maintain white supremacy but to transfer spots from the over-represented Asians to under-represented Blacks and Hispanics. Something similar is happening to Jews too. Objectors are vilified as defenders of “white-adjacent” privilege, a term concocted to obscure what it cannot explain. From a DEI perspective, mental acumen is less important than its wrapping in visibly diverse skin.

The core problem isn’t privilege; it’s performance. Statistically, Asians significantly outperform Whites in math, who significantly outperform Hispanics, who outperform Blacks. To be clear, each of these broad groupings includes subgroups that defy these norms, and there is no evidence that performance is frozen by genes or culture. None of the differentials, whether in sports or academics, justify abridgment of individual rights or disregard for outliers in the bell curves of distributions. However, those bell curves do imply that at the top levels of math competence, Asians so outnumber Blacks and Hispanics that it is ludicrous to assign some ostensibly equitable quota. Furthermore, there are no signs that this gap will disappear this century, even if, as seems wholly feasible, average Black and Hispanic competence greatly improves.

Estimates of Math Margins

The National Assessment of Educational Progress (NAEP) has tested the math proficiency of US 4^th and 8^th graders fifteen times since 1990, with sample sizes on the order of 100,000. The scores, unlike those for IQ tests or SATs, do not explicitly constrain the ratio of standard deviation to mean, better known as the coefficient of variation or CV. Nevertheless, the aggregate CVs consistently cluster near 12%, well within the typical range for common human traits. While this essay focuses on differences in group means, let me emphasize that most seem environmental or attributable to modest pressures from natural selection, e.g., a civilization with a strong engineering emphasis shifting the mean an average ~0.01 standard deviations a generation. Our human species stands out far more for its similarities than its differences.

As discussed in previous essays, my core Margin measure refers to the mean difference from a benchmark group divided by the standard deviation of the benchmark. Here my benchmark is the math performance of US Whites. Its standard deviation typically runs 10%-15% less than the overall standard deviation.

Margins for Asian, Hispanic, and Black 4^th graders are charted below. As Asians were not distinguished from Pacific Islanders until 2011, and as Pacific Islanders have Margins similar to Hispanics, the dotted line bumps up the pre-2011 Asian values by an ad hoc 0.15. Current Margins are approximately 0.4 for Asians, -0.65 for Hispanics, and -0.85 for Blacks. From 1990 to 2015 Margins trended up by about a few tenths but apart from a Covid-related drop (Hispanics and Blacks were disproportionately hurt by school closures) have been relatively stable since.

NAEP evidence on 8^th graders is even more pessimistic about narrowing. Through 2015 the patterns roughly matched those for 4^th graders. Since 2015 the gaps have widened significantly on a combination of 0.3 higher Asian Margin and 0.2-0.3 lower Hispanic and Black Margins. This raises the mean Asian outperformance over Hispanics and Blacks to over 1.6 standard deviations.

Is this widening merely an artifact of peculiar shifts in the NAEP? To check, I calculated Margins on tests for 4^th and 8^th graders administered by Trends in International Mathematics and Science Study (TIMMS). These cover a smaller US sample but come to similar conclusions as NAEP. For 4^th graders, the average outperformance for Asians over Hispanics and Blacks shrank from 1.3 in 2015 to 1.0 in 2023. For 8^th graders, the corresponding outperformance grew from 1.2 in 2015 to 1.5 in 2023.

Since both outperformance and underperformance with respect to Whites are widening, and since US White performance relative to OECD means shows no strong trends, at least two explanations are needed. Here are the two candidates that strike me as most plausible:

• Asian families increasingly recognize math as an area of comparative advantage and are encouraging their children to cultivate it through extra attention and study.

• Math standards for most Hispanic and Black students have dropped despite more attention to basic arithmetic in elementary school.

Disparities at the Top

NYC’s elite schools admit about 1.0% of eligible students. By comparing admissions with relevant pools and assuming normal distributions, I can compare the implied tail thresholds for each racial/ethnic group. Here are the imputed Margins versus Whites inferred from admissions: 0.5 for Asians, -0.9 for Hispanics, and -1.1 for Blacks. Given the differences in coverage, these are remarkably close to the previous estimates of math Margins for 8^th graders. The main difference is the lesser outperformance for Asians, which might reflect their known tendencies to outperform less in verbal competence than in math.

Let us next compare high performances by ethnic group on math tests administered by the College Board. Shaken by criticism that “SAT math scores mirror and maintain racial inequity”, the College Board reports results less transparently than it used to. However, this report allowed me to infer fractions of percentiles for scores of 750 or higher. If math SAT takers are representative of their ethnic pools, the implied Margins are 1.0 for Asians, -0.8 for Hispanics and -1.0 for Blacks. However, these estimates are perturbed by self-selection: the least qualified students tend not to take the SAT. This shrinks both the differences in means and the standard deviations, with no guarantee that the offsetting effects cancel out.

