RACE and IQ RELATED GENES

by Sean Last

Summary: Over the past 5 years, a variety of specific genes have been shown to influence IQ. Versions of these genes which lead to higher intelligence are consistently found to be more common in Whites than in Blacks. The difference between Asians and Whites is less clear. The consistency of these findings, which have been replicated across two genetic databases, makes an egalitarian view of racial intelligence differences implausible. This research also suggests that selection, instead of genetic drift, causes racial intelligence differences because IQ-related genes are more racially differentiated than most genes are. In fact, they are more racially differentiated than height related genes are, suggesting that selection for racial intelligence differences was stronger than selection for racial height differences.

When I first started looking into race realism, no one had any idea what specific genes might explain racial differences in intelligence. Happily, that situation is changing. A ground breaking 2015 paper by David Piffer of the Ulster Institute for Social Research looked at population level differences in the frequency of all 9 gene variants which, as of July 2015, had been reliably associated with intelligence. This post will describe that paper.

1. IQ-related Genes

To begin with, let’s look at how these 9 genetic variants, called SNPs, were shown to relate to IQ. These SNPs were studies across 7 studies, each of which is described below:

Reitvald et al (2013) measured two phenotypes. The first, years of education completed, had a discovery sample size of 101,069. The second, a binary variable measuring whether or not the person has completed college, had a sample size of 95,427. Both samples only included Whites. 3 SNPs were found to predict both phenotypes in the discovery and replication samples. These three SNPs explained .022% of educational variance. Each SNP was associated with an extra month of education and a 1.8% increase in the chance of completing college.

Ward (2014) had a sample of 5,979 individuals, who were genotyped and took an English SAT test, while 6,145 were genotyped and took a math SAT. The participants were between 13 and 14 years old. All participants were White and 4 principal components of SNP co-variance were extracted. The top 3 SNPs identified with Rietvald et al (2013) were found to significantly predict SAT scores. Each allele was associated with an average increase of .041 standard deviations in English scores and .028 standard deviations increase in math scores.

Reitvald et al (2014) involved three samples. One contained 107,736 individuals and measured educational attainment. Another contained 24,189 individual and measured general intelligence. A third sample contained 8,652 old individuals for whom dementia and memory was measured. It found three SNPs associated with educational attainment which were also found, in a separate sample, to be associated with intelligence. Each SNP predicted an increase of 0.3 IQ points. However, one of these SNPs was highly correlated with other SNPs in this analysis and was therefore dropped.

Davies et al (2015) utilized three samples (Ns= 53,949, 5,976, and 5,487) which were genotyped and had their general intelligence measured. All participants were White, sex was controlled for, and 4 principal components of SNP co-variance were extracted. 13 SNPs were found to predict cognitive ability at P<.0000001. These SNPs were spread across three chromosome regions. The top three SNPs, one from each region, explained between .19% and 1.27% of intelligence variance in the replication samples. These three SNPs were used in Piffer’s analysis.

Ibrahim-Verbaas (2015) utilized 40 samples. 20 of them were involved in the discovery phase and ranged in size from 5,429 to 32,070. The other 20 were used as replication samples and ranged in size from 1,311 to 21,860. The phenotypes looked at were executive functioning and processing speed. Each sample was entirely made of one race, however, some were all Black and others were all White. One SNP, rs17518584, was found to be a significant and replicated predictor.

Davies et al (2011)'s discovery sample was 3,511 individuals and its replication samples was 670 people. All participants were White. The phenotype measured was fluid intelligence. A single SNP was found to be predictive in the discovery sample but it failed to replicate.

Benjamin (2013)'s discovery sample size was 12,441 and its replication sample size was 5,548. The phenotype measured was general intelligence. All the samples were White and 4 principal components of SNP co-variance were extracted. The only significant allele, the same as in Davies et al 2011, was found to be a significant predictor of G. Thus, though the SNP failed to replicate in Davies et al (2011), it replicated here with a superior sample.

2. Racial Differences in IQ-related Genes

To see how these 9 SNPs differed between populations, Piffer utilized data on 23 populations from the public genetic database 1000 Genomes. National IQ data was mostly taken from Lynn and Vanhanen (2012).

