An important point that I don't think is obvious to most non-experts is that the GWAS SNP studies and limits that Sasha points to all fundamentally assume a purely linear model. Yes, that's how all genetic prediction is done today but it's not a fundamental limitation on heritability or our potential ability to improve on this in the future (Sasha believes this is unlikely/far off while I'd argue that the fundamental technical advance we've made in AI recently is the ability to fit large, highly interdependent/non-linear models to massive data sets via nueral net techniques)
There is every reason to believe that the effects of genes on IQ is highly non-linear (especially if mediated by complex environmental effects) and whether it matters these limits assume a certain limited model depends alot on why you care.
I don't think the argument here is throw out twin studies (or at least it shouldn't be). It's just that it seems likely they capture some passive GxE which is likely more relevant to the expression of cognitive phenotypes. So I do think whatever the "true" heritability of intelligence is will be meaningfully lower than physical traits like height (This isn't saying that genes don't matter - they of course do).
The thing that draws my concern with some of the above is the same missing heritability discrepancy exists for autism, where there obviously is significantly less room for GxE or other confounders (measurement issues still a thing though). Now, rare variation likely matters more for autism, but there's quite a mystery here.
You'll have to flesh out your point a bit more. There isn't a general reduction in expression in autistic brians. Plenty of genetic models of ASD show different patterns of DGE. Some of this is perhaps a function of calling different types of neurodev issues autism but even if we look at autism without known rare/de novo variants I'd imagine this same pattern is observed. There are just many ways to disrupt finely tuned social phenotypes.
>>The error bars for the population and within-sibship IQ h^2 overlapped. Thus, this study does not move the boulder of twin studies.
Is not true. Non-overlapping error bars means beyond the 1%, with uneven error bars and partial overlap, it's hard to say what the p value is without doing the math. Fortunately, the authors did this in their supplementary table, sheet 5, which gives p = .011 for this comparison. So Gusev is right there is some shrinkage at the 5% significance, though because of the uncertainty, it's hard to say how large it is exactly.
Per the same study, figure 4 shows the shrinkage in polygenic scores, which has 2 estimates for cognition, one of which overlaps with 0, and one of which doesn't. Best guess for shrinkage there is about 20%, the value for height is about 10% (significant), so cognition is closer to height than to education in amount of confounding (or the proportion of direct effects).
I'm getting from a simple Z test (.24-.14)/sqrt((.09^2 + .06^2)/(1.96^2)) = 1.81 -> p > 0.05. Their "heterogeneity p" is probably chosen because it gives them lower values. Researcher degrees of freedom and replication crisis, etc
Even though the bars overlap, there is a statistically significant difference of p=0.011 between the environmentally confounded population heritability estimate for cognition and the within-sibship estimate, which controls for demographic and other confounds. So Sasha is right.
"The within-sibship SNP heritability point estimate for educational attainment attenuated by 76% from the population estimate (population h2: 0.13; within-sibship h2: 0.04; difference P = 5.3 × 10−26), with attenuations also observed for cognitive ability (population h2: 0.24; within-sibship h2: 0.14; attenuation 44%; difference P = 0.011), ever smoking (population h2: 0.10; within-sibship h2: 0.07; attenuation 25%; difference P = 0.029) and height (population h2: 0.41; within-sibship h2: 0.34; attenuation 17%; difference P = 1.6 × 10−3)."
>Even though the bars overlap, there is a statistically significant difference of p=0.011 between the environmentally confounded population heritability estimate for cognition and the within-sibship estimate, which controls for demographic and other confounds. So Sasha is right.
It's more like 3% or 4% with a better test than the authors used, and you're avoiding the point that the between families height PGS was more confounded, when Sasha said otherwise and his point was that IQ is unique in being confounded between families.
Look at supplementary table 2 for the study. The sample size for height (149K) was much larger than for cognition (27K), which explains why the p value for height was lower.
Sasha shows that there’s a much larger difference between the cognition and heritability point estimates. The point estimate is the best measure of a parameter, so what Sasha is saying valid.
