Despite the US Supreme Court recently overturning affirmative action (AA), many scholars believe that AA bans in higher education hurt minorities’ opportunities because affirmative action actually delivered on its promises. Although the findings lack consistency, the impact of AA on the outcomes of under-represented minorities (URM) is generally either ambiguous or slightly negative. The bans exert a less negative effect on more competitive fields such as STEM. The picture is even worse when one considers that AA comes at the cost of lowering the chance of other, more capable minorities, such as Asians, and does not greatly impact the intended targets, i.e., the impoverished families among URMs.
This post is a quick update on my in-depth review of group differences in the SAT, occasioned by the publication of the College Board’s reports on the SAT scores of the 2023 high school graduate cohort. Using national-level test data as well as data from selected states, I will examine how the most recent results relate to the trends that I previously identified.
This article reports racial gaps in the LSAT scores for Canada (2019-2023). By using the threshold method designed by La Griffe du Lion (2007), the standardized effect sizes are computed from the proportions of members of each groups who attain specific score ranges. Results are compared with U.S. gaps in the Law School Admission Test (LSAT) and Medical College Admission Test (MCAT) scores. Details of the analysis is available here.
We recently published the IQ scores for major ethnic groups, based on the broadly representative Adolescent Brain Cognitive Development sample. These ethnic averages correlated very strongly (r = .90 to .94) with scholastic aptitude scores (SAT or ACT scores) based on nationally representative samples of American-born college students between the years 2012 and 2020. The aptitude scores came from the NPSAS surveys, which, unfortunately, have a limited number of ethnic classifications.
As Dalliard noted, understanding racial/ethnic differences in aptitude tests is important since it is a guide to the composition of the USA’s future cognitive elite. Since different ethnic groups have different political interests, which, in turn, shape policy, understanding the cognitive capital of ethnic groups is essential to predicting the trajectory of the USA in the coming century.
While no open-source national surveys provide data on SAT/ACT scores decomposed by detailed ethnic groups, Common Application provides some data for USA citizens and residents. Common App is an undergraduate college admission application service that allows one to apply to over one thousand member colleges in the USA. While the data sample is large, with over 1 million applicants each year, several issues have been reported by Freeman et al. (2021):
- The percentage of applicants reporting a test score fell from 70 and 73% in 2018-19 and 2019-20 to 40% in 2020-21. This is likely due to 89% of Common App’s members (900+ colleges) no longer requiring SAT/ACT for admission in 2020-21.
- Nearly 60% of applicants applied from ZIP Codes in the top 20% of the median household income distribution. The decline in reporting rates between 2019-20 and 2020-21 was greater in lower-income communities.
- Underrepresented minorities (not including Asians) report test scores at lower rates than non-underrepresented minorities (71% vs 77%). The drop in reporting rates between 2019-20 and 2020-21 was larger for underrepresented minorities (31% vs 47%).
- In 2019-20, 78% of non-first-generation students reported test scores versus 69% of first-generation students while in 2020-2021 the rates were 48% and 30%.
- Applications to private, more selective institutions were the most likely to include test scores (83% in 2019–20 and 44% in 2020–21), while applications to private, less selective institutions were the least likely to include test scores (67% and 28%).
The high rates of missing test scores, especially for low-achieving groups, may mean that certain group averages are biased. Moreover, many ethnic groups suffer from ethnic attrition (Emeka, 2019), in which case group identification is correlated with aptitude. To illustrate, in the case of Nigerian-Americans, Emeka (2019) noticed that Nigerians residing in poor families with parents who have not completed high school or college degrees are much more likely to drop out of the Nigerian group in favor of the African American or Black group. This is because for them, “it is not Nigerian not to go to college”.
Those caveats noted, with respect to test scores, which are reported by Kim et al. (2022), the numbers are more or less as expected. Scores for the average Asian group, average White, average Black, and Native Hawaiian or Pacific Islander are 1382, 1278, 1108, and 1181, respectively. Applicants who did not report racial/ethnic information score (1378) higher than the non-underrepresented minorities (1297). Among Asian ethnicities, Asians from India, China, Korea, Japan, Malaysia, and “None provided” scored substantially higher (around 1400 vs 1300 or less) than Asians from Cambodia, Philippines, Vietnam, Pakistan, Other Southeast, or Other South Asia (Figure 2a).
