This is the first part of a three part response to a blog post entitled “Modern High-Throughput Genomics Versus Race Realism” written at the blog “Debunking Denialism” by the blogger Emil Karlsson. For those that don’t know, in Karlsson’s own words:”Debunking Denialism is a website dedicated to the refutation of pseudoscience and denialism by applying scientific skepticism and defending evidence-based science.”
Unfortunately, Karlsson thinks that race realism is a form of psuedo science and so has dedicated many blogs to arguing against it. In this most recent blog post, Karlsson responds to a series of questions about race realism sent to him by a viewer of his blog. He touches on a vareity of topics. This first response will deal with the existence of race and the connection between race and intelligence. In part two I will deal with race and crime. In the third response I will look at how race relates to national differences in success, both in the present, and over the long course of history.
The validity of race
Karlsson denies that there are multiple human races, and his first basic argument is as follows:
” Only a tiny minority of genetic variation occurs between continental groups. Thus, the available scientific evidence strongly disagree with the race realist position. Instead, human genetic diversity is better described as mostly continuous clines, with a few rare exceptions (Serre and Pääbo, 2004).”
Let’s unpack this statement one point at a time, beginning with this stuff about the amount of genetic variation within and between continental groups.
This argument, first forwarded Richard Lewontin in the early seventies, relies on measures of genetic differentiation such as the fixation index, of FST values. An FST value tells us how much the probability that two genes variants picked at random will not be the same decreases when the two variants are picked from the same race instead of from humanity at large.
Suppose there is some gene, A, which comes in two versions, A1 and A2. If you pick two people at random from the world and look at their A gene, the chances of them having different versions is 30%. However, if you pick two people at random from the same race and look at their A gene, the chances of them having different versions is only 15%. In this case, the FST value would be (.3-.15)/.3 = .5 , or 50%, because the probability of two randomly picked gene variants being different decreases by 50% if they are picked from the same race instead of from humanity at large.
Some people like to say that an FST value is equal to the proportion of variation contained “between” races, and so in this case 50% of genetic variation would be between them. I don’t think this wording is very clear, but that is what it means.
What we’ve just gone through only involves one gene locus. When you see an FST number reported, it is actually an average based on many genes, or another kind of genetic marker (micro satellites, SNPs, ect). So an FST value tells you much variation in an average genetic variant in a species is contained between its races.
In reality, the FST of humans is probably something like .12. This is not especially high or low, and is similar to the FST value of many other animals with subspecies.
Jackson et al. 2014, Lorenzen et al. 2008, Pierpaoli (2003), Lorenzen et al. (2007), Jordana 2003, Hofft et al. 2000, Schwartz et al 2002, Williams (2004), and Elhaik (2012).
Lewontin thought that this number (or, .6, which was the figure of his original study) implied that human races were of “virtually no taxonomic or practical significance”. However, he never explained his reasoning behind this, he just asserted it, even though this was obviously not a mainstream way of interpreting this number.
For much of the 20th century, races were considered subspecies and the most common criterion for subspecies was the “75% rule”, meaning that a subspecies would only be valid if a researcher could look at an organism’s traits without knowing the population it came from and predict which subspecies it belonged to with an accuracy of at least 75%.
Based on this background, some biologists thought that Lewontin may have been arguing that you could not predict a person’s race well based on their genetics.
This is true if you only consider one genetic variant at once. However, by considering many loci at the same time you can predict someone’s self identified race with an accuracy of more than 99% (Rosenberg et al. 2002; Tang et al. 2005; Rosenberg et al. 2005). Lewontin ignored this fact, and this mistake in reasoning is often called “Lewontin’s Fallacy”.
An analogy may help readers better understand Lewontin’s fallacy. Suppose you were told a person’s height, shoe size, career, hair length, and lifting capability. With any single piece of information, you would probably have a pretty mediocre chance of correctly guessing their sex. Moreover, the average predictive power of a single item would probably be pretty low. However, if you considered every piece of information at once you would have a much higher change of predicting the person’s sex.
