Tuesday, April 27, 2021

Polygenic scores and Black Americans


Sunday Morning in Virginia (1877), by Winslow Homer. 

The ability to acquire language may be the mental domain where people of sub-Saharan African descent have undergone the most cognitive evolution since their separation from other humans.




If we look at SNP alleles associated with educational attainment, we see differences between Europeans and sub-Saharan Africans. In a previous post I asked whether the cause was genetic drift or natural selection (Frost 2021).


That post brought a comment on Twitter:


Or maybe the fact that educational attainment is based on whiteness and familial wealth in the USA but not in Africa? And familial wealth tends to be concentrated in specific closed groupings of people who only breed with each other?


I don’t think so. First, the alleles were identified in subjects from the Netherlands Twin Registry, the Finnish Twin Cohort, the Swedish Twin Registry, the Avon Longitudinal Study of Parents and Children, the UK Biobank and 23andme. Of those sources, only 23andme had American subjects.


Second, let's suppose that those alleles are incidentally related to educational attainment. Maybe they are just something that wealthy Europeans share with each other through inbreeding, a bit like the Habsburg jaw. Those alleles should therefore be useless for predicting educational attainment in other populations. Are they?


Let me answer that question by discussing two recent studies:



The Guo et al. study


Guo et al. (2019) used the same alleles to predict success on a cognitive test (verbal ability) by 8,078 Americans of different ethnic backgrounds. Two polygenic scores were calculated: one based on alleles associated with educational attainment (education PGS) and the other based on alleles associated with IQ (IQ PGS).


The polygenic scores significantly correlated with test results for all major ethnic backgrounds, except one:


The education PGS was significantly predictive of verbal ability in all estimated models and its coefficients were similar in size except for the black sample in which the coefficient was much smaller. The IQ PGS significantly predicted verbal ability in all samples except the black sample. (Guo et al. 2019)


[...] The incremental R2 s or the R2 s of "pure" PGS effects were 1.8%, 0.1%, 1%, 1.8%, 1.7%, and 1% for whites, blacks, Asians, Hispanic whites, the combined sample and the overall sample, respectively.


The literature is showing a consistent trend: polygenic scores have much less power to predict cognitive ability in people of sub-Saharan African descent than in people of European or Asian descent. In this case, the polygenic scores were ten to eighteen times worse at predicting verbal ability in Black Americans than they were at predicting verbal ability in White, Asian, and Hispanic White Americans.


Why? The reason may be that Eurasians and sub-Saharan Africans have different gene pools. Some alleles for higher cognitive ability are available in one gene pool but not in the other. There is undoubtedly overlap between the two, but not total overlap. Intelligent Nigerians, for instance, may owe their intelligence to alleles that exist only in sub-Saharan Africa.


To return to the Twitter comment, it seems clear that polygenic scores are predicting something that correlates with cognitive ability, and that "something" is not an artefact of wealthy people being related to each other and sharing the same genes. It's already a stretch to believe that close family ties are shared by high achievers throughout the United Kingdom, the Netherlands, Sweden, and Finland. Does the same family clique also include high achievers of Asian American origin? 



The Rabinowitz et al. study


Rabinowitz et al. (2019) used an education PGS to predict cognitive ability in Black American participants, specifically three cohorts from first grade to young adulthood (at which point their DNA was collected and analyzed).


The results? The PGS significantly correlated with pursuit of postsecondary education. The correlation was weak or insignificant, however, for performance on school tests. The PGS did not predict performance on a standardized reading test for any of the three cohorts, and it predicted performance on a standardized math test for only one of them. In addition, the PGS negatively correlated with having a criminal record (but only in the male subjects).


A problem here may be the young age of the participants. Cognitive ability seems to become less malleable and more hardwired with age. We can help children do better on IQ tests, but the improvement tends to disappear by adulthood (Frost 2008). Consequently, academic success in childhood may be too clouded by environmental factors to show a significant correlation with genetic factors.


