Thank you. I recently started a graph to keep track of the failures and propaganda of the Eugenics movement, and I'm glad to have your article in there. This will grow, and I plan to always have the product to share as I fill it in.
Thank you for an excellent post. I dimly remember being taught not take classical heritability estimates seriously in genetics class. After reading this series, I think I understand why. And I got updated on some really cool research that I would never have read by myself, let alone be able to fully comprehend. We need more science communication like this.
Great post, extremely helpful — thank you Sasha, we needed that.
A question for you: are these newer models still mostly linear? How would you answer the objection that genome expression might be highly non-linear?
(Apart from saying that twin studies also assume linearity... I'm asking the question from the broader perspective of understanding the limitations of the state of the art.)
Yes, almost all of these models are only really interpretable under an assumption that the phenotype is fairly linear and normal-like; with some of the heritability models explicitly assuming the phenotype comes from a multivariate normal. That's on the phenotype side. On the genetics side, however, the evidence for non-linearity of genetic effects *on top* of additive genetic effects is very slim. The heritability of dominance is basically zero, for example. And a recent very large study of height found negligible gene-gene interaction when looking at top variants. Additivity on the genetic side does not mean additivity on the biological side though, it just means that the effect is additive *on average*, each individual almost certainly has extensive biological non-linearity.
There might be a deep explanation for the normal-like behavior (eg, homeostatis can't afford too much non-linearity). This might also be an aspect where social "phenotypes" differ from raw biological phenotypes: wealth can be Pareto, but height can't be. Intelligence might sit somewhere in between, as it involves a complex edifice of cultural scaffolding.
Thanks for writing this article, but If intelligence is not highly genetically heritable, how was it naturally selected for in our past? A trait needs to have meaningful (as in, impacts the organisms life and reproductive success) variance within a population to be selected for.
There's little to no evidence that intelligence has been under strong selection and the evidence that is there suggests stabilizing selection (which is itself very slow). For example from a study led by David Reich:
"Thus, our results provide evidence against a model in which one or a few mutations were responsible for the rapid developments in human behaviour in the last 50,000 years. Instead, changes in lifestyles due to cultural innovation or exposure to new environments are likely to have been driving forces behind the rapid transformations in human behaviour in the last 50,000 years"
So it may have been under slow selection for a very long time (100s of thousands of years), or the variants that were under selection fixed quickly prior to 50kya and were then present/absent in nearly all humans, or it is under complex stabilizing selection that is very hard to identify.
> our results provide evidence against a model in which one or a few mutations were responsible for the rapid developments in human behaviour in the last 50,000 years. Instead, changes in lifestyles due to cultural innovation or exposure to new environments are likely to have been driving forces behind the rapid transformations in human behaviour in the last 50,000 years
I already didn't believe what that quote argues against, and it doesn't actually address my question very well. I don't think your sentence is backed up by what's written.
> There's little to no evidence that intelligence has been under strong selection and the evidence that is there suggests stabilizing selection (which is itself very slow). For example from a study led by David Reich:
I don't have access to the article, and the abstract doesn't help that line of reasoning. If part of it supports your above sentence can you quote it?
Given all the factors that go into intelligence, (eg something as unrelated as immune system, uptake of any nutrient shown to impact intelligence... all of them?, etc.), it seems implausible that genes impacting intelligence could have ALL reached fixation across groups. In fact I think the abstract helps my case here:
> Our analysis reveals key features of the landscape of human genome variation, including that the rate of accumulation of mutations has accelerated by about 5% in non-Africans compared to Africans since divergence. We show that the ancestors of some pairs of present-day human populations were substantially separated by 100,000 years ago, well before the archaeologically attested onset of behavioural modernity. We also demonstrate that indigenous Australians, New Guineans and Andamanese do not derive substantial ancestry from an early dispersal of modern humans; instead, their modern human ancestry is consistent with coming from the same source as that of other non-Africans.
Hmm, I guess I don't understand your question if a section titled "No species-wide sweeps in modern humans" is not addressing it. Perhaps you can clarify what specific examples of selection on intelligence you are talking about.
I think we have seen this in everyday life. Einstein's family wasn't spectacular, there was a series of failed business enterprises etc. his son, Hans Albert was an esteemed professor but not really anyone famous. His other son ended up in an insane asylum.
We just don't see these "Nobel Prize clans" anywhere. James Woods, IQ 190, mother ran a preschool, father was a CIA officer. And so on. What we see is moderately intellectual parents who probably did emphasize education, but not stellar successes.
> Surprisingly, while these findings made a big splash among geneticists and were seen as essentially a coup de grâce in the heritability debate
This is a big claim and perhaps the most important claim in your post. Yet it is not substantiated with any sources. What's the evidence that the view "among geneticists" is that this paper has ended the "heritability debate"?