So, what did Piffer find with these 9 SNPs? Well, for one thing, he found that they all correlated with each other to a high degree. A single factor was extracted which explained 61% of SNP frequency variance. The average factor loading strength was .76 and the strength of the factor loading ranged from .35 to .97.

A polygenic score, calculated by taking the average frequency of these SNPs in each population, correlated with national IQ at .91. It’s worth noting that these SNPs predicted national IQ better than random SNPs did and continued to predict national IQ after the general genetic distance between populations was controlled for.

The average SNP frequency by race was 36% for Blacks, 53% for Whites, and 60% for Asians. It thus mirrored the racial IQ hierarchy. On average, IQ-related SNPs were 17.4% more common among Whites than Blacks, 23.7% more common among Asians than among Blacks, and 6.2% more common among Asians than Whites.

3. What are the Chances?

Some people will respond to this evidence by pointing out, correctly, that thousands of SNPs are involved in explaining IQ variations and concluding from this that information about 9 SNPs isn’t really a big deal. After all, it could be that the next 9 SNPs researchers find favor Blacks over Whites.

But consider this: all 9 SNPs differed between Blacks and Whites in a way that would predict Whites having higher intelligence. What is the probability of that happening if, as the egalitarian would have us believe, the changes of an IQ-related SNP favoring Whites over Blacks is actually 50%. (Given that we know that the SNP frequencies differ by race, this is the only way that equal genotypic intelligence could be produced).

This is a simple question of binomial probability. The probability of finding the same outcome in 9 out of 9 traits when the probability of that outcomes occurring is actually 50% is .00195%. Put another way, the chances of the first 9 IQ-related SNPs we discover all favoring Whites if there is no bias in favor of Whites among IQ-related SNPs generally is less than 1 in 500. In fact, the probability of finding what Piffer found would not have been as high as 50% unless there was a 92.6% chance that a random IQ-related SNP will be more frequent among Whites than Blacks.

4. Signs of Selection

So IQ-related SNPs are most common among Asians, followed by Whites, followed by Blacks. But why? Well, there is some pretty good evidence that natural selection is the root cause of these differences. We can test this because, due to random drift, there will be a certain degree of genetic distance between populations just due to geographic distance and mating barriers. However, if an allele is made to differ between populations via selection then it will differ between populations more than average.

Piffer looked at this by analyzing the standard deviation in SNP frequencies across populations. He found that sets of random SNPs had an average standard deviation of .046, compared to .06 for a set of 66 SNPs related to height, and .088 for the 9 IQ-related SNPs. Thus, not only do IQ-related SNPs show signs of selection, the signs of selection for IQ-related SNPs are stronger than those for genetic variation related to height.

5. Find the genes

Thus, molecular genetic evidence on IQ-related SNPs supports the race realist position. This certainly strengthens our case, but some people will regard this as the only evidence worth considering. And this, the “find the genes” argument, is a mistake.

It should be obvious that we can infer that differences are caused by genes without knowing what “the genes” are. Prior to the 20th century, no one even knew what a gene was. Were they, therefore, unjustified in thinking that any difference, whether between pairs of people or humans and non humans, was innate?

Even today, we don’t know most of the genes that explain variation in traits like height, which are widely accepted as heritable. Nor do we know the genes which explain all the neurological differences between humans and chimps. Should we therefore withhold judgment on whether or not said differences are caused by the environment?

Recall that science is about forming a hypothesis that makes testable predictions. The fact of the matter is that hereditarian hypotheses make many predictions that have nothing directly to do with the distribution of trait related genes.

For instance, the hereditarian view on race and IQ makes predictions about how universal racial IQ gaps should be, how they should relate to sub-test and SES heritability, and whether or not they should disappear when the environment is held constant.

Clearly then, there are many valid, scientific, ways of testing hereditarian views that aren’t about “finding the genes”. Of course, the genes seem to be on our side, but our case was strong even before we had confirmed this.


The Alternative Hypothesis