The heritabilities in Supp Table 5 are SNP-heritabilities, ie they only consider the variation found in SNP array data, not all genetic variation. Everyone in the field knows this, but outsiders might not. It is surely misleading not to mention this.
Just out of curiosity, did you read the post? It is clear that I am talking about SNP heritabilities from the beginning to the end. I explicitly contrast them with twin heritabilities. Let’s go on the record here. I take it your opinion is that the close to zero heritabilities and the absolute zero polygenic score performance will both improve with better SNP technology? On what time frame do you expect that to happen?
Yes, I did read the post. The relevant quote is “The median DGE heritability for behavioral phenotypes is .048. Let that sink in for a second. How different would the modern history of behavior genetics be if back in the 80s one study after another had shown that the heritability of behavior was around .05? When Arthur Jensen wrote about IQ, he usually used a figure of .8 for the heritability of intelligence. I know that the relationship between twin heritabilities and SNP heritabilities is complicated…” That gives the misleading impression that Jensen (or, say, the more commonly used figure of 50% ish heritability for educational attainment) is wildly wrong because of the much lower SNP figure. But this is an apples-to-oranges comparison. People in the field know that, but most people won’t. You only acknowledge it by saying the relationship “is complicated”.
Actually if this argument held good, you wouldn’t even need to mention Tan et al. The population SNP-heritabilities on the x axis of Figure 3 are already mostly below 0.1! So by this logic, GWAS has already doomed behaviour genetics. I don’t find either argument convincing, and I don’t think the field does either - though sure, maybe they have grants to protect.
I don’t have the expertise to predict future performance of PGS from whole-genome data, but I think the solution to missing heritability is to look for it, not give up. But my point is simply that your argument is stated is likely to mislead the general reader.
I don't know. I think I was pretty clear that the difference between SNP and twin heritabilities is a challenge, a mystery. I don't think, and didn't imply, that SNP heritabilities are right, and twin heritabilities are wrong. There is no "right" heritability, it isn't a population parameter we are trying to estimate accurately. As we all know its value depends on where you look and how you measure it.
My point is that you can't have it both ways, taking high heritabilities when they occur as a basis for establishing the importance of genetics, and then just ignoring them when they are small. If the article had had one paragraph saying, uh oh, this suggests that individual-level genomics still isn't very useful in understanding normal range differences in behavior, I would have no problem.