I read through both articles, though in this one I skipped over some sections, and I did not see a lot of discussion on IQ test design itself and the variability it can introduce (e.g., SAT vs Raven's). You do mention how tests are normalized to a population, but I didn't see anything discussing how many "IQ" tests have strong assumptions about education/culture baked into the tests themselves.
The worst I have seen of this is people calling the SAT an "IQ test" because it is "correlated with positive life outcomes", even though it is essentially a knowledge test at best, memorization test at worst.
Has any of the types of analysis you discuss been done on a limited dataset that tries to only use tests with as little dependence on life experience as possible (like Raven's)? An ideal test seems like one you can give to a child that is old enough to see and point at things and can understand the concept of "correct answer", but otherwise has no educational/cultural history. If so, do the results come out the same?
I agree that IQ tests have very strong assumptions and it is difficult to make any definitive statements because the tests can be very different. Even though they are not intended to test "general knowledge" they are often still highly correlated with culturally-specific questions: http://gusevlab.org/projects/hsq/#h.pqhr9q81wgd5
I talk about measurement error a bit at the end here:
I've yet to see an example where GWAS heritability increased substantially because a better IQ test was used. In instances where people have looked at more "fluid" like subtests, like memory or reaction time, the heritability of those is *much* lower than that of an overall IQ test or a general factor. For example, see Supp Table 2 of https://pubmed.ncbi.nlm.nih.gov/32895543/, where the Verbal Numeric Reasoning test (which includes general knowledge / SAT-type questions) has much higher heritability than all of the more basic subtests. We do not have direct effect estimates for these subtests but in general it looks like the less cultural a test is, the less heritable it is.
I think an underrated aspect of this discussion is that even if the 'true' heritability were substantially higher (due to e.g. private variants within families), given the information we currently have there would be strong limits on our ability to build predictors anyway (which is more-or-less the whole point of this estimand).
Right, one cold hard truth is that we currently do not have any methods or data to outperform the prediction accuracy set by the GWAS heritability + the rare burden heritability. Even for people who believe there is missing heritability still out there, we would not currently have any way to leverage it for prediction. I do think the broader biological questions of how much rare variant heritability there is in truth and how much Gene-Environment interaction are interesting, of course.
Thank you. I recently started a graph to keep track of the failures and propaganda of the Eugenics movement, and I'm glad to have your article in there. This will grow, and I plan to always have the product to share as I fill it in.
https://embed.kumu.io/8ef99af0bd6e6efa398aef8698828a5a
"Robert Plomin tends to make shit up"
Sometimes the truth hurts.
Thank you for an excellent post. I dimly remember being taught not take classical heritability estimates seriously in genetics class. After reading this series, I think I understand why. And I got updated on some really cool research that I would never have read by myself, let alone be able to fully comprehend. We need more science communication like this.
Great post, extremely helpful — thank you Sasha, we needed that.
A question for you: are these newer models still mostly linear? How would you answer the objection that genome expression might be highly non-linear?
(Apart from saying that twin studies also assume linearity... I'm asking the question from the broader perspective of understanding the limitations of the state of the art.)
Yes, almost all of these models are only really interpretable under an assumption that the phenotype is fairly linear and normal-like; with some of the heritability models explicitly assuming the phenotype comes from a multivariate normal. That's on the phenotype side. On the genetics side, however, the evidence for non-linearity of genetic effects *on top* of additive genetic effects is very slim. The heritability of dominance is basically zero, for example. And a recent very large study of height found negligible gene-gene interaction when looking at top variants. Additivity on the genetic side does not mean additivity on the biological side though, it just means that the effect is additive *on average*, each individual almost certainly has extensive biological non-linearity.
Thank you Sasha, very clear and informative.
There might be a deep explanation for the normal-like behavior (eg, homeostatis can't afford too much non-linearity). This might also be an aspect where social "phenotypes" differ from raw biological phenotypes: wealth can be Pareto, but height can't be. Intelligence might sit somewhere in between, as it involves a complex edifice of cultural scaffolding.
Thanks for writing this article, but If intelligence is not highly genetically heritable, how was it naturally selected for in our past? A trait needs to have meaningful (as in, impacts the organisms life and reproductive success) variance within a population to be selected for.
There's little to no evidence that intelligence has been under strong selection and the evidence that is there suggests stabilizing selection (which is itself very slow). For example from a study led by David Reich:
"Thus, our results provide evidence against a model in which one or a few mutations were responsible for the rapid developments in human behaviour in the last 50,000 years. Instead, changes in lifestyles due to cultural innovation or exposure to new environments are likely to have been driving forces behind the rapid transformations in human behaviour in the last 50,000 years"
https://www.nature.com/articles/nature18964#Sec7
So it may have been under slow selection for a very long time (100s of thousands of years), or the variants that were under selection fixed quickly prior to 50kya and were then present/absent in nearly all humans, or it is under complex stabilizing selection that is very hard to identify.