On the other hand, the PGS did predict some things better than others. It predicted general academic success (pursuit of postsecondary education) and compliance with rules (absence of a criminal record). For actual school tests, it had some power to predict success on the reading test but none at all on the math test. The ability to acquire language may be the mental domain where people of sub-Saharan African descent have undergone the most cognitive evolution since their separation from other humans. The PGS cannot predict superior reading ability among Black Americans because too many of the relevant alleles are exclusive to the sub-Saharan African gene pool and remain to be identified by scientific studies.


The take-home message? At present, we can create polygenic scores that provide a rough idea of cognitive ability in people of sub-Saharan African descent. To get more than a rough idea, we need to identify the relevant alleles specific to that population.





Frost, P. (2008). IQ: Interaction between race and age. The Unz Review, May 20



Frost, P. (2021). The mismeasure of genetic differentiation. Evo and Proud, April 13



Guo, G., Lin, M.J., and K.M. Harris. (2019). Socioeconomic and Genomic Roots of Verbal Ability. bioRxiv, 544411.



Rabinowitz, J.A., S.I.C. Kuo, W. Felder, R.J. Musci, A. Bettencourt, K. Benke, ... and A. Kouzis. (2019). Associations between an educational attainment polygenic score with educational attainment in an African American sample. Genes, Brain and Behavior, e12558.



Interested said...

Maybe it's less predictive in blacks because of range restriction. Only the educated tend to volunteer their DNA for analysis and for blacks, educated people might be an elite demographic.

Rob said...

I have heard on the internet, but never seen in a respectable source, even what passes for respectability with hbd folk, nor an academic source, that IQ and income were not (as?) correlated for blacks as they are for whites in the US. So maybe educational attainment for black men correlates well with athletes ability (to get on a sports team) and rule-following (to stay on the team) than for cognitive ability. I don’t do numbers, at least not for blog comments, but a big chunk of the black men in college are there for sports.

Also, perhaps the floors on the achievement tests were too high?

Peter Frost said...


That criticism doesn't apply to either of the two studies I cited:

"We used data from the National Longitudinal Study of Adolescent to Adult Health (Add
Health) (http://www.cpc.unc.edu/projects/addhealth/), which is an ongoing longitudinal study of a nationally representative sample of more than 20,000 adolescents in grades 7-12 in 1994-95 in the United States who have been followed for more than 20 years." (Guo et al. 2019)

"The study’s analytic sample was drawn from three cohorts of participants in a series of randomized controlled trials of elementary school-based universal preventive interventions targeting early aggression and academic achievement. The trials were carried out within a single urban school district in a Mid-Atlantic region of the United States when children were in first grade. In terms of exclusion criteria, children had to attend one of the participating schools, be in first grade, and be in a mainstream as opposed to a self-contained special education classroom." Rabinowitz et al. (2019)


Your point is discussed by Guo et al. (2019):

"The analysis by ethnicity showed that cognitive ability had a much smaller effect and educational attainment had a much larger effect for African Americans than for white Americans. Examining about 10,000 Wisconsin high school students followed since their graduation in 1957, Hauser (2010) concluded that the associations of cognitive ability with job performance, occupational prestige, income, and wealth largely disappeared once levels of schooling were adjusted."

It looks like success within the Black American community is due to something that is measured by schooling but not by IQ. That finding can be interpreted in various ways. It looks like sub-Saharan Africans have a different package of mental aptitudes, perhaps weighted more toward verbal ability than toward abstract thinking. Or one can say that IQ tests are "culture-biased" against Black Americans.

Rob said...

