Posted on August 12, 2020

No Skin in the Game

James Thompson, Unz Review, August 5, 2020

Diagram on White Admixture, Skin Color, and Intelligence in African-Americans

It is perfectly reasonable for critics to ask, every so often, if there is any work showing that genes make a contribution to intellectual differences between genetic groups. I assume it can be accepted that genes make a difference within a genetic group, and the animus arises only when genetic groups are being compared.

One approach is to begin by mentioning the major findings of the last century of research, which set the context of the debate. Although reasonable, the reading involved may seem unreasonable to those who want immediate answers. A brief summary would be that, despite many interventions over seven decades, African Americans remain roughly one standard deviation behind European Americans. Whatever the reason for the ability gap, it has not proved malleable.

Another approach is to discuss a selection of recent papers. This ought to be welcome, but unfortunately for those who want to skip the reading, it is necessary to go back to some old debates.

For example, although differences in intelligence between racial groups could be caused by different genetics, they might also be caused by trivial aspects of race like skin colour, which then triggers non-trivial bad treatment by other races. This bad treatment, the argument goes, could cause intellectual under-performance, either by denying educational access and quality, occupational opportunity, sufficient encouragement and mentoring, or leads to some other broad, unfair impositions. So, if a research finding is to be believed, it must distinguish between deep intrinsic causes and superficial social ones.

Furthermore, if some ancestral backgrounds really create higher intelligence in offspring, then those who have less of that advantageous genetic material should have progressively lower intellectual ability; that is to say, there must be a linear dose-response relationship. Purely European children must be brighter than those with a little less European in their genetic mix; and they must be brighter in turn than those with even less European blood. No excuses allowed.

Global Ancestry and Cognitive Ability.
Jordan Lasker, Bryan J. Pesta, John G. R. Fuerst and Emil O. W. Kirkegaard
Psych 2019, 1(1), 431-459;
30 August 2019

This is a very detailed paper, which makes use of a natural experiment. Since Europeans, African Americans, and the children of European/African American parents have varying amount of European ancestry, it ought to be possible to check whether that genetic mix predicts intelligence, and whether is does so better than superficial characteristics like skin colour.

The paper is set out in a series of logical steps, each countering objections commonly raised against the hereditarian hypothesis.

They say:

The present work uses a population-representative Philadelphia-based sample, the Philadelphia Neurodevelopmental Cohort (PNC), otherwise known as the Trajectories of Complex Phenotypes study (TCP). Due to the location, the results are directly comparable to those of Scarr et al.[52]. Our analysis has numerous advantages compared to earlier admixture studies.

First, participants all came from the same location, so geographic confounding is not an issue.

Second, we assessed measurement invariance (MI) for the cognitive test battery using multi-group confirmatory factor analysis (MGCFA;[64]).

Third, the heritabilities of the g factor and subtest scores have already been estimated for this sample. Specifically, Mollon et al.[65] reported heritabilities for g of 0.61 (standard error (S.E.) = 0.14) and 0.72 (S.E. = 0.07) for the non-Hispanic African and European-Americans in this sample respectively.

Fourth, we included estimates of skin, hair, and eye color to evaluate phenotypic discrimination (i.e., colorism) models of the observed differences.

Fifth, we validated polygenic scores (PGS) associated with cognitive ability for both the African- and European-American samples and we examined to what extent cognitive ability- and education-related PGS (eduPGS), could account for group differences.

Sixth, we tested for Jensen effects in relation to ancestry, heritability, and eduPGS.

Seventh, we examined whether MI was tenable across the full range of European ancestry using local structural equation modeling (LSEM).

Measurement invariance means that the tests are testing the same things in all populations. This is done by carrying out confirmatory factor analyses in both populations. The study already has calculations of the heritability of intelligence in different racial groups. The fourth point is a great addition: they have predictions of what people looked like in racial terms, so one can test if people have been treated differently because of skin-colour and hair-type superficial characteristics. If intelligence is affected by racism, then these superficial appearances will be useful predictors of the size of the deleterious effect on intelligence. Fifth, on the basis of DNA taken from most of the subjects, polygenic risk scores have been calculated, which show the genetic estimates for intelligence for each person. The sixth and seventh points are further tests for whether the genetic explanation is tenable for the test scores in these different racial groups.

There is a great deal in this paper, so I will pick out the main features only, and the technical details are all there in the text, many of them dealing with possible methodological objections.

The total sample includes data from 9421 genotyped participants assessed primarily from 2010 to 2013. Demographically, the sample was 51.7% female, 55.8% European-American, 32.9% African-American, and 11.4% Other, with a mean age of 14.2 (standard deviation (SD) = 3.7) years of age. Participants were recruited from the Philadelphia area. Persons with severe cognitive or medical impairments were excluded from the final sample. The subjects were English-speaking people aged 8–21 years at the time of testing.