By the same principle, considering many genes at once allows us to predict someone’s race far better than any single gene does.
Karlsson has heard race realists talk about “Lewontin’s fallacy”, but he misunderstand what said fallacy actually is. He writes “Lewontin’s argument was incomplete as his analysis was on the level of a single locus. Critics, such as A. W. F. Edwards, lamented that there could be correlations between different loci and that this could offer a justification for traditional racial categories. Modern studies, such as Li et al. (2008) and Rosenberg et al. (2002), that look at 300+ loci and 650 000 single-nucleotide polymorphisms show that the vast majority of human genetic variation (e. g. 93-95%) is to be found within human population and only a tiny fraction between them (e. g. 3-5%). So although the original argument by Lewontin had an important limitation, his conclusion is supported by modern genetic research.”
Karlsson seems to think that Lewontin’s study was flawed because he only analyzed one locus. This is wrong. Lewontin (1972) analyzed 16 loci. The problem wasn’t that Lewontin didn’t use enough loci. To return to the previous analogy, the average predictive power of sex predicting differences would not necessarily increase if we used more sex predicting differences. For instance, if we included life expectancy along with the items we already had and then found the average predictive power of the items it is not obvious that it would increase. Certainty, it would not necessarily increase just because we used more “markers” of sex.
Similarly, the problem with race is not that the average isn’t based on enough genetic markers, but rather than the averaging process fails to make use of all the data at once.
Is Human Variation Racial or Continuous?
Let’s now turn to the second part of Karlsson’s statement. He writes “Instead, human genetic diversity is better described as mostly continuous clines, with a few rare exceptions (Serre and Pääbo, 2004).” Karrlson also refers readers to another blog of his in which he expands upon this point by writing:
“It is easy to create the illusion of distinct genetic clusters by having a low sampling density: sample a few people from Sweden, some from Nigeria, some from Bolivia and some from China and plot the results and distinct genetic clusters appear. However, if you sample many populations living in between these groups, the emerging pattern does not lend itself to such a simplistic interpretation and do not support the existence of distinct genetic clusters (Serre and Pääbo, 2004).”
In a genetic cluster analysis, you give a computer program information on ton of people’s DNA and you tell it to sort the data into X number of groups, called clusters, so that the genetic differences within each cluster are minimized while the genetic differences between clusters is maximized. When you do this and tell the computer to group human genetic variation into 4 – 6 “clusters”, the clusters end up mirroring the races such that researchers can predict someone’s race based on which cluster they are assigned to with a 99%+ level of accuracy (Rosenberg et al. 2002; Tang et al. 2005; Rosenberg et al. 2005).
The point Karlsson is making is that human genetic variation changes gradually as a function of the geographic distance between populations. You can only make clusters appear by having unrepresentative samples of humanity.
To use an analogy, the color spectrum is an example of perfectly continuous variation.
However, if you only looked at samples of colors at the 400nm, 500nm, 600nm, and 700nm, mark, it would seem the color spectrum was characterized by huge sudden jumps in color and “clusters” in-between them. Karrlson thinks that this is what is going on with cluster studies of humans and in both blogs he cites Serre and Paabo (2004) to establish this.
Serre and Paabo (2004) was shown to be mistaken over a decade ago. Recall before that I said that averaging the predictive power from a large number of “sex markers” would not increase our predictive power. This is true. However, using more “sex markers” at once clearly would. For instance, if, in addition to our previous list, we also knew if the person liked sci-fi, and we considered all our information together, this would increase our predictive power.
Serre and Paabo made a possibly valid critique of the cluster studies that existed in 2004, their samples were not representative of the full range of human variation. To make up for this, Serra and Paabo sampled people from 52 populations around the world. However, while making up for this mistake they introduced one of their own by considering far fewer genetic variants than past studies.
This was all cleared up by Rosenberg et al. (2005). Rosenberg looked at a sample of individuals from the same 52 populations that Serre and Paabo looked at and analyzed nearly 400 genetic markers, compared to the 20 genetic markers analyzed by Serre and Paabo. The results? Human genetic variation forms clear clusters, and they match up with continental ancestry. Moreover, two populations of the same race are, on average, more genetically similar than two populations of different races, even when both population pairs are equally far from one another geographically.