A study that finds a heritability of 0.19 for cognitive performance, about half of the heritability for height, scarcely seems like a ringing vindication of hereditarianism and supports Sashas's point that IQ is much less heritable than height. I wrote a substantial article pointing to converging evidence that IQ heritability is much lower than twin studies estimated. You are welcome to comment on it. https://open.substack.com/pub/eclecticinquiries/p/twin-studies-exaggerate-iq-heritability?r=4952v2&utm_campaign=post&utm_medium=web
Height heritability was also probably overestimated by twin studies, as I review in my article. The argument goes:
"Advances in GWAS methodology will probably lead to higher heritability estimates for IQ, but it is virtually certain that they will be much lower than those derived from twin/adoption studies. GWAS are already measuring the influence of millions of gene variants. GWAS have already found a moderately strong heritability for height of 45%. There is every reason to believe that if twin/adoption studies were correct about IQ heritability, GWAS would have found higher heritability by now."
>Note two things. First, the p value for height is GREATER than that of IQ in this study, yet Gusev represents it like this:
I forgot this part. No, the height one is 0.0016, so it is smaller. It's a bit unclear if you meant "greater significance" which presumably means greater log10(-p) value meaning more evidence.
Flynn Effect with it been provided not to be the full story, the iq peaking. Do you think height is the same way where we had a peak and now it is falling due to Genetics of the people having kids?
What does this tell you about the future of the West (and Asia)?
I think the increase of Height and Iq in the west have (I willing to be told I am wrong on this) as the calories and other things increase in the 20th century, western people could reach their maximum genetic peak (in terms of height, iq and other things such as sports abilities).
The thing is that genetics of who is having kids are on average shorter and lower iqs (including within Race), calories especially in the west have not decrease that much.
A big part of it I am whiling to be proven wrong is on this is the type of women having children. Studies on this already proven that taller women are having less children and also higher iq women are also having less children. Yes height and iq are not fully related only 0.2, but the height, iq and other things are strongly more related to the mother than the father.
It is interesting that now the calories and freedom given to women is changing the gene pool. The future even without the food supply changing is shorter lower iq people (At least until the famine, war or other things change it).
Flynn Effect didn't vary by race so it's irrelevant to race differences, as is everything in the present article. Average polygenic scores by race consistently align with race differences in phenotypic IQ.
India is probably 95, but Pakistan is definitely lower due to a 60+% 1st cousin marriage rate. Indian IQ in England has been measured at 98 and Pakistani IQ in England at 91.
I don't think the effect of kinship marriage on the intelligence of the population is severe, the Ashkenazi Jews did it for a long time without becoming less intelligent.
Depending on the ethnic group I looked at, Pakistanis had very different polygenic scores for intelligence, for example, Pashtuns were a full standard deviation below whites, while Kalash people were no different from whites.
But most of the Pakistani immigrants should be Punjabi, so 90 to 95.
The whole of the argument of Jensen and that crowd depends on the notion that IQ is strongly heritable. That's why they spend so much time talking about twin studies. And there are more and more reasons over time to think that the estimates of heritability from twin studies were hugely exaggerated. The Flynn Effect is one of these, the GWAS studies another. Twin studies weren't capable of measuring the most important environmental effects on our IQ, the individual and social multipliers that I discuss in my article.
Flynn effect was probably caused by improve nutrition or more familiarity with testing which impacted both blacks and whites equally. There may be individual multiplication through the creation of one's own environment but this is downstream of genes and properly counted as a part of broad sense heritability in twin studies.
If you think that IQ scores are mostly measuring test taking skills, then you don't believe it is strongly heritable either. What change in nutrition could have caused a 20 point rise in Dutch IQs between 1952 and 1982? If anyone has a theory about that I'd like to know. I think the more likely explanation is that IQ is mostly a developed rather than an innate trait and doesn't really measure intelligence. The GWAS studies are just confirming that the twin studies hugely exaggerate heritability.