> our results provide evidence against a model in which one or a few mutations were responsible for the rapid developments in human behaviour in the last 50,000 years. Instead, changes in lifestyles due to cultural innovation or exposure to new environments are likely to have been driving forces behind the rapid transformations in human behaviour in the last 50,000 years
I already didn't believe what that quote argues against, and it doesn't actually address my question very well. I don't think your sentence is backed up by what's written.
> There's little to no evidence that intelligence has been under strong selection and the evidence that is there suggests stabilizing selection (which is itself very slow). For example from a study led by David Reich:
I don't have access to the article, and the abstract doesn't help that line of reasoning. If part of it supports your above sentence can you quote it?
Given all the factors that go into intelligence, (eg something as unrelated as immune system, uptake of any nutrient shown to impact intelligence... all of them?, etc.), it seems implausible that genes impacting intelligence could have ALL reached fixation across groups. In fact I think the abstract helps my case here:
> Our analysis reveals key features of the landscape of human genome variation, including that the rate of accumulation of mutations has accelerated by about 5% in non-Africans compared to Africans since divergence. We show that the ancestors of some pairs of present-day human populations were substantially separated by 100,000 years ago, well before the archaeologically attested onset of behavioural modernity. We also demonstrate that indigenous Australians, New Guineans and Andamanese do not derive substantial ancestry from an early dispersal of modern humans; instead, their modern human ancestry is consistent with coming from the same source as that of other non-Africans.
Hmm, I guess I don't understand your question if a section titled "No species-wide sweeps in modern humans" is not addressing it. Perhaps you can clarify what specific examples of selection on intelligence you are talking about.
I think we have seen this in everyday life. Einstein's family wasn't spectacular, there was a series of failed business enterprises etc. his son, Hans Albert was an esteemed professor but not really anyone famous. His other son ended up in an insane asylum.
We just don't see these "Nobel Prize clans" anywhere. James Woods, IQ 190, mother ran a preschool, father was a CIA officer. And so on. What we see is moderately intellectual parents who probably did emphasize education, but not stellar successes.
> Surprisingly, while these findings made a big splash among geneticists and were seen as essentially a coup de grâce in the heritability debate
This is a big claim and perhaps the most important claim in your post. Yet it is not substantiated with any sources. What's the evidence that the view "among geneticists" is that this paper has ended the "heritability debate"?
I read through both articles, though in this one I skipped over some sections, and I did not see a lot of discussion on IQ test design itself and the variability it can introduce (e.g., SAT vs Raven's). You do mention how tests are normalized to a population, but I didn't see anything discussing how many "IQ" tests have strong assumptions about education/culture baked into the tests themselves.
The worst I have seen of this is people calling the SAT an "IQ test" because it is "correlated with positive life outcomes", even though it is essentially a knowledge test at best, memorization test at worst.
Has any of the types of analysis you discuss been done on a limited dataset that tries to only use tests with as little dependence on life experience as possible (like Raven's)? An ideal test seems like one you can give to a child that is old enough to see and point at things and can understand the concept of "correct answer", but otherwise has no educational/cultural history. If so, do the results come out the same?
I agree that IQ tests have very strong assumptions and it is difficult to make any definitive statements because the tests can be very different. Even though they are not intended to test "general knowledge" they are often still highly correlated with culturally-specific questions: http://gusevlab.org/projects/hsq/#h.pqhr9q81wgd5
I talk about measurement error a bit at the end here:
https://theinfinitesimal.substack.com/i/148251755/what-about-measurement-error
I've yet to see an example where GWAS heritability increased substantially because a better IQ test was used. In instances where people have looked at more "fluid" like subtests, like memory or reaction time, the heritability of those is *much* lower than that of an overall IQ test or a general factor. For example, see Supp Table 2 of https://pubmed.ncbi.nlm.nih.gov/32895543/, where the Verbal Numeric Reasoning test (which includes general knowledge / SAT-type questions) has much higher heritability than all of the more basic subtests. We do not have direct effect estimates for these subtests but in general it looks like the less cultural a test is, the less heritable it is.
I think an underrated aspect of this discussion is that even if the 'true' heritability were substantially higher (due to e.g. private variants within families), given the information we currently have there would be strong limits on our ability to build predictors anyway (which is more-or-less the whole point of this estimand).
Right, one cold hard truth is that we currently do not have any methods or data to outperform the prediction accuracy set by the GWAS heritability + the rare burden heritability. Even for people who believe there is missing heritability still out there, we would not currently have any way to leverage it for prediction. I do think the broader biological questions of how much rare variant heritability there is in truth and how much Gene-Environment interaction are interesting, of course.