Really solid work using actual temperature data instead of latitude. The height PGS maintaining significance while cognitive didn't makes sense - Bergmann's rule has way more evolutionary time to work with. I ran into similar linkage disequilibrium issues with cross-ancestral data couple years ago, where measurement error kills statistical power. The 30 degree example helps clarify the effect sizes niceley.
I’d like to add a bit of nuance to what my co-author wrote. Principal components and polygenic scores capture very different things. PCs are an unsupervised summary of genome-wide covariation: they describe how allele frequencies co-vary across individuals and populations, without regard to what those alleles do phenotypically. Polygenic scores, by contrast, are constructed from effect-size–weighted variants, so they are explicitly anchored to a trait; they can therefore be informative about selection on that trait because they aggregate allele-frequency differences in the direction predicted to change the phenotype.
For that reason, PCs are not a direct test of natural selection. On their own, they mainly reflect population structure shaped by ancestry, drift, and demographic history. PCs can become relevant to selection only indirectly, for example, when particular axes of ancestry correlate with geography, climate, migration routes, or other environmental gradients. In that case, you may observe associations between PCs and a trait (or a PGS), but interpreting that as selection requires an additional argument about why that demographic/geographic structure should map onto differential fitness for the trait, rather than being a byproduct of history.
Good post. Just one question though: my understanding is that the people who believe in Cold Winters Theory posit it as a way of explaining race differences in intelligence. If latitude is the mechanism by which different groups differentiate in intelligence, does controlling for ancestry really make sense? I’m not sure if people who believe in Cold Winters Theory necessarily believe it would explain differences within the same group given factors like internal migration.
Because we’re not trying to rank ancestries or groups by intelligence. Our goal is to identify the climatic selection pressures. Natural selection predicts differences above and beyond ancestry alone.
Ancestry is just genes. Same goes for any pgs, which also reflects genes.
Furthermore, ancestry is quantitative or continuous, or rather arbitrary. Your family is an ancestry group relative to other families. As is your extended family, community, nation, race, etc.
So what does it mean to "control" for ancestry? It'd be meaningless, there would be no variance, because ancestry could mean up to 100% of the variance.
So, as a result, what does it mean when we "control for ancestry" via PCs, Fsts, etc? Or rather, to what extent does one control for ancestry? Continental level? Group level?
Each PC is the representation of the main axes of variance of all genes. Despite each PC being uncorrelated to each other, each PC is related to both ancestry and pgs. As one incrementally adds PCs, you're losing real evolutionary signal that has accumulated via selection, because you're working with less variance overall. Control for more PCs again, and theoretically you're working with an even smaller ancestry groups and less variance. Add more PCs after that, again and again, eventually you're working with something ridiculous, think of the variance within your extended family, for example.
So really, your threshold for the number of PCs depends on what sort of level you're expecting an effect to persist. IMO, if the effect disappeared after controlling for PC 1 and PC 2, that would be very suspicious. Those axes mainly reflect lat/long, so it would imply the effect is only on a superficial, geographical level. I think cold winters folks should be controlling for the at least this. As Piffer says, natural selection supersedes, it applies to everyone; from the individual to the group.
So I think it should be controlled for, but with awareness of what it means to do so, etc.
Really solid work using actual temperature data instead of latitude. The height PGS maintaining significance while cognitive didn't makes sense - Bergmann's rule has way more evolutionary time to work with. I ran into similar linkage disequilibrium issues with cross-ancestral data couple years ago, where measurement error kills statistical power. The 30 degree example helps clarify the effect sizes niceley.
I’d like to add a bit of nuance to what my co-author wrote. Principal components and polygenic scores capture very different things. PCs are an unsupervised summary of genome-wide covariation: they describe how allele frequencies co-vary across individuals and populations, without regard to what those alleles do phenotypically. Polygenic scores, by contrast, are constructed from effect-size–weighted variants, so they are explicitly anchored to a trait; they can therefore be informative about selection on that trait because they aggregate allele-frequency differences in the direction predicted to change the phenotype.
For that reason, PCs are not a direct test of natural selection. On their own, they mainly reflect population structure shaped by ancestry, drift, and demographic history. PCs can become relevant to selection only indirectly, for example, when particular axes of ancestry correlate with geography, climate, migration routes, or other environmental gradients. In that case, you may observe associations between PCs and a trait (or a PGS), but interpreting that as selection requires an additional argument about why that demographic/geographic structure should map onto differential fitness for the trait, rather than being a byproduct of history.
Good post. Just one question though: my understanding is that the people who believe in Cold Winters Theory posit it as a way of explaining race differences in intelligence. If latitude is the mechanism by which different groups differentiate in intelligence, does controlling for ancestry really make sense? I’m not sure if people who believe in Cold Winters Theory necessarily believe it would explain differences within the same group given factors like internal migration.
Because we’re not trying to rank ancestries or groups by intelligence. Our goal is to identify the climatic selection pressures. Natural selection predicts differences above and beyond ancestry alone.
This is my take on controlling for ancestry.
Ancestry is just genes. Same goes for any pgs, which also reflects genes.
Furthermore, ancestry is quantitative or continuous, or rather arbitrary. Your family is an ancestry group relative to other families. As is your extended family, community, nation, race, etc.
So what does it mean to "control" for ancestry? It'd be meaningless, there would be no variance, because ancestry could mean up to 100% of the variance.
So, as a result, what does it mean when we "control for ancestry" via PCs, Fsts, etc? Or rather, to what extent does one control for ancestry? Continental level? Group level?
Each PC is the representation of the main axes of variance of all genes. Despite each PC being uncorrelated to each other, each PC is related to both ancestry and pgs. As one incrementally adds PCs, you're losing real evolutionary signal that has accumulated via selection, because you're working with less variance overall. Control for more PCs again, and theoretically you're working with an even smaller ancestry groups and less variance. Add more PCs after that, again and again, eventually you're working with something ridiculous, think of the variance within your extended family, for example.
So really, your threshold for the number of PCs depends on what sort of level you're expecting an effect to persist. IMO, if the effect disappeared after controlling for PC 1 and PC 2, that would be very suspicious. Those axes mainly reflect lat/long, so it would imply the effect is only on a superficial, geographical level. I think cold winters folks should be controlling for the at least this. As Piffer says, natural selection supersedes, it applies to everyone; from the individual to the group.
So I think it should be controlled for, but with awareness of what it means to do so, etc.
I hope it answers your question adequately.