Flynn did the math, and it's pretty hard to believe. Was there really a four sigma improvement in environment? I feel like I have the right to ask someone to show me in detail how that happened:
Dutch 18-year-old men gained 20 [IQ] points (1.33 SDs) between 1952 and 1982. By this logic [that is, the logic of strong heritability], a minimum of a 2.67 standard deviation gain in environmental quality would be necessary to account for their IQ gains. If we take into account that the passage of time cannot have had much effect on some significant fraction of environmental causes, the necessary gain for relevant environmental factors might be as much or more than 4 standard deviations. So, assuming a normal distribution for environments, the average Dutch man of 1982 must have had an environment whose quality was well into the highest percentile of the 1952 Dutch distribution.
An important point that I don't think is obvious to most non-experts is that the GWAS SNP studies and limits that Sasha points to all fundamentally assume a purely linear model. Yes, that's how all genetic prediction is done today but it's not a fundamental limitation on heritability or our potential ability to improve on this in the future (Sasha believes this is unlikely/far off while I'd argue that the fundamental technical advance we've made in AI recently is the ability to fit large, highly interdependent/non-linear models to massive data sets via nueral net techniques)
There is every reason to believe that the effects of genes on IQ is highly non-linear (especially if mediated by complex environmental effects) and whether it matters these limits assume a certain limited model depends alot on why you care.
I don't think the argument here is throw out twin studies (or at least it shouldn't be). It's just that it seems likely they capture some passive GxE which is likely more relevant to the expression of cognitive phenotypes. So I do think whatever the "true" heritability of intelligence is will be meaningfully lower than physical traits like height (This isn't saying that genes don't matter - they of course do).
The thing that draws my concern with some of the above is the same missing heritability discrepancy exists for autism, where there obviously is significantly less room for GxE or other confounders (measurement issues still a thing though). Now, rare variation likely matters more for autism, but there's quite a mystery here.
It's likely due to various factors affecting gene expression, for both IQ and autism.
https://www.researchgate.net/publication/271507751_A_Theory_of_General_Reduction_of_Gene-Expression_Manifesting_as_Autism_1993_with_2014_revisions_of_presentation_-_now_2020_rev
You'll have to flesh out your point a bit more. There isn't a general reduction in expression in autistic brians. Plenty of genetic models of ASD show different patterns of DGE. Some of this is perhaps a function of calling different types of neurodev issues autism but even if we look at autism without known rare/de novo variants I'd imagine this same pattern is observed. There are just many ways to disrupt finely tuned social phenotypes.
>>The error bars for the population and within-sibship IQ h^2 overlapped. Thus, this study does not move the boulder of twin studies.
Is not true. Non-overlapping error bars means beyond the 1%, with uneven error bars and partial overlap, it's hard to say what the p value is without doing the math. Fortunately, the authors did this in their supplementary table, sheet 5, which gives p = .011 for this comparison. So Gusev is right there is some shrinkage at the 5% significance, though because of the uncertainty, it's hard to say how large it is exactly.
Per the same study, figure 4 shows the shrinkage in polygenic scores, which has 2 estimates for cognition, one of which overlaps with 0, and one of which doesn't. Best guess for shrinkage there is about 20%, the value for height is about 10% (significant), so cognition is closer to height than to education in amount of confounding (or the proportion of direct effects).
I'm getting from a simple Z test (.24-.14)/sqrt((.09^2 + .06^2)/(1.96^2)) = 1.81 -> p > 0.05. Their "heterogeneity p" is probably chosen because it gives them lower values. Researcher degrees of freedom and replication crisis, etc
What is a simple z-test and why would you think it's correct here?
https://en.wikipedia.org/wiki/Z-test
On my simulation the bootstrapped distribution of R^2s they get for cognition should be very close to normally distributed since it's far away from 0.
Z test gives the mean of the distribution with SD = 1 of what you would expect to see if you subtracted the bootstrapped distribution of X and Y.
So the mean of the Z distribution is 1.81 with SD 1.
Although this should probably be 1 tailed so it would be p =~ 3%.
Even though the bars overlap, there is a statistically significant difference of p=0.011 between the environmentally confounded population heritability estimate for cognition and the within-sibship estimate, which controls for demographic and other confounds. So Sasha is right.