The following table reports the SAT/ACT means by race/ethnicity, including mixed races, as well as Hispanic groups by both region and race. The columns display the N (unadjusted), % of reports, average SAT/ACT, GPA, N (adjusted for % of reports), SAT/ACT in IQ metrics. The IQ metric SAT/ACT scores were computed using the NPSAS20 total SAT/ACT standard deviations.
|Other East Asia||2800||0.59||1411||95||1652||110.0|
|Other South Asia||5620||0.45||1309||92||2529||102.3|
|Other Southeast Asia||4130||0.44||1261||92||1817||98.7|
|Other (Excl. Philippines)||570||0.30||1204||86||171||94.5|
|OK Citizen Potawatomi||20||0.29||1338||90||6||104.5|
|OK Muscogee (Creek) Nation||50||0.59||1241||91||30||97.2|
|MI Sault Ste. Marie||40||0.57||1192||88||23||93.5|
|NY Saint Regis||50||0.16||1170||86||8||91.9|
|SD Oglala Sioux||20||0.18||1123||90||4||88.3|
|NC Eastern Cherokee||40||0.41||1079||87||16||85.1|
|Asian & White||25400||0.60||1354||95||15240||105.7|
|Asian & Pacific Islander||1020||0.43||1278||93||439||100.0|
|Asian & American Indian||140||0.45||1266||88||63||99.1|
|White & Pacific Islander||1010||0.48||1265||92||485||99.0|
|White & Native American||4620||0.50||1248||91||2310||97.7|
|Three or More Races||3610||0.41||1241||90||1480||97.2|
|Asian & African Am.||2680||0.43||1224||90||1152||96.0|
|White & African Am.||15680||0.40||1192||88||6272||93.5|
|African Am. & Pacific Isl.||40||0.19||1118||83||8||88.0|
|Native Am. & Pacific Isl.||1540||0.31||1095||84||477||86.2|
|African Am. & Native Am.||380||0.34||1093||84||129||86.1|
|Hispanic or Latinx Only||60870||0.27||1146||88||16435||90.1|
²Real N estimated by multiplying N column by % Reports
Despite issues with the data that were pointed out above, one clearly notices the strong similarity between these IQ-metric SAT/ACT scores by race/ethnicity, including various mixed-race categories, and the IQ-metric SAT/ACT estimates from the NPSAS reported in a previous post.
The full dataset made available by Common App can be found at the following link. See also the following report. Additional data is included in the file such as AP scores, the number of academic honors reported, and household income.
- Freeman, M., Magouirk, P., & Kajikawa, T. (2020). Applying to college in a test‐optional admissions landscape: Trends from Common App data.
- Kim, B.H., Freeman, M., Kajikawa, T., Karimi, H., & Magouirk, P. (2022). Unpacking applicant race and ethnicity, part 2: disparities in key indicators of applicant readiness and resources across detailed backgrounds. Common Application.
In this analysis of the Project Talent data, the g factor model as represented by the Spearman’s Hypothesis (SH) was confirmed for the black-white cognitive difference but not for the sex difference. Results from MGCFA and MCV corroborate each other. MGCFA detected small-modest bias with respect to race but strong bias with respect to sex cognitive difference. Full result is available at OSF.
Chuck recently published the IQ estimates for almost 30 ethnic groups/subgroups in the ABCD of the 10-year old US children. The post was an astounding hit. However, a few commenters complained that the sample sizes of some subgroups were small. I responded that if one could replicate the values and the rank order, one would have more confidence in these estimates. And this is exactly what we did here (full result available).
The SAT is the most popular standardized test used for college admissions in the United States. In principle, SAT scores offer a good way to gauge racial and ethnic differences in cognitive ability. This is because, psychometrically, the SAT is just another IQ test–that is, it is a set of items the responses to which can be objectively marked as correct or incorrect.[Note 1] Unsurprisingly, SAT scores correlate strongly with scores from other IQ tests (Frey & Detterman, 2004). It is also advantageous that SAT-takers are generally motivated to get good scores, and that large numbers of young people from all backgrounds take the test each year, enabling precise estimation of population means.
However, the SAT has at least two major limitations when used for group comparisons. Firstly, it is a high-stakes college entrance test, which means that it is a target for intense test preparation activities in ways that conventional IQ tests are not, potentially jeopardizing its validity as a measure of cognitive ability. Secondly, taking the SAT is voluntary, which means that the participant sample is not representative but rather consists of people who tend to be smarter and more motivated than the average.
This post will attempt to address these shortcomings. I will investigate whether racial and ethnic gaps in the SAT are best understood as cognitive ability gaps, or if other factors make a significant contribution, too. An important method here is to compare the SAT to other tests that are not subject to extraneous influences such as test prepping. Another goal of the post is to come up with estimates of racial/ethnic gaps in the test that are minimally affected by selection bias. This can be done with data from states where entire high school graduate cohorts take the SAT. Other topics that will receive some attention in the post include ceiling effects, predictive validity, and measurement invariance.
Because racial/ethnic gaps in the SAT have changed over time, an essential part of the analysis is understanding these temporal trends. In particular, Asian-Americans have performed extraordinarily well in the test in recent years. Getting a better handle on that phenomenon was a major motivation for the post. Comparisons of trends in the national and state-level results turned out to be informative with respect to this question.
