Why do some groups of people use chopsticks and others do not? Years back, created a thought experiment. So he found a few hundred students from a university and gathered DNA samples from their cheeks which were then mapped for candidate genes associated with chopstick use. Come to find out, one of the associated genetic markers was associated with chopstick use—accounting for 50 percent of the variation in the trait (Hamer and Sirota, 2000). The effect even replicated many times and was highly significant: but it was biologically meaningless.
One may look at East Asians and say “Why do they use chopsticks” or “Why are they so good at using them while Americans aren’t?” and come to such ridiculous studies such as the one described above. They may even find an association between the trait/behavior and a genetic marker. They may even find that it replicates and is a significant hit. But, it can all be for naught, since population stratification reared its head. Population stratification “refers to differences in allele frequencies between cases and controls due to systematic differences in ancestry rather than association of genes with disease” (Freedman et al, 2004). It “is a potential cause of false associations in genetic association studies” (Oetjens et al, 2016).
Such population stratification in the chopsticks gene study described above should have been anticipated since they studied two different populations. Kaplan (2000: 67-68) described this well:
A similar argument, bu the way, holds true for molecular studies. Basically, it is easy to mistake mere statistical associations for a causal connection if one is not careful to properly partition one’s samples. Hamer and Copeland develop and amusing example of some hypothetical, badly misguided researchers searching for the “successful use of selected hand instruments” (SUSHI) gene (hypothesized to be associated with chopstick usage) between residents in Tokyo and Indianapolis. Hamer and Copeland note that while you would be almost certain to find a gene “associated with chopstick usage” if you did this, the design of such a hypothetical study would be badly flawed. What would be likely to happen here is that a genetic marker associated with the heterogeneity of the group involved (Japanese versus Caucasian) would be found, and the heterogeneity of the group involved would independently account for the differences in the trait; in this case, there is a cultural tendency for more people who grow up in Japan than people who grow up in Indianapolis to learn how to use chopsticks. That is, growing up in Japan is the causally important factor in using chopsticks; having a certain genetic marker is only associated with chopstick use in a statistical way, and only because those people who grow up in Japan are also more likely to have the marker than those who grew up in Indianapolis. The genetic marker is in no way causally related to chopstick use! That the marker ends up associated with chopstick use is therefore just an accident of design (Hamer and Copeland, 1998, 43; Bailey 1997 develops a similar example).
In this way, most—if not all—of the results of genome-wide association studies (GWASs) can be accounted for by population stratification. Hamer and Sirota (2000) is a warning to psychiatric geneticists to not be quick to ascribe function and causation to hits on certain genes from association studies (of which GWASs are).
Many studies, for example, Sniekers et al (2017), Savage et al (2018) purport to “account for” less than 10 percent of the variance in a trait, like “intelligence” (derived from non-construct valid IQ tests). Other GWA studies purport to show genes that affect testosterone production and that those who have a certain variant are more likely to have low testosterone (Ohlsson et al, 2011). Population stratification can have an effect here in these studies, too. GWASs; they give rise to spurious correlations that arise due to population structure—which is what GWASs are actually measuring, they are measuring social class, and not a “trait” (Richardson, 2017b; Richardson and Jones, 2019). Note that correcting for socioeconomic status (SES) fails, as the two are distinct (Richardson, 2002). (Note that GWASs lead to PGSs, which are, of course, flawed too.)
Such papers presume that correlations are causes and that interactions between genes and environment either don’t exist or are irrelevant (see Gottfredson, 2009 and my reply). Both of these claims are false. Correlations can, of course, lead to figuring out causes, but, like with the chopstick example above, attributing causation to things that are even “replicable” and “strongly significant” will still lead to false positives due to that same population stratification. Of course, GWAS and similar studies are attempting to account for the heriatbility estimates gleaned from twin, family, and adoption studies. Though, the assumptions used in these kinds of studies are shown to be false and, therefore, heritability estimates are highly exaggerated (and flawed) which lead to “looking for genes” that aren’t there (Charney, 2012; Joseph et al, 2016; Richardson, 2017a).
Richardson’s (2017b) argument is simple: (1) there is genetic stratification in human populations which will correlate with social class; (2) since there is genetic stratification in human populations which will correlate with social class, the genetic stratification will be associated with the “cognitive” variation; (3) if (1) and (2) then what GWA studies are finding are not “genetic differences” between groups in terms of “intelligence” (as shown by “IQ tests”), but population stratification between social classes. Population stratification still persists even in “homogeneous” populations (see references in Richardson and Jones, 2019), and so, the “corrections for” population stratification are anything but.
So what accounts for the small pittance of “variance explained” in GWASs and other similar association studies (Sniekers et al, 2017 “explained” less than 5 percent of variance in IQ)? Population stratification—specifically it is capturing genetic differences that occurred through migration. GWA studies use huge samples in order to find the genetic signals of the genes of small effect that underline the complex trait that is being studied. Take what Noble (2018) says:
As with the results of GWAS (genome-wide association studies) generally, the associations at the genome sequence level are remarkably weak and, with the exception of certain rare genetic diseases, may even be meaningless (13, 21). The reason is that if you gather a sufficiently large data set, it is a mathematical necessity that you will find correlations, even if the data set was generated randomly so that the correlations must be spurious. The bigger the data set, the more spurious correlations will be found (3).
Calude and Longo (2016; emphasis theirs) “prove that very large databases have to contain arbitrary correlations. These correlations appear only due to the size, not the nature, of data. They can be found in “randomly” generated, large enough databases, which — as we will prove — implies that most correlations are spurious.”
So why should we take association studies seriously when they fall prey to the problem of population stratification (measuring differences between social classes and other populations) along with the fact that big datasets lead to spurious correlations? I fail to think of a good reason why we should take these studies seriously. The chopsticks gene example perfectly illustrates the current problems we have with GWASs for complex traits: we are just seeing what is due to social—and other—stratification between populations and not any “genetic” differences in the trait that is being looked at.