This post will document the research done thus far looking at population differences in the frequency of gene variants known to predict cognitive ability. I will also evaluate the strength of this research as evidence for the view that genes are significantly involved in population differences in cognitive ability.
This research began around 2013 and has consistently resulted in strong correlations between national IQ and frequency of CA-related alleles in national populations. These correlates can be seen in rows 2, 5, and 8 of table 41 from The Intelligence of Nations:
The most recent publication in this literature produced a correlation of .89 in a sample of 23 nations between national IQ and a population mean polygenic score based on 1,267 SNPs previously associated with cognitive ability. This polygenic score thus accounted for 79% of the variance in IQ between nations. (EDU is the polygenic score)
Some of the largest residuals were for American Blacks whose IQs were significantly higher than what was predicted on the basis of their polygenic score.
The opposite occurred for Hispanics, whose IQs were over-predicted by their polygenic score.
The association between national IQ and polygenic score remained largely intact after controlling for measures of economic development and malnutrition. In fact, the polygenic score was a much stronger predictor of national IQ than were these other variables.
A second regression was run predicting the IQ’s of 8 major populations which produced a correlation between polygenic score and mean IQ of .979, meaning that nearly all the variance in IQ between these populations was predictable on the basis of their polygenic score. The most notable residual was Ashekanzi Jews, whose IQs were higher than what their polygenic score would have predicted.
The national mean on this polygenic score was also shown to almost perfectly correlate with three previous national polygenic score estimates. This shows that national differences in these polygenic scores are remaining stable as more CA related alleles are discovered.
The paper also showed that the correlation between these polygenic scores and national IQ is well in excess of the correlations between randomly chosen sets of SNPs and national IQ.
Turning to with-nation analyses, Lasker et al. (2019) analyzed the relationship between a polygenic score constructed from 10,000 gene variants previously associated with intelligence, and the black-white IQ gap in a sample of 9,421 American adolescents. This polygenic score was shown to predict IQ among both African Americans and European Americans, though its predictive validity was 51% greater for Europeans.
Within the sample, the typical gap of 15 IQ points between African Americans and White Americans was found. Moreover, a Black-white gap of 1.79 SD was observed for the polygenic score, and bi-racial Americans were found to posses intermediate polygenic scores, exhibiting a .59 SD gap with whites.
There was also shown to be less variance in polygenic score among African Americans, a finding mirroring the lower variance in IQ scores among African Americans that has been reported on for decades (Jensen, 1998).
The polygenic score was also shown to predict IQ after holding genetic European ancestry, self identified race, and skin color, constant.
The Black-White gap in this polygenic score predicts a phenotypic IQ gap of .23 SD, meaning it can account for roughly 25% of the total Black-white IQ gap. Thus, even if we interpret this relationship as totally causal, the majority of the black-white IQ gap must be explained either by other genetic variants, or by non-genetic factors.
Finally, the paper showed a strong tendencies such that the more a given cognitive ability was predictable on the basis on their polygenic score, the larger the black-white gap in that specific cognitive ability tended to be.
A third notable paper is Dunkel et al. (2019), who looked at the distribution of IQ and IQ-related polygenic scores among American Christians and Jews. They found that the polygenic score gap between Christians and Jews significantly mediated their IQ gap.
Aside from research comparing populations and nations at a point in time, there is also some research comparing groups over time. The main paper worth mentioning here is Woodley et al. (2017), who analyzed a sample of 503 modern European genomes, and 99 ancient Eurasian genomes. This paper found a significant correlation between the age of the genomes and their polygenic score, which implied that genetic cognitive ability has risen in Eurasia over the course of recorded history.
On Various Criticisms
On its face, this research seems to provide strong support for a genetic view of group differences. However, recently a series of papers have been published arguing against using polygenic scores in this way. Rosenberg et al.(2018) is typical of this literature.
The paper spends time talking about how we shouldn’t simply assume that group differences are due to genetics and about how this sort of analysis can’t reliably be applied to specific pairs of individuals, as if there are a significant number of people who both don’t know this and read papers in biology journals on new methods for analyzing the causes of population differences. There aren’t and so this choice strikes me as odd.
The paper also devotes time to worrying about the prevalence of GxE and GxG interactions differing between populations, but fails to mention that for most human behavioral traits the underlying genetic variation is almost entirely additive (e.g. Polderman et al., 2015). This certainly seems true of intelligence, which is probably the single most studied behavioral trait in humans.
The Rosenberg paper also states that “Although for many genetically complex phenotypes, polygenic scores currently explain too small a fraction of variation in the phenotype to be clinically meaningful”.
This statement is not substantiated or argued for despite the fact that it appears obviously false. Polygenic scores for many traits are more predictive of phenotype than are classic environmental risk factors, and they pass typical statistical guidelines for practical significance.
|Citation||# of Variants||Type||Phenotype||Variance Explained|
|Yengo et al. (2018)||3,290||SNP||Height||25%|
|Yengo et al. (2018)||941||SNP||BMI||22%|
|Lee et al. (2018)||1,271||SNP||Educational Attainment||11%|
|Sniekers et al. (2017)||336||SNP||Intelligence||5%|
|Davies et al. (2018)||148||Loci||Intelligence||4%|
|Li et al. (2017)||128||SNP||Schizophrenia||4%|
|Wray et al. (2018)||44||Loci||Depression||4%|
With respect specifically to intelligence, this paper also ignores the fact that the predictive validity of polygenic scores has increased several fold over the last few years but, and as pointed out in the Piffer paper, the polygenic differences between nations have remained the same.