The post is multipronged and somewhat sprawling. This is because the (publicly available) SAT data do not yield straightforward answers to many important questions about racial and ethnic differences. The only way to get some clarity on these issues is to examine them from multiple angles, none of which alone supplies definitive answers, but which together, hopefully, paint a reasonably clear picture. I have relegated many technical details of the statistical methods used, as well as many ancillary analyses, to footnotes so as make the main text less heavy-going. The data and R code needed to reproduce all the calculations, tables, and graphs in this post are included at the end of each chapter, or, in the case of some ancillary analyses, in the footnotes.
Transracial adoption studies, especially ones which examine the performance of adopted blacks, are lacking since the prominent Minnesota Transracial Study of Black adoptees (Weinberg et al., 1992, Table 2). To fill this gap I analyzed the HSLS data, and found that the math abilities of transracial adoptees do not depend on the adoptive parents’ race. Continue reading
Note: We computed these results based on multiple versions of the ABCD data (v2.01 & v3.01) and with different inclusion/exclusion criteria. I originally posted a version based on the ABCD 2.01 data filtered for missing admixture, and other scores. However, after looking, I found a version that uses the maximum 3.01 sample with age-corrected NIHTBX scores (N = 11474). While the scores for the two versions correlate at r = .98, in some cases (e.g., Vietnamese), there is a notable difference. I have now replaced the original table with the one based on N = 11474 and moved the original table to the end of the post. For replicability and modifiability, I attached the latest R code which I had in my file.
In our manuscript, titled “Reply to Warne,” we present average eduPGS and NIH Toolbox composite scores from the ABCD study, categorized by ethnic and religious groups. In our analyses, we used unweighted means instead of sample weighted scores, since we were only interested in the correlation between mean eduPGS and cognitive ability. However, we also computed weighted NIH Toolbox scores, which may be of interest to some readers.
These scores were computed using the survey package for R as recommended by Heeringa and Berglund (2020). These weighted scores, reported below, represent the “neuropsychological performance” scores, measured between 2016 and 2018, of broadly representative samples of 10-year-old American children. (Though, children were excluded, by the ABCD consortium, if they were not fluent in English or if one of their parents were not fluent in either English or Spanish.) The first three columns, after the group labels, display the sample size, means, and standard deviations, respectively. The fourth column presents the scores normalized with the non-Hispanic White mean set to 100.00 and standard deviations set to 15.00. To norm scores, we pooled the standard deviations across all groups (pooled SD = 16.45) and transformed the values using the pooled SD. On a reader request, I added average years of parental education, which I previously outputted, in the fifth column.
The ethnic groups are mutually exclusive, and the specific variables used to code them are provided in the supplementary materials of the manuscript. Classifications are based on the race/ethnicity of the child as reported by the responding parent in conjunction with the nationality and immigrant status of the parents; see the Parent Demographics Survey for specific variables and the second table for definitions. To be clear, some of the definitions do not perfectly overlap with ones commonly used in the social sciences. For example, the classification “USA Blacks” refers to children who were identified as being Black, not being White, not being Hispanic, but also not having an immigrant parent or grandparent. This was done because, when computing the scores, we were interested in mutually exclusive ethnocultural groups.
Bear in mind that the sample sizes are often small and so the corresponding estimates are imprecise and also that the NIH Toolbox battery is fluid-intelligence loaded. For comparison, Sailer, in 2009, reported cognitive abilities of legal American immigrants based on the digit span backwards test. Additionally, in 2015, I summarized scores by immigrant generation and ethnic groups mostly based on scholastic tests.
|Ethnic/National Group||N||M||SD||IQ-Metric Score||Parental Education (Years)|
|Korean & Japanese||33||115.13||19.15||110.05||16.36|
|White & Asian Indian||44||114.66||14.19||109.62||16.75|
|White & Korean/Japanese||78||111.41||18.02||106.65||15.89|
|White & Chinese||77||109.77||18.16||105.16||16.42|
|White & Filipino||60||109.67||18.09||105.07||16.16|
|White & Pacific Islander||25||103.74||16.9||99.66||15.47|
|N. Africa & Mid. East||47||100.33||20.01||96.56||14.92|
|White & Native American||144||99.32||15.36||95.63||14.51|
|Central & South American||352||98.36||16.98||94.76||14.15|
|NH Black & White||418||94.93||16.91||91.63||14.14|
|White Puerto Rican||133||94.22||17.23||90.98||13.74|
|Black & Other Puerto Rican||90||90.61||15.49||87.69||13.22|
When observable measures such as socio-economic and health factors are adjusted, the IQ gap is substantially reduced yet a non-trivial difference remains. And while it is known that environmental factors are influenced by genetic factors and therefore should be not treated as pure environmental effects, an outcome that is typically ignored is that the education-matched blacks fall further behind in the IQ scores when education level increases.