This point about the number of genetic markers used and our ability to predict race was made especially clear by Alloco et al (2007):
Thus, human genetic variation is not like the color scheme. Not completely anyway. But let’s say that it was, would this be a problem for race realism?
Not at all: organizing continuous variation into discrete categories can help us predict and explain the world, and, so, can be a valid way of scientifically categorizing something. In fact, scientists often group continuous variation into discrete categories. Consider, for instance:
- Medical researchers break continuous blood pressure variation into discrete categories such as “high” and “low”.
- Physicists group continuous variation across the color spectrum into discrete colors such as “blue” and “green”.
- Social scientists break continuous variation in income into discrete categories such as “poor” and “rich”.
To relate this more strongly to race, consider skin color. Skin color changes pretty continuously throughout the world. So what? Does knowing whether someone is White, Black, or Asian, not help me predict and explain their skin color? Of course it does, and so this whole argument is actually irrelevant to the validity and existence of race.
A History Forgotten
It’s actually worse than that though. Like many race deniers, Karlsson commonly references “traditional racial categories” without ever citing a traditional racial thinker and explaining what they thought race was. This is important because race deniers often state obvious facts which have always been known about races and then act as if they have just shown that races do not exist. The semi-continuous nature of human variation is an example of this. Consider these three early racial thinkers:
Johann Bumlenbach was a founder of modern anthropology and, in the late 1700’s, devised a 6 race categorization of humans that still matches common sense today. He was one of the first writers to speak of a Causation race which included both Europe and north Africa and an early proponent of the idea that the Races all had a common ancestor. In 1775 he described the way that traits vary geographically from one race to another as an “imperceptible transition”.
“No variety exists, whether of color, countenance, or stature, so singular as not to be connected with others of the same kind by such an imperceptible transition, that it is very they are all related, or only differ from each other in degree.” – Blumenbach 1775
Comte de Buffon was the most important popularizer of the idea that a species should be defined as a population that can produce fertile offspring. He was also another early proponent of the idea that all the races have a common ancestor. Here, we can see Buffoon describing populations as differing by “imperceptible degrees”. This is clearly consistent with the modern terminology describing genetic variation as continuous.
“Man descends, by imperceptible degrees, from the most enlightened and polished nations, to people of less genius and industry; from the latter to others more gross, but still subject to kings and laws, and these, again, to savages” – Buffon 1753 (page 186)
Finally, we have Darwin. Darwin’s significance in general biology hardly needs to be explained. It is worth noting, though, that Darwin was probably the single most important historical advocate of the idea that human races are subspecies, rather than species. Here, we can see him using the continuous nature of human genetic variation as evidence for this claim:
“But the most weighty of all the arguments against treating the races of man as distinct species, is that they graduate into each other, independently in many cases, as far as we can judge, of their having intercrossed.” – Darwin 1871 (page 226)
To sum up, human genetic variation is not completely continuous. But it is somewhat continuous. Race realists have always known this and it has never been a problem.
Karlsson also describes his position as being the “mainstream” one:
“this is the mainstream scientific position with regards to human genetic diversity”
However this is not what surveys of the relative scientific communities actually show:
A few things to note about these charts:
- Researchers outside of Western Europe are more likely to believe in race
- Biologists are more likely than anthropologists to believe in race
- Young researchers are more likely to believe in race than middle age ones, and the use of race in textbooks is increasing, suggesting that belief in race is on the rise in academia
- The only place that has a consensus on race is China. The consensus is that race exists.
To be clear, I’m not saying that my position on race is mainstream while Karlsson’s is not. Rather, I am saying that this is actually a pretty open debate within academia.
Does Recent Selection Cast Doubt on Race Realism?
Moving forward, let’s say for the sake of argument that races exist. Do important genetic differences between them in traits like intelligence exist? One starting point to answering this question is to look at what kinds of genes show the highest levels of racial differentiation, a sign of recent selection.