Participants were administered the Penn Computerized Neuro-cognitive Battery. This battery was built to be highly-reliable, psychometrically-robust, and to incorporate tasks linked to specific brain systems. The battery consists of 14 tests grouped into five broad behavioral domains: Executive Control, Episodic Memory, Complex Cognition, Social Cognition, and Sensori-motor Speed.

The tests in the battery are as follows: Penn Conditional Exclusion Test (meant to assess Mental Flexibility), Penn Continuous Performance Test (Attention), Letter N-Back Task (Working Memory), Penn Word Memory Task (Verbal Memory), Penn Face Memory Task (Face Memory), Visual Object Learning Test (Spatial Memory), Penn Verbal Reasoning Test (Language Reasoning), Penn Matrix Reasoning Test (Nonverbal Reasoning), Penn Line Orientation Test (Spatial Ability), Penn Emotion Identification Test (Emotion Identification), Penn Emotion Differentiation Test (Emotion Differentiation), Motor Praxis Test (Sensorimotor Speed), Finger Tapping (Sensori-motor Speed), and the Penn Age Differentiation Test (Age Differentiation). The sample also completed the Wide Range Achievement Test, which is a highly-reliable broad ability measure.

Of the included participants, there were 5183 European-Americans, 3155 African-Americans, and 242 biracial African-European-Americans.

Since we were only concerned with European and African-Americans, we ran ADMIXTURE with k = 2 genetic clusters. Some subjects either had no genotypes available or their data failed quality control. Thus, the final sample size was reduced to 7399. Consistent with previous research (e.g.,[59], SIRE is strongly associated with genetic ancestry. This can be seen in Figure 1 which shows the probability of identifying with a particular SIRE group as a function of European admixture.


As previously found, people in the US know which racial groups they are in, and those social-construct descriptions match with their DNA.

And now, to see if there is some skin in this particular game:

The data did not include measures of appearance, so we opted to impute these based on genotypes. We used the HIrisPlex-S web application to do this ( This application was developed by the U.S. Department of Justice for use in forensic investigations. It imputes skin, hair, and eye color probabilities with a high degree of accuracy based on 41 SNPs (with overlapping variants; 6 for eye color, 22 for hair color, and 36 for skin color). This tool has been validated on thousands of people from diverse regions of the world[79].

We focus on skin color since this trait is given primacy by colorist theorists (e.g., [80,81]) and because we were able to calculate skin color scores for a larger sub-sample than for hair or eye color.
The correlation between our imputed color score and European ancestry was −0.87 for the combined sample (N = 5585) and −0.39 for the African-Americans only sample (N = 1557).


They then calculated 4 cognitive ability polygenic risk scores, to provide DNA-based intelligence estimates. Of course, these currently capture only part of the effects caused by genes, but the interest is to see whether these partial measures are better predictors of intelligence than skin colour. If even a partial measure is as good as, or better, than skin colour, then the case for a genetic cause of racial differences in intelligence is strengthened. Those comparisons are discussed a bit later.

Here are the simple summary statistics for the sample:


American Africans are 19% European, and when an AA marries a European their children are 80% European.

Socio-economic status is highest in Europeans, then Biracial AA-Europeans, then further down, Africans.

The intelligence data, collected in 2010-2013 from these teenagers confirms the usual 1 standard deviation difference between Europeans and American Africans. Although a narrowing of the intelligence gap is often suggested, that is not the case on this very large sample, recently measured on a variety of mental tasks.

The groups differ in the colour score. In general, Biracials are more like Europeans than American Africans.

Does racial ancestry predict intelligence? The authors construct regression equations to test this hypothesis. They describe their results thus:

In models that only included monoracial African-Americans, we found that European ancestry was always strongly and significantly related to cognitive ability. Skin color (assessed genetically with the highly accurate predictor [79,93] was associated with cognitive ability (Model 1b, Table 5), but made no significant incremental contribution when ancestry was also in the model (Model 2, Table 5).

Results could still be due to phenotypic confounding from other appearance variables. To test this possibility, we fitted a number of models including skin, hair, and eye color. We found that none of these features had significant effects on their own, except for brown eye color, which was positively related to cognitive ability, but with a large standard error. These results are shown in the R notebook.

The last monoracial African-American model (Model 3, Table 5) included SES, which had a considerable effect on cognitive ability. However, the so-called sociologist’s fallacy [2,5] may be at play here. That is, controlling for parental SES also controls for genetic effects on SES which may be shared with cognitive ability.

They also look at the power of education polygenic risk scores when used across genetic groups.