"The within-sibship SNP heritability point estimate for educational attainment attenuated by 76% from the population estimate (population h2: 0.13; within-sibship h2: 0.04; difference P = 5.3 × 10−26), with attenuations also observed for cognitive ability (population h2: 0.24; within-sibship h2: 0.14; attenuation 44%; difference P = 0.011), ever smoking (population h2: 0.10; within-sibship h2: 0.07; attenuation 25%; difference P = 0.029) and height (population h2: 0.41; within-sibship h2: 0.34; attenuation 17%; difference P = 1.6 × 10−3)."
For more on GWAS and how we can be confident that twin studies overestimate heritability, see my latest: https://open.substack.com/pub/eclecticinquiries/p/twin-studies-exaggerate-iq-heritability?r=4952v2&utm_campaign=post&utm_medium=web
>Even though the bars overlap, there is a statistically significant difference of p=0.011 between the environmentally confounded population heritability estimate for cognition and the within-sibship estimate, which controls for demographic and other confounds. So Sasha is right.
It's more like 3% or 4% with a better test than the authors used, and you're avoiding the point that the between families height PGS was more confounded, when Sasha said otherwise and his point was that IQ is unique in being confounded between families.
Look at supplementary table 2 for the study. The sample size for height (149K) was much larger than for cognition (27K), which explains why the p value for height was lower.
Sasha shows that there’s a much larger difference between the cognition and heritability point estimates. The point estimate is the best measure of a parameter, so what Sasha is saying valid.
Looks like you were wrong: https://x.com/BronskiJoseph/status/1843419025185050686
I recommend this book for frequentism: https://egrcc.github.io/docs/math/all-of-statistics.pdf
A study that finds a heritability of 0.19 for cognitive performance, about half of the heritability for height, scarcely seems like a ringing vindication of hereditarianism and supports Sashas's point that IQ is much less heritable than height. I wrote a substantial article pointing to converging evidence that IQ heritability is much lower than twin studies estimated. You are welcome to comment on it. https://open.substack.com/pub/eclecticinquiries/p/twin-studies-exaggerate-iq-heritability?r=4952v2&utm_campaign=post&utm_medium=web
https://pbs.twimg.com/media/GZYCRHwXAAAeiFE?format=jpg&name=medium
Height heritability was also probably overestimated by twin studies, as I review in my article. The argument goes:
"Advances in GWAS methodology will probably lead to higher heritability estimates for IQ, but it is virtually certain that they will be much lower than those derived from twin/adoption studies. GWAS are already measuring the influence of millions of gene variants. GWAS have already found a moderately strong heritability for height of 45%. There is every reason to believe that if twin/adoption studies were correct about IQ heritability, GWAS would have found higher heritability by now."
>Note two things. First, the p value for height is GREATER than that of IQ in this study, yet Gusev represents it like this:
I forgot this part. No, the height one is 0.0016, so it is smaller. It's a bit unclear if you meant "greater significance" which presumably means greater log10(-p) value meaning more evidence.
> It's a bit unclear if you meant "greater significance" which presumably means greater log10(-p) value meaning more evidence.
Yes, I meant this, this is a typo. fixed
Flynn Effect with it been provided not to be the full story, the iq peaking. Do you think height is the same way where we had a peak and now it is falling due to Genetics of the people having kids?
What does this tell you about the future of the West (and Asia)?
An article on the average height (https://priceonomics.com/where-are-the-tallest-people-in-the-world/). It say that the USA average height has not gone up since the 1960s and in some estimates it is falling.
I think the increase of Height and Iq in the west have (I willing to be told I am wrong on this) as the calories and other things increase in the 20th century, western people could reach their maximum genetic peak (in terms of height, iq and other things such as sports abilities).
The thing is that genetics of who is having kids are on average shorter and lower iqs (including within Race), calories especially in the west have not decrease that much.
A big part of it I am whiling to be proven wrong is on this is the type of women having children. Studies on this already proven that taller women are having less children and also higher iq women are also having less children. Yes height and iq are not fully related only 0.2, but the height, iq and other things are strongly more related to the mother than the father.