Thanks for responding. Why read the paper when I can you? Is that effect symmetrical? Like if they controlled for cognitive ability, would education not explain much? Perhaps IQ and EA are more strongly correlated with each other than for whites? I could see this. There are lots of bright white kids who fall through the cracks/do not have the personality for school. I imagine every black kid who is close to bright enough for higher education gets picked out young. Perhaps it is educational prestige? There are a lot of white kids smart enough for the Ivies/near-Ivies/hidden Ivies who go to lower-prestige state schools. Affirmative action and black ability being what they are, every black kid smart enough for then Ivies goes there. But these effects must have been smaller in the 40’s and 59’s. How much of a factor was affirmative action back during their college or careers? Did they just track high school graduates?

I wonder how the white IQ, EA job performance occupation prestige look for whites across the IQ spectrum. I have read that the g factor is largely caused by strong correlation at lower IQ, and that once you get into the 130 and above group, subtests are less correlated. So maybe the correlation here is the opposite of that, and black correlation is smaller because there are fewer blacks in the portion that drives the white correlation? What happens when the whites are sub-sampled to match black IQ? In The G Favtor, Jensen suggested doing research with pseudo races, samples of one group chosen to match another group’s SES, educational attainment, or IQ. Follow them, and see how they compare to the population with those traits. Like, say your studying iQ’s regression to the mean, do blacks regress to a lower mean. Your sample of black parents has 75% single moms, has IQ and so of 85 and sd 12. You take cut your sample of whites down to a mean of 85 and a sd of 12, with 85% single moms. Do the blacks and the white pseudn-blacks regress to different means? The three results, white, black, and pseudo-black could all be different.

Finally, perhaps the floor was not low enough, and blacks higher rate of criminal convictions also obscures the result?

People have noticed that blacks are well-spoken, fluid speakers for their IQ scores.

Rob said...

I have a general question about GWAS, especially for IQ And EA. It seems that researchers mask (or some other term I can’t remember) alleles from the first n (8, maybe?) principle components. These are genes that correlate with each other within the sample, right? How much heritability is hiding within those components? They never say in the papers. Even if they cannot determine which of those alleles is causal, they can tell how much heritability each component accounts for? My understanding is the adult, narrow sense heritability of IQ is .4 to .5, and GWAS scores predict something like 5-15% of the variability. Is the genetic component ‘hiding’ in those first however many principle components possible to find? I mean, which specific alleles of the bunch of correlated alleles are actually causal? Of the top of my head, I can think of a few ways that might be possible, but am not enough of a stats-hound to know.

For traits like IQ or EA, where there is strong assortstive mating on the trait, alleles that are causal but uncorrelated seem like they would be odd ducks.

1) Take component by component. Mask the first n, and do the regression. Then adjust individuals’ IQ by the amount predicted for each causal SNP, then mask the first n-1 components, and and do the regression, rinse and repeat. Would this in effect treat the allelic predictions as principle components? Perhaps this could be tested by comparing results from masking the first 9 components, then doing the 9-1 comparison as above, and comparing those results to a straight 8 components masked.

2) When you have a very large sample, it must be possible to find sub samples with smaller or different principle components, or at least different genes in each of the components. At the simplest, if two alleles are always found together, it would be easy to find people who are all homozygotic for both, and then that controls for those two genes. Are there just too many alleles in each PC to do that? It seems if you could just ‘pop’ one allele out of, say, the 1st component out of the suppressed ones, then you could do the GWAS for genes after the 8th component plus that allele. Do this over and over, and you have results for every gene, regardless of how correlated it is with the others?

Are these reasonable, or am I not even wrong?

I apologize if this posts multiple times.

Anonymous said...

Examining about 10,000 Wisconsin high school students followed since their graduation in 1957, Hauser (2010) concluded that the associations of cognitive ability with job performance, occupational prestige, income, and wealth largely disappeared once levels of schooling were adjusted."

Blacks who graduated high school in 1957 might be a pretty range restricted sample, kind of like black PhDs today.

Off topic; does anyone know anything about the polygenic scores of arctic people? They're an anomaly because they have huge brains yet score lower than whites on IQ tests. Could they be held back by their environment?