3 Comments

>"Now we just need them to repeat this analysis with siblings from all ancestries, not merely the British."

I'd actually expect these results be informative for continental ancestry more generally. They computed over 20 PCs in this sample, and ancestry variation is largely clinal, so if PCs from the 1000 genomes project were projected into this UKBB sample and compared to those from within the sample, I'm sure that at least the first 5 or so 1000-genomes PCs could be estimated pretty well from the UKBB ones.

Suppose you have two questions you're trying to answer:

A) Would a sample of Blacks be as smart as Whites if they were CRISPR'd into having 100%-Euro ancestry?

B) Would a sample of Whites be as dumb as Blacks if they were CRISPR'd into having 100%-African ancestry?

It strikes me that really, these are both actually the same question.

As a non-sequitur, I've been thinking, and I also don't think that the effect of assortative mating is to have symmetric/cancelling biases on inferences of between-group heritability.

So, assume a scenario where Whites and Blacks are equally intelligent but where racemixing whites are dumber than average. This should create a negative correlation between intelligence and european ancestry, yes? Similarly, if the opposite were the case, this would create a positive correlation between the two, right? Well, what happens when racemixing Blacks are smarter than avg in such a scenario? The more African ancestry you have, the less Euro ancestry you have, and so a positive correlation between intelligence and African ancestry is really a negative correlation between intelligence and Euro ancestry. Although you flip the direction of the effect, you also flip which group the effect applies to, and so really, the two effects don't cancel.

Seeing as the expected IQ-ancestry correlation under a scenario of equality isn't zero but is rather negative, this would mean that assortative mating causes between-group heritability to be underestimated.

Expand full comment

By looking only on white british, this study eliminated almost all the variance along the racial dimensions. This has two implications:

* It's kind of dubious that the racial dimensions can be found among the study's top PCs,

* Even if they can be found there, the statistical power to measure their effects would be massively lowered and therefore unlikely to say anything about black-white differences.

Expand full comment
author

Well, it did find race among the first dimensions, but just English (Germanic I guess) vs. Scottish vs. Welsh. We don't know what the latter dimensions mean, maybe they relate to global races, who knows. I guess we could try redoing the study and seeing if they do.

Expand full comment