4.2. eduPGS Findings and Past Research

We also evaluated the transethnic validity of eduPGS. We found that the eduPGS with the highest validity for g in both the African- and European-American samples was the MTAG_EA_10K set. Although the validity in the African-American sample was approximately half of that in the European-American sample (rAA = 0.1115; rEA = 0.2269), the relations were statistically significant in both populations (p < 0.0001). As with Piffer[111] we found large African-/European-American differences in these eduPGS (d = 1.89). Using the beta in the African-American sample and controlling for the effect of European ancestry (B = 0.124; Model 1b; Table 10), we estimate that the known eduPGS can naïvely explain as much as 20%–25% of the African-/European-American intelligence gap.

Removing variants with low Minor Allele Frequency in African 1000 Genomes lineages had little effect on the validity of MTAG_10K eduPGS among African and European-Americans. Thus, contrary to some arguments, European-specific alleles do not seem to be biasing prediction with eduPGS. Our results corroborate those of Piffer[111] who found a strong ecological correlation between MTAG-derived SNPs and population IQ (r = 0.86). We also found, using both regression and path analysis that, while the eduPGS mediates the association between European ancestry and cognitive ability, skin color scores do not.

The authors are now coming towards their conclusion, and outline further work which can be done to test their findings.

Our data are compatible with a between-group heritability (variance explained by European ancestry) of between 50% and 70% depending on the model chosen (see Scarr et al.,p. 85). This estimate of between-group heritability is consistent with Rushton and Jensen’s[9] hereditarian model, according to which 50%–80% of the African-/European-American cognitive difference is due to genetic differences.

While the statistical mediation by PGS scores suggests that genetic factors may be involved, as discussed in detail by Kirkegaard et al. we cannot rule out many types of confounding environmental variables with this research design. Global admixture analysis results are suggestive and should only be considered a first step for investigating the effects of admixture on a trait. We suggest to attempt replication of the current results using a nationally representative sample and then, if these findings are confirmed, proceed to admixture mapping (local admixture analysis). This is the standard approach taken in medical epidemiology.

We suggest two approaches to further reduce the uncertainty regarding the causes of the African-/European-American cognitive ability gap.

First, attempts should be made to replicate the current results using other samples (e.g., the Add Health study).

Second, local ancestry analysis/admixture mapping to examine the regions of the genome where the association with ancestry is most pronounced are a natural follow-on project.

The rationale of such an analysis has been explained by others (e.g., [17,24,113,114]). It would also be worthwhile to attempt to replicate these results in admixed American populations outside of the United States (e.g., Brazil, Colombia, etc.). This research project can and should be expanded to other ethnic groups both in and outside of the United States. Examples include Aborigines in Australia, Cape Coloureds in South Africa, and the Mestizo population in Mexico. For some groups, it may be of utility to examine differences in broad abilities (e.g., spatial or mathematical ability) instead of general ability as done here, as differences may not be general.

The claim that 50% to 70% of the black-white difference is due to genetics is virtually the same as that concluded by Rushton and Jensen in 2005, when they did a review of 30 years of research on the topic. Testing this more recent finding on the Add Health sample could disprove it, so this ought to be done quickly.

Also, I am strongly in favour of these findings being tested on a Brazilian sample. Brazil has always had a more relaxed attitude to race, and had far higher levels of inter-marriage than was the case in the US. I clearly remember in 1964 reading Time magazine about segregation and civil rights marches in the US while taking un-segregated buses into the centre of Sao Paulo to teach English to un-segregated classes, and later listening to Brazilian singers in un-segregated bars and night clubs. Research in Brazil should show lesser effects on intelligence if part of the intelligence gap is due to racial intolerance from Europeans.

The authors conclude:

4.4. General Conclusion

Rushton and Jensen[9] called for modern genetic studies to test the hereditarian model. They predicted that “for those Black individuals who possess more White genes, their physical, behavioral, and other characteristics will approach those of Whites” ([9], p. 262). In the present study, we confirmed that this was the case for general cognitive ability. Moreover, we showed that the association between European ancestry and g was substantially mediated by eduPGS rather than skin color PGS. These results provide support for a hereditarian model.

We conducted several analyses aimed at testing a genetic hypothesis for the African-/European-American difference in cognitive ability. We found that European ancestry was a consistent predictor of cognitive ability, even after entering various controls into our models. The large observed eduPGS differences were capable of predicting substantial proportions of the cognitive ability gap, which suggests the possibility of mediation. Future assessments with cross-racially valid PGS should attempt to assess this possibility more clearly. Our global admixture-based results suggest a contribution of admixture to the African-/European-American cognitive ability gap and should encourage future investigations at the level of local admixture.

This is a very important study. I have had to summarize, and the detail about dealing with precise methods and possible confounding is in the text of the paper. Does this paper wrap up the issue of genetic factors in racial differences in intelligence? It is hard to see what else the authors could have done to carefully test the genetic hypothesis. It appears to be a solid result. Testing it in other samples should happen quickly, so that if it does not replicate, we can discard it. Meanwhile, it stands as a clear indicator that at least half of the black-white difference is probably of genetic origin.