To this end, Karlsson cites Barreiro et al (2008) who produce the following result:
Based on this he states “When scientists look at what genes have undergone recent positive selection in some populations, but not others, they mostly find genes related to the human immune system, superficial morphological traits, DNA repair, insulin regulation, metabolic pathways for ethanol and so on (Barreiro et al., 2008). Although some of the genes they identified are completely unknown, the available evidence is not consistent with the race realist claim about the genetic differences between populations, such as the flawed claim that whites evolved more intelligence, whereas blacks evolved more athletic ability. This was demolished in the comment section to a previous article on Debunking Denialism.”
(More on the referenced comments section in a moment)
In the first place, I don’t see why this evidence would be inconsistent with race realist views on intelligence and athletic ability. After all, metabolism and DNA repair are obviously involved in both these traits. (Not to mention the “unkown genes”).
In the second place, this paper is not about racial differentiation. To measure genetic differentiation, the paper used FST values, a common practice. Recall that an FST value tells you how much genetic variation is contained between a set of populations. These populations are usually races or subspecies, but in this paper they were as follows : Chinese, Japanese, African, and European. Of course, Chinese and Japanese people are of the same race, and so this paper is not about differentiation between races.
However, this paper, which is based on the same data-set, is:
“Other genes that showed higher levels of population differentiation include those involved in pigmentation, spermatid, nervous system and organ development, and some metabolic pathways, but few involved with the immune system.” – Wu and Zhang (2011) (emphasis added)
And so we see that gene variants involved in the brain are among the most racially differentiated. Given this, if anything, it would be surprising if the races did not differ psychologically for genetic reasons.
Race and IQ Related Genes
Karlsson doesn’t bring this up, but I think it’s silly to sit around speculating about how genes vaguely related to various physical systems may or may not impact intelligence while ignoring population differences in genes directly related to intelligence and educational attainment.
As of the latest paper on this topic, nine gene variants had been repeatedly associated with IQ or EA differences within the general population. (Since then far more have been discovered but their population differences have not been analyzed). In all 9 cases, the gene variants differed between Blacks and Whites in the way you would expect given the racial IQ differences we observe.
This same study also found that IQ related SNPs vary more across populations than random SNPs do or SNPs related to height, suggesting that the racially differentiating selection pressures for intelligence were stronger than those for height.
Race and IQ in Under 5 Minutes
Very briefly, I want to touch on some other evidence favoring the hereditarian view that racial intelligence differences are caused by a mix of genes and environment.
First, the basics. East Asians score higher than Whites on IQ tests and White score higher than Blacks. This has been documented all over the world, with millions of subjects, for over 100 years.
The gap between Blacks and Whites in the U.S., the most studied data in this area, is, in many studies, fully present at age three:
|1960s and Prior||12||85.57|
|1990s and beyond||13||86.67|
Other research indicates that it may be somewhat more variable in early ages, though a large gap is certainly there. Moreover, especially among adults, the gap has been fairly stable for the last 100 years:
This is in spite of the fact that racism has obviously declined over the last century.
And no, poverty cannot account for much of the gap either.
We also have individual level admixture studies. These studies show that mixed race individuals have IQs inbetween the means of the two races of their parents. For example:
|Race||Verbal IQ||Sample Size|
|Race||Math Score||Reading Score|
Studies also show that blacks with lighter skin have higher IQs than blacks with darker skin. This is significant because skin color is a proxy for white genetic admixture. For instance:
Similarly, if you break a nation down into regions, or compare nations, the one’s in which the mean person is whiter tend to be smarter. For instance here is an analysis of Mexican districts:
|% European||# of districts||RPM 2002||RPM 2005||Avg.|
And here is the relationship between mean whiteness and IQ across 29 American countries:
And here is the mean IQ of 55 nations grouped by skin color:
|Skin Color Rating||Average IQ||Number of Countries|
|0 to 1.99||98.4||18|
|2 to 3.99||95.4||15|
|4 to 5.99||84.9||8|
Another piece of evidence comes from subtest heritability. Basically, the more heritable an IQ subtest is the larger the racial difference in it tends to be, and this is best explained by the hereditarian view:
Next, then there’s data on brain size. As reviewed in this article, lots of research shows that smarter people tend to have larger brains. Moreover, the same genes that impact IQ also impact brain size, changes in a person’s brain size over their life predicts changes in their IQ, breeding experiments show that animals bred for more intelligence end up with larger brains, larger brained siblings are smarter than smaller brained siblings (meaning that brain size covariance with SES cannot explain the association), and over time, whether we are comparing different species of humans or generations within the 20th century, humans have gotten larger brains as we’ve gotten smarter. Given all this, brain size probably has a causal relationship with intelligence.