It is interesting that now the calories and freedom given to women is changing the gene pool. The future even without the food supply changing is shorter lower iq people (At least until the famine, war or other things change it).
The articles on women having children
Taller women having fewer children
(https://pubmed.ncbi.nlm.nih.gov/22410858/#:~:text=Conclusions%3A%20We%20conclude%20that%20shorter,of%20height%20on%20child%20survival.)
Higher iq people having less children
(https://pubmed.ncbi.nlm.nih.gov/25131282/)
Iq is more related to the mother
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5800314/
Call me simple-minded but I think that the reason that IQ always shows lower heritability than height in GWAS studies is that it's just less heritable. Even before these GWAS studies came out, adoption studies and the Flynn Effect showed that estimates of IQ heritability derived from twin studies had to be wrong, as I argue here: https://open.substack.com/pub/eclecticinquiries/p/on-race-racism-iq-and-heritability?r=4952v2&utm_campaign=post&utm_medium=web
Flynn Effect didn't vary by race so it's irrelevant to race differences, as is everything in the present article. Average polygenic scores by race consistently align with race differences in phenotypic IQ.
You mention the gap narrowing also but it's not true https://www.emilkirkegaard.com/p/a-plethora-of-evidence-for-genetic
The polygenic scores do not fully align with phenotypic IQ .
For example, the IQ of Indians and Pakistanis has been reported as 80, but their polygenic score for intelligence is around 95.
India is probably 95, but Pakistan is definitely lower due to a 60+% 1st cousin marriage rate. Indian IQ in England has been measured at 98 and Pakistani IQ in England at 91.
I don't think the effect of kinship marriage on the intelligence of the population is severe, the Ashkenazi Jews did it for a long time without becoming less intelligent.
Depending on the ethnic group I looked at, Pakistanis had very different polygenic scores for intelligence, for example, Pashtuns were a full standard deviation below whites, while Kalash people were no different from whites.
But most of the Pakistani immigrants should be Punjabi, so 90 to 95.
Can you give source for Kalash and Pashtuns scores?
The whole of the argument of Jensen and that crowd depends on the notion that IQ is strongly heritable. That's why they spend so much time talking about twin studies. And there are more and more reasons over time to think that the estimates of heritability from twin studies were hugely exaggerated. The Flynn Effect is one of these, the GWAS studies another. Twin studies weren't capable of measuring the most important environmental effects on our IQ, the individual and social multipliers that I discuss in my article.
Flynn effect was probably caused by improve nutrition or more familiarity with testing which impacted both blacks and whites equally. There may be individual multiplication through the creation of one's own environment but this is downstream of genes and properly counted as a part of broad sense heritability in twin studies.
https://www.sciencedirect.com/science/article/pii/S0160289622000241
If you think that IQ scores are mostly measuring test taking skills, then you don't believe it is strongly heritable either. What change in nutrition could have caused a 20 point rise in Dutch IQs between 1952 and 1982? If anyone has a theory about that I'd like to know. I think the more likely explanation is that IQ is mostly a developed rather than an innate trait and doesn't really measure intelligence. The GWAS studies are just confirming that the twin studies hugely exaggerate heritability.
>What change in nutrition could have caused a 20 point rise in Dutch IQs between 1952 and 1982?
Same one that increased height by multiple inches
Flynn did the math, and it's pretty hard to believe. Was there really a four sigma improvement in environment? I feel like I have the right to ask someone to show me in detail how that happened:
Dutch 18-year-old men gained 20 [IQ] points (1.33 SDs) between 1952 and 1982. By this logic [that is, the logic of strong heritability], a minimum of a 2.67 standard deviation gain in environmental quality would be necessary to account for their IQ gains. If we take into account that the passage of time cannot have had much effect on some significant fraction of environmental causes, the necessary gain for relevant environmental factors might be as much or more than 4 standard deviations. So, assuming a normal distribution for environments, the average Dutch man of 1982 must have had an environment whose quality was well into the highest percentile of the 1952 Dutch distribution.
https://open.substack.com/pub/eclecticinquiries/p/on-race-racism-iq-and-heritability?r=4952v2&utm_campaign=post&utm_medium=web