Are Racial Classifications Arbitrary?
Responds to criticism of my and Bo Winegard’s review of Angela Saini’s new book, ‘Superior: The Return of Race Science’
Bo Winegard and I wrote a review of Angela Saini’s new book, Superior: The Return of Race Science, which provoked a rather hostile reaction from some people on Twitter. Unsurprisingly, the vast majority of criticisms were ad hominem attacks against us, or against the publisher, Quillette. You know, it was the usual stuff: ‘racist’, ‘fascist’, ‘Nazi’, etc. However, there were a few criticisms that went beyond simple name-calling, and I would like to respond to those here.
1. Skull shape and phrenology
A number of rather obtuse commentators claimed that our article was promoting “phrenology”. The statement of ours that prompted this specific accusation was as follows:
In fact, researchers can classify human variation by continent quite accurately using only data from the human skull. (They are able to correctly classify human skulls into black and white Americans with about 80% accuracy, using only two variables)
The latter claim was taken from a 2009 paper by Stephen Ousley and his colleagues (which does argue against the “classical biological race concept”). Here is the exact quote from that paper:
DFA [Discriminant function analysis] using just two variables, basion-nasion length (BNL) and basion-prosthion length (BPL), separates American blacks and whites about 80% correctly
As at least two people have already pointed out on Twitter, craniometry (measuring the size and shape of the skull) is completely different from phrenology (attempting to correlate bumps in the skull with personality traits). Moreover, craniometric classification is standard practice in forensic science, and can be done to a reasonably high degree of accuracy just using two-dimensional images. For example, Rachel Murphy and her colleagues note that:
Black crania were characterized by a more prognathic maxilla, more sloping frontal bone, and a shorter cranial height relative to width […] In comparison, White crania on average had less maxilla projection (i.e., more orthognathic) and a more rounded cranial vault
Here is Figure 2 from that paper, which shows that the first principal component of variation in Murphy and colleagues’ data was able to demarcate the black and white skulls quite successfully:
2. Skull shape and intelligence
According to one commentator, our review “argues that […] skull shape can predict intelligence”. We did not actually defend that particular claim, and I’m not aware of any individual-level studies on the relationship between skull shape and intelligence. However, it should be noted that there is a well-documented relationship between overall brain size and intelligence. For example, a recent meta-analysis reported a correlation between brain volume and IQ of r = .31–.39. Another recent study, using a more robust causal design, reported a slightly lower figure of ρ = 0.19.
3. Genes and intelligence
According to one commentator who opined on our review, “There’s no link between IQ and genes. That’s the problem. End of story”. In response, I would merely point out that there are several decades worth of research to the contrary, including twin studies, adoption studies, and studies based on actual genetic data. Here is Figure 2 from the paper by Gail Davies and her colleagues, which shows all the SNPs that were significantly associated with general cognitive function in a sample of >300,000 individuals:
4. Race and intelligence
According to one commentator, we claimed that “whites have superior IQs”. (That same individual quoted a pseudonymous reader comment in attempt to argue that we and/or Quillette are engaged in some sort of “project”.) While we did note that mean IQ in the United States is higher among whites than among blacks (and provided supporting quotations from several leading textbooks), we purposefully avoided using the word ‘superior’. This is because ‘superior’ could be taken to imply some kind of metaphysical superiority, whereas we were simply concerned with the statistical difference between black and white Americans on one psychological trait.
5. Races and racial classification
Several commentators took issue with our discussion of race. One individual even went to the trouble of writing a blog post, in spite of the fact that he has “better things to do with [his] time”, such as taking care of “laundry that needs to be folded”. Since writing the blog post was evidently such a burden on this individual’s time, it only seems fair to try to engage with his arguments. My co-author Bo Winegard already wrote a lengthy reply to the blog post on Twitter, so I will limit my remarks to the first three points raised therein.
According to our critic, we “would have [our readers] believe” that different populations are “fixed” because of their divergent geographic origins. I’m not even sure what it would mean for different populations to be “fixed”. We explicitly invoked the mechanisms of natural selection and genetic drift to explain differences between human populations, which implies that we believe human populations can and do change under the influence of such mechanisms. Since changing is the opposite of staying fixed, our critic’s first charge is false.
Referring to our discussion of Rosenberg and colleagues’ 2002 paper, our critic writes the following:
Quillette argues that Blumenbach’s division was correct […] Without getting into the details of Rosenberg’s study, let us suppose, counter-factually, that Rosenberg had found a different number. Pretty much whatever number you suppose, Quillette could have declared that the study proved that race was real by choosing a different 18th or 19th century race scientist.
He then quotes Darwin, who noted that various Enlightenment philosophers had postulated different numbers of human races. Our critic goes on to conclude that “whatever racial division you come up with, there is no principled reason to choose it over any other”.
The first thing to say about disagreement over the number of human races is that it is hardly a unique situation. Zoologists often disagree about the number of ‘races’ or subspecies there are within a particular species. For example, David Mech notes that:
Scientists who classified wolves in North America were splitters. Old World scientists had pretty well recognized that there were 8 geographically distinct races, or subspecies of wolves in Europe and Asia. However, North American scientists split New World wolves into 24 subspecies
Needless to say, this does not mean that subspecies is a social construct in wolf.
The second thing to say is that, contrary to our critic’s assertion, we did not claim Blumenbach’s division was “correct”. We merely claimed that it was not “arbitrary”. To see why a division can be non-arbitrary without necessarily being “correct”, let us review the basic facts about human bio-demography over the last 250,000 years, drawing on recent research from population genetics.
It is generally understood that humans evolved in Sub-Saharan Africa around 250,000 years ago, and that the ancestors of modern non-Africans began leaving Africa around 75,000 years ago. The first major split in the human lineage was when African hunter-gatherers (the ancestors of San people) diverged from other Africans, something that may have happened more than 200,000 years ago. After leaving Africa, the ancestors of modern non-Africans split-off into a number of populations, which then migrated to different parts of the world. Some humans went to Europe; some went to South Asia; some went to East Asia; some went to Australia, and some went all the way to America.
During the course of our dispersal across the globe, human populations in particular regions interbred with other hominin species that they encountered. All modern non-Africans (but only Africans with Eurasian admixture) have some Neanderthal DNA, which implies that inter-breeding with Neanderthals took place after humans left Africa, and to a large extent before they split-off into the populations that migrated to different continents. Similarly, all modern Australasians have some Denisovan DNA, which implies that inter-breeding with Denisovans took place after humans split off into the ancestors of modern day Eurasians and modern-day Australasians. (East Asians also have a small amount of Denisovan DNA.) The preceding events are depicted in Figure 3 of the 2016 paper by David Reich and his colleagues:
Once they had settled in different continents, gene flow between the various human populations remained low for substantial periods of time. This is because natural barriers, such as oceans, deserts and mountain ranges, prevented inter-breeding between different populations. To quote from a 2009 editorial by Bruce Lahn and Lanny Ebenstein:
Anatomically modern humans first appeared in eastern Africa about 200,000 years ago. Some members migrated out of Africa by 50,000 years ago to populate Asia, Australia, Europe and eventually the Americas. During this period, geographic barriers separated humanity into several major groups, largely along continental lines, which greatly reduced gene flow among them. Geographic and cultural barriers also existed within major groups, although to lesser degrees.
Some of the most significant natural barriers were: the Sahara desert (separating Sub-Saharan Africans from Eurasians), the Himalayan mountains (separating South Asians from East Asians), the Indian Ocean (separating East Asians from Australasians), the Pacific Ocean (separating East Asians from Amerindians), and the Atlantic Ocean (separating Europeans from Amerindians). Other important natural barriers were: the Mediterranean sea (separating Europeans from North Africans), the Ural mountains (separating Europeans from Siberians), the Caucasus mountains (separating Europeans from Middle-East Eurasians), the Kalahari desert (separating San people from other Sub-Saharan Africans), and the Mozambique channel (separating Madagascans from other Sub-Saharan Africans).
The reduction in gene flow between different populations during the last 75,000 years gave rise to multiple levels of population structure. Natural barriers like the Sahara and the Himalayas gave rise to population structure at the continental level, whereas smaller-scale barriers (e.g., rivers) and diverse cultural practices (e.g., marriage rules) gave rise to population structure at sub-continental levels. In addition, gradients or ‘clines’ of genetic variation formed within each broad geographic region. As a result, global genetic variation exhibits a nested structure, whereby fine-scale clusters are nested within regional clusters, which are in turn nested within major continental clusters. Although much of the variation within major continental clusters is gradual, there are small discontinuities between them.
Of course, the natural barriers that impeded gene flow between human populations were by no means impenetrable, and studies of ancient DNA indicate that there were significant population movements in some parts of the world. For example, early European hunter-gatherers were substantially replaced by populations originating in Anatolia and the Caucasus during the Neolithic revolution. There were then further incursions into Europe by populations originating in the Caspian-Pontic Steppes, beginning around 4,800 years ago. As a result of events such as these, the nested structure of global genetic variation is partly lattice-shaped.
The preceding discussion implies that classifying genetic variation into broad geographic regions is in no way “arbitrary”, even if a five-category scheme is not necessarily “correct”. As a matter of fact, the classification that emerges from cluster analysis at different values of K (the number of pre-specified clusters) almost always accords with major geographic divisions. For example, consulting the 2016 paper by David Reich and his colleagues (Extended Data Figure 3, in particular), we see that: K = 2 separates Africans from non-Africans; K = 4 separates Africans, West Eurasians, Australasians, and other non-Africans; and that K = 6 separates Africans, Amerindians, South Asians, West Eurasians, Australasians, and East Asians.
Moving from K = 2 to K = 12, none of the classifications seem “arbitrary”. Moreover, when all the various sub-populations are plotted on a principal components graph, the major continental populations are clearly separable. (Note that their separation would be even clearer if the third principal component were plotted too.) Here is Extended Data Figure 2b from Reich and colleagues’ paper:
Regardless of whether one uses the term ‘race’ to refer to the major continental populations (K = 5–7), it is quite evident that these are not “arbitrary”. An arbitrary classification would be one that grouped San and Yoruba together with Druze and Iranian, or one that grouped together Finnish and Russian together with Australian and Papuan. This is palpably not what we see.
Of course, phylogenetic analyses (such as those described above) are based on genetic markers that are approximately neutral. Yet human populations not only differ at these neutral loci; they also differ at loci that have been under strong selection. The most obvious examples are the loci coding for skin, hair and eye colour, but there are many other interesting examples, both at the continental level and at various sub-continental levels. As we wrote in our review:
The humans that settled in different geographic regions subsequently came under different selection pressures […] When there is limited gene flow between populations that have come under different selection pressures, we would expect them to gradually diverge from one another over via the processes of genetic drift and natural selection
Our critic’s final point relates to the classification of people by their phenotypic traits. In this regard, he notes that “scientifically speaking there is no reason why any particular trait should be singled out”. He proceeds to ask, “what traits do you choose? And how do you justify those choices?” The answer to our critic is quite simple, and derives from a general principle of discriminant analysis, namely that the greater the number of traits you choose, the more accurate your classification will be. Hence if you try to classify people using only a single trait (e.g., skin colour), you are unlikely to be successful. But if you try to do so using many correlated traits (e.g., skin colour, body height, hair texture etc.), you will generally be quite successful.