There are also extremely well established racial differences in brain size such that East Asians and Whites have larger brains than blacks. East Asians have larger brains than Whites if you adjust for body size. Whites have larger brains than East Asians if you don’t.
- Brain size variation within populations is almost completely due to genes
- Racial brain size differences are present at birth
- Black/White brain size differences have been stable for over 100 years
- Racial brain size differences are found all over the world
- Mulattoes have brain sizes in-between those of Whites and Blacks
- Many traits which tend to co-evolve with larger brains also differ racially in a way that mirrors the body size adjusted racial brain size pattern
This last point takes a little explaining. Rushton and Rushton (2003) looked at 37 anatomical features which 3 textbooks on human evolution identified as tending to co-evolve in the hominid line with larger brains. For instance, larger pelvic size tends to co-evolve with brain size so that mothers can give birth to larger brained infants. Rushton then utilized 5 forensic anthropology textbooks to look at racial differences in these traits. These traits followed the East Asian>White>Black pattern in 25 out of 31 cases. The probability of this happening at random is .000000001.
Similarly, Rushton (2004) showed that, across 234 mammalian species, brain size correlates with longevity, gestation time, birth weight, litter size, age of first mating, body weight, and body length. Various studies have shown that each of these variables also differ between the races in a way that, based on what we find across the animal kingdom, would predict the body size corrected brain size differences we observe (Rushton, 1995; Templer 2006; Rushton and Templer, 2009;).
In sum, the evidence suggests that the races differ in brain size for genetic reasons and this probably explains some of the racial IQ gaps we see around the world.
The final piece of evidence I will mention here is that the genetic distance between two nations not only correlates with their IQ difference, but is a better predictor of their IQ difference than their geographic distance is.
So, there’s lots of evidence to suggest that racial IQ differences are partly caused by genes. Now that we’ve finished our tour of Race and IQ, back to Debunking Denialism.
A Final Confusion About FST Values
As you may recall, Karlsson claimed that this race and IQ stuff was already debunked in the comments section of a previous post of his. So I went there and found this:
“Since the between-population variation in humans is around 11% and even if we assume a heritability of sprinting ability of 0.5, that would only translate into 5.5% of the variation in sprint talent being accounted for by genetic variation.”
This is a serious misunderstanding of the relationship between FST values and the heritability of a group difference. In reality, these two measures have little to do with each-other
There are multiple possible ways to demonstrate this, but here is the simplest: consider population differences in skin color. According to this logic, at most, 11% of population differences in skin color are caused by genetics.
Reflecting on the concepts, it’s hard to see any connection between an FST values and the heritability of a population difference. An FST value does not directly measure how strong gene variants differ in frequency between populations. And the degree of gene variant frequency difference does not measure the amount of phenotypic difference caused by genetics since a difference in alleles can produce a small or large physical difference depending on the specific gene variants.
In summary then, this argument is based on fallacious reasoning. Karlsson’s comments about Lewontin’s fallacy betray a second misunderstanding. His references about the continuous nature of human genetic variation are both outdated and irrelevant to historical conceptions of race. And, like many other sources of evidence, the kinds of genes which show the greatest levels of racial differentiation actually lend support to the hereditarian view on race and IQ.
That’s all for part 1. Be sure to come back and see parts 2 and 3!