The only criticism of this research that is serious is argument that the alleles which predict CA in European samples, which are the sorts of samples in which the correlations between these alleles and CA were initially established, may not predict IQ in non-European populations
The primary reason that this might occur is that the alleles which we are measuring might not themselves directly impact CA, but, rather, they might correlate with alleles that do, and these correlations between alleles may differ across populations.
Evolutionary, this is to be expected, as these correlations are produced by the coincidences of mating, and the sorts of people who happen to survive major population reductions, and these factors will differ between populations. However, we cannot know a priori whether this problem will be great enough to significantly inhibit the validity of this line of research.
Importantly, the Rosenberg paper, like most papers critiquing PGS research, doesn’t actually attempt to empirically asses the degree of this problem. They merely suggest that it might be a problem and leave it at that.
Now this issue will be most severe when comparing highly genetically distant populations. Given this, it is instructive to look at research comparing to predictive validity of polygenic scores among Blacks and whites.
Empirically, polygenic scores for cognitive ability do predict cognitive (and criminal) outcomes among African Americans.
However, the predictive validity of recent polygenic scores for CA among African Americans is only about half what it is among whites.
This decrement in predictive validity is a potential problem, but the difference is not so great that it is self evidently a fatal flaw in this research. It is plausible that the polygenic-score research design could work even with this less-than-ideal level of validity.
Typically, discussions of this problem are done as if we don’t actually have research on how polygenic scores differ between populations, as if we don’t know that national IQ correlates extremely well with national polygenic scores and that the more races differ in a given cognitive ability the more that cognitive ability is predictable on the basis of such polygenic scores, and this radically obscures the quality of this line of evidence.
The fact that these scores produce results that are so theoretically sensible suggests that this the research has significant validity. National differences in IQ, including for Black Africans, differences between historical populations, and differences between religious groups, are all predicted quite well by polygenic scores. It is logically possible that these polygenic scores are not valid across populations, and so these results were produced by random chance, but this is highly improbable.
This is a point that critics seem to often miss, the contention that these scores cannot be used to compare populations does not in any way explain why these results keep coming out exactly as the genetic model would predict. That is, their hypothesis, that PGS’s are invalid across populations, does not in anyway predict the evidence we are observing.
Assertions that these scores might not be valid across populations does not in any way substantiate the claim that they actually are not valid across populations. It is a hypothetical problem which would be a great deal more compelling if we didn’t actually have these results since these are results that a genetic view can easily explain while a non-genetic view cannot.
Because these results are significantly more likely given a genetic explanation of population differences than given a non-genetic one, even given the fact that these polygenic scores have somewhat lesser validity among blacks, these results still constitute empirical evidence that makes a genetic explanation more probable than it would be in the absence of this line of research.
Another paper worth mentioning is Birney et al (2019). This is a blog post, but worth discussing because it was written by some very high profile geneticists. It offers some of the same arguments that have already been addressed, but makes what is, so far as I know, a unique (and, perhaps, uniquely bad) argument in addition to these. Namely, these authors seem to imagine that population differences cannot be driven by alleles which are present in each race. They suggest the following facts give us reason to doubt genes-based explanations of group differences:
“The genetic variants that are most strongly associated with IQ in Europeans are no more population-specific than any other trait. To put it bluntly, the same genetic variants associated with purportedly higher IQ in Europeans are also present in Africans, and have not emerged, or been obviously selected for, in recent evolutionary history outside Africa.”
Of course, this is simply nonsense. An allele which is present in 40% of one population and 60% of another can cause just as much of a phenotypic difference as one that is present in 20% of one population and 0% of another. This is conceptually obvious, but also attested to by Lasker et al. (2019) who found that black-white differences in polygenic score were similar whether looking at recent or old gene variants, and when excluding or including population-specific variants.
Moreover, the authors factual claim, that gene variants associated with IQ lack population specificity, is something for which they provide no citation, and is something they seemingly just made up. Piffer (2019) finds that a significant minority of IQ related gene variants do appear to be present on in some populations.
The authors then make the obviously fallacious argument that because lots of genes are involved in intelligence and only a few are involved in skin color, there can be no correlation between these two sets of genes:
“the genetic variation related to IQ is broadly distributed across the genome, rather than being clustered around a few spots, as is the nature of the variation responsible for skin pigmentation. These very different patterns for these two traits mean that the genes responsible for determining skin pigmentation cannot be meaningfully associated with the genes currently known to be linked to IQ. These observations alone rule out some of the cruder racial narratives about the genetics of intelligence: it is virtually inconceivable that the primary determinant of racial categories – that is skin colour – is strongly associated with the genetic architecture that relates to intelligence. “
This is wrong for many reasons. First, skin color is not the primary determinant of race. Ancestry is. If this was not true, many Indians would be black and albino Africans would be white. Secondly, even if the genes impacting intelligence are not the same as the genes impacting skin color this does not imply that they cannot be strongly associated with one another due to random linkage or due to common selective pressures. Thirdly, the fact that there are many more genes which impact intelligence than impact skin color implies that there must be many genes that impact intelligence that don’t impact skin color, but it does not imply that there are any genes that impact skin color which don’t also impact intelligence. To be clear, I don’t actually think that variance in skin color and intelligence share common genetic causes, I’m simply pointing out that the argument given here is roughly as fallacious as it is possible for an argument to be.
The authors go on to complain about variables like SES and nutrition, seemingly unaware that the relevant research has already begone including these variables in models along side polygenic scores. The paper makes many other misinformed statements about race, genetics, and intelligence, but they are not relevant to the specific line of research that is the topic of this post and so I won’t comment on them here.
In conclusion, the various criticisms of research linking IQ related gene variants to racial differences in cognitive ability are not compelling, and so such research counts as evidence in favor of the view that genetics is involved in such group differences.