First off, a big thank you for everyone who read and retweeted the blog post by myself, @JenniferRaff, @AdamRutherford and @aylwyn_scally on Race, Genetics and IQ (link here: ewanbirney.com/2019/10/race-g…). As well as the praise it has also attracted some complaints
This is a slightly tiresome world of having people who are seemingly quite convinced that our thesis is wrong/misguided/doesn't represent the facts and throw a variety of stones at this. Most is pretty incoherent, but some is worth stepping people back through arguments.
As with the original blog post, I hold out little hope in changing the mind of these people - they are quite adamant in their positions and delight in the smallest chink of uncertainty or cases where there are unanswered questions or statistical model shortfalls.
However, I do hope stepping people through the arguments help provide people with a more nagging "I still think there must be something in this genetics + race + IQ thing, even if it is a low level" more explanation of what is going on
(Finally meta-comment; it is amusing to me that some people claim we don't understand genetics and human population history. I am aware of my intellectual limitations - which is why I always enjoy hanging out with people who know more than me in all sorts of topics >>
in terms of population history - that's @aylwyn_scally and @JenniferRaff ; in terms of IQ genetics it's @StuartJRitchie and KCL/Edinbrugh crew(s). And ... I do 'alright' as a broad based genomicist / geneticist, knowing my human and animal quantitative trait stuff pretty well)
Anyway, onwards into this mess. Many people who support 'human biodiversity' (modern name) or 'biological basis of race' (old school) point to a variety of papers published outside of mainstream journals but still looking pretty scientific (abstract, methods, plots, pvalues etc)
There are three main types in my view, all with a similar vein. Ones which plot mean IQ levels in groups with labels - labels from the study or from countries, sometimes regressing to distance to mean IQ from East Africa.
Another does skin pigmentation levels as a proxy for ancestry (!) against IQ in US individuals; another does actual SNP chip ancestry against IQ in US individuals. All of these have a "African countries/African ancestry has less IQ" and something else (eg, Chinese) as far better.
To be clear, I think these are deeply flawed studies in terms of saying anything about genetics. I think they are probably also pretty deeply flawed about social/educational process exploration as well, but I am pretty sure of their flaws with genetics.
The first, and most major flaw, is that human genetics (like all genetics) just doesn't map onto groups (any groups) - genetics maps onto individuals with a genome that has a series of tree-like (ancestral recombination graph) relationships to other individuals. (also see below)
The other major flaw in all of these is that we *know* there is more than genetics going on for educational processes - between (global) countries this is huge and obvious; inside of western world (US, UK, France, etc) this is also very clear.
As such these studies just ... can't tell you about genetic basis of means differences. I get this is annoying if this was the thing you wanted to study, but such is the world of observational studies - quite often the world is not set up the way you want to ask your question
At this point, you might scratch your head and ask "well, how do people do genetics of IQ anyway then?" and for the more GWAS cognescent people you might bristle at the idea that genetics is not done on groups - "surely GWAS does this all the time?"
This is best stepped through in my view from the perspective of animal breeding genetics first, then go to human. In animal breeding they have hideously complex relationships between individuals, and, crucially, they can make a strong assumption of randomisation of environment
(basically - Cows don't choose where they live; Cows also don't get to choose their mates; and as an extra bonus for genetics, single male bulls have insane numbers of children, often on very different farms and fields via artificial insemination)
Because of this, in animal breeding they can model a pretty full representation of the genetic structure of the individuals, and be confident that when that model fits, it is due to genetics not due to something else correlating with this genetic structure.
This process is so good that they can throw out the "foreground" genetics - aka trying to find the actual loci of interest - and let this "background genetic model" do all the heavy lifting (this is called "genomic selection"
In humans we absolutely cannot bring a strong randomisation of the environment prior to humans wrt to their genetics. This falls down on even a trivial look; it falls down massively in big samples of people (which for example span different self identified ethnicities)
In the late 90s/early 2000s there were some spectacular fails of trying to do population scale genetics without thinking this through which had the whole genome lit up as "somehow involved in this process" - something we've learnt as diagnostic of "statistical models doesn't fit"
But, people realised that if one subsetted samples of humans to the "biggest, modal bit of genetics in my sample" and one threw in lots of principle components from the genetics (an even cruder "background model" of genetics) then it might work...
The subsetting does two things (a) it produces a more consistent environment of people, basically saying that complex ancestries of humans are likely to also have complex environmental differences - think - wealth, discrimination (b) allows this PC trick to work by being smoother
It's important to realise that although both of these techniques leverage genetic data, it is really to homogenise and model environmental effects in human society, and by "environmental effects" we mean - wealth, discrimination etc as well as water quality
How are human geneticists reasonable sure this modelling works? The key reason is that the outputted results of a GWAS stop implicating *all* of the genome, but rather implicate only a subset, and that the "null model" of genetics being independent of environment fits the rest.
(this is visualised in the QQ plot, which is why human geneticists flash up their "good" QQ plots a lot, and as the statsitical sample size grows you have to move it into a modelling framework to test right - so called LD score regression and friends)
Frustratingly human geneticists have used some everyday words often used to describe ethnicity "European" or "East Asian" to talk about this subsetting process. Self identified ethnicity is non-random wrt to these groups, but I think this use of these everyday words is misleading
It's misleading for two reasons. Firstly because it makes people think that this subsetting is important about modelling the *genetics* of these people, when in fact the modelling is aiming to achieve randomisation of the environment wrt to the genetics. >>
Secondly it is misleading because a large fraction (15-30%) of people who confidently self identify in these groups, eg "White European", "European American", "Han Chinese" don't get include in this subsetting procedure (they are *very different* things).
As my long time followers know, I think we'd be better off change our words on the science side here; my current stop gap solution is the pluralise ancestry, eg, "European Ancestries" though ... this is not perfect
So - this "subset and Principle Components" trick works for ... heart attacks, type II diabetes, height, and ... pretty much any IQ or educational measure one might want to look at (IQ scores if measured, standardised exam results, number of years in education)
In fact, all these educational measures look reasonably standard in terms of their genetic trait architecture (very polygenic - no surprise - nothing too odd going on). The big thing though is that it is clear that parental environment is a big factor
Because parents both create the environment for their offspring (aka "being parents") *and* pass down 50% of their genetic material, the parental environment and direct genetics are confounded. As a tangent - It's actually quite cool we can tease this apart - but not for now
You might be skeptical the same "trick" of subsetting people and using PCs can "effectively randomise" the educational environment with all its complexity for people (say, in the UK, or Denmark, or South Korea). Unsurprisingly IQ/EA geneticists worry about this *alot*
Probably the most convincing evidence that they have something biologically solid is that is that genetic models built in one country (eg, UK) correlate/predict (synonymous for genetics) in another (eg, Denmark).
For this to be due to non-random environment/genetic effects one has to have correlated non-random environment/genetic effects at *specific points in the genome* between these countries. Thats... hard to construct something that would give rise to that.
Circling back: there is strong evidence for a genetic component to IQ/Eductional Attainment. Genetics is also involved in visible characterisitics (skin pigmentation, hair type), *but* these two features *do not mean* that skin pigmentation/hair type can predict IQ/Education.