Another problem with the math SAT is that it has been intentionally dumbed down, which helps it mask both individual and group disparities. The highest math competence tested by the College Board involves the Calculus BC exam, with maximum score of 5. The College Board stopped publishing detailed breakdowns of AP performance by race or gender and withdrew some previously published data from circulation. However, I managed to find relevant data here for 1997-2019 and supplemented it with data I had previously downloaded for 2020.

Between 1997 and 2020 the number of Calculus BC test takers soared, which signals a healthily growing interest in STEM. Despite a roughly 10% downturn in 2020 due to Covid, the multipliers were 4 for Whites, 6 for Asians, 7 for Blacks and 19 for Hispanics. Asians consistently maintained both higher rates of participation and higher rates of top scores. If we picked students randomly, Asians would be 7 times as likely as Whites, 29 times as likely as Hispanics, and 76 times as likely as Blacks to receive a 5 in Calculus BC. When the pools are estimated as 1/12 of all K-12 enrollments, the estimated Margins versus Whites in 2020 are 0.9 for Asians, -0.5 for Hispanics, and -0.8 for Blacks.

One objection is that inferior access to calculus instruction directs Blacks and Hispanics toward the less advanced Calculus AB exam. To check, I redid the calculations to count scores of 5 on either Calculus AP exam. The Margins round to the same values as for Calculus BC alone.

If Gaussian tails hold for a generation, the top 1% of math competence among US citizens is projected to be 48% White, 44% Asian, 7% Hispanic, and 1% Black. In the top 0.1%, Asians are projected to outnumber Whites nearly two to one, with Blacks and Hispanics comprising 4%. If we take into account the 10% broader diversity of Asians—most of the outperformers in math are east Asian, while southeast and central Asians tend to underperform Whites—the Asian share at the top end could be higher. The main offset will likely be Ashkenazi Jews, who also exhibit a high Margin but are not separately tabulated here.

Since math-heavy Silicon Valley talent is already heavily Asian and Jewish, all these projections do is acknowledge current trends and anticipate their continuation. I emphasize them not to propose de facto ethnic quotas but to ridicule them. Consider the roughly 1:1 ratio of Black-or-Hispanic to Asian American students that many top universities gravitated toward before the Supreme Court called their bluff. From a DEI perspective, this was a very modest aim since the corresponding K-12 population ratio exceeds 7:1. From a math perspective it was grossly anti-Asian. In 2020, Asian scores of 5 outnumbered Black-or-Hispanic scores of 5 by 4.6:1 on Calculus BC and by 3.4:1 on all Calculus APs. Furthermore, the “visual diversity” mantra left Filipinos and Hmong out in the cold, since their typically lower scores were compared to east Asian standards while Black and Hispanic scores were not.

In practice, it is hard to ignore persistent 1+ differences in Margins within ostensibly egalitarian settings. Both sides of the divide get demoralized: leaders feel held back and under-rewarded while laggards feel overburdened and scorned. One consequence is extra attrition among the laggards. Among college students who start in a STEM major, a Department of Education study found that 65% of Blacks and 50% of Hispanics left STEM within 6 years (nearly half of which dropped out of college) versus 32% of Asians. A STEM program that took heed of poorer preparation and allowed a slower pace would likely have suited them better.

Many universities and private firms try to compensate for ethnic disparities among senior researchers by hiring more Blacks and Hispanics in ancillary roles. That includes staffing a DEI department with darker-complexioned liberal arts grads who preach to the rest of the organization but hopefully don’t intervene in research proper. The problem with this tokenism is that it compounds the underlying biases and resentment it ostensibly aims to correct.

The best approach is to emulate college and professional sports, which nowadays generally ignore background and color to focus on merit and expect their fans to do the same. Which will be more effective: preaching the excellence of diversity or showcasing the diversity of excellence? Let’s be as proud of Asian and Jewish math stars, and inspired by them, as we are of Black football and basketball stars and Hispanic baseball stars.

Visceral Objections

Let me close by acknowledging and responding to some visceral objections to a meritocratic approach in STEM:

• Meritocracy is just a cover for America’s historic racism toward the darker skinned.

America was historically racist toward Asians too (and Jews). And meritocracy is more an enemy of racism than a defender. It defeated white supremacy in sports. It is defeating white supremacy in STEM. So-called “anti-racism” in STEM is just a call to discriminate against both Whites and Asians.

• While individuals can differ in inherited math acumen, ethnic groups cannot.

It is ludicrous to argue that inheritance affects every characteristic except the various forms of human intelligence. Denial reminds of the Bhagavad Gita’s comment that people observe other people dying without realizing that they will eventually die too. However, humans are such inherently social animals that much of their inheritance reflects the social web they grow up in.

On the one hand, I believe that most people can learn calculus given good training and that this ability distinguishes us from every other earthly species. On the other hand, I see no sign that every ethnic group is equally disposed to learn it well. In both the TIMMS tests mentioned earlier and the math tests for 15-year-olds administered by the OECD’s Programme for International Student Assessment (PISA), the standard deviations across the means of different countries average about two-thirds of the standard deviations within countries. And neither TIMMS nor PISA cover many poor countries, whose means are generally much lower and would greatly widen the international spread.

• Any major differences in ethnic performance reflect imperialist, white supremacist oppression.

This is patently untrue. The country with the highest mean score on PISA is Singapore, a former British colony inhabited mostly by ethnic Chinese. In the Soviet Union, central Asians significantly underperformed Russians in math by a Margin of roughly 0.5 despite half a century of Communist rule ostensibly committed to multinational equality.

Far from exaggerating disparities, and despite its numerous shortcomings, US math education has had a tempering effect. The chart below displays Margins versus US Whites from the PISA 2022 math tests. The gaps between US groups are much narrower than the gaps between east Asia and most of southeast Asia or Latin America.

• Focus on objective measures like test scores enshrines class privilege.

Math competence is indeed significantly correlated with income and other measures of socioeconomic status. This should not surprise. Individual competence can be approximated as a product of material aid, social encouragement, inherent talent, and effort. Furthermore, math smarts help families gain wealth and status. It is hard to disentangle the various contributions. Generally, the best we can do is to test competence in some imperfect way and adjust encouragement appropriately. Competitive exams helped China develop a relatively meritocratic bureaucracy that allowed some upward mobility. The Mathematical Tripos helped Cambridge develop a high-quality, competitive STEM culture. Shifting back to more subjective standards overseen by non-STEM administrators is bound to dilute merit with patronage and social standing.

The NAEP provides a “racial/ethnic score gap tool” that regresses disparities against students’ socioeconomic status (including parents’ highest level of education and number of books in home), academic behaviors (including absenteeism and taking a college-track course), and approaches to learning (including self-reported persistence and self-discipline). Half to three-quarters of mean gaps can be attributed to a linear combination of these measures. However, the measures fold in a lot of social/genetic influences that should be more respected than resented. If children of bookish, STEM-educated parents attend school regularly, do their homework, and take AP math, naturally they will tend to outperform on math tests but they should not be penalized for it.

For comparison, 5% of NBA players and 3.4% of NFL players reportedly have fathers who also played in the NBA or NFL. Statistically, the fathers’ playing history gave a thousand-fold boost to their sons’ professional chances. Should Arch Manning, currently a quarterback for the Texas Longhorns, be penalized in a future NFL draft due to his grandfather and two uncles quarterbacking in the NFL for an average 16 seasons each?

• Acknowledgment of disparities, other than to excuse them, impedes the self-esteem vital to overcoming them.

While performance and self-esteem generally move in the same direction, obsession with the latter seems peculiarly American. The US Department of Education maintains a vast database of education research called ERIC. A search for “Black differentials in math” returns 1189 results. Of the first 90 listings I checked, the vast majority emphasized Black students’ self-perceptions, teachers’ perceptions of Black students, and author’s perceptions of how to improve them. Only a handful of articles explored spurs to better achievement. This might help explain a curious phenomenon: controlling for grades and college plans, Blacks score highest in self-esteem and Asians lowest. In PISA math comparisons, South Korea and Japan rank among the highest performers yet below average in self-confidence about math. Why not focus on imparting the skills that justify self-esteem?

• Attention to group disparities is premature until school reform and DEI policies get more time to work.

Math performance differs considerably across US states even among the same ethnic group. In 2024, the mean 8^th grade NAEP math score for Whites was 1.0 standard deviation higher in New Jersey than in West Virginia. If we use mean White scores as proxies for instruction quality across states, ideally we want Blacks and Hispanics to improve even more than Whites in better-taught states so that performance gaps narrow. Instead we find only slight improvements overall. A 10 point higher White score is associated with a 2 point higher Black score and a 1 point higher Hispanic score. In contrast, Asian scores move largely in tandem with White scores.

In short, Whites and Asians currently tend to benefit more from better math instruction than Blacks and Hispanics do. One likely explanation is a combination of stabler families, more consistent homework, better attendance, and more advanced STEM courses. Another is the dumbing-down of academic standards in inner-city schools. Mean group disparities were greater in 22 large urban school districts studied by NAEP than on average for 50 states, and Chicago’s disparities exceeded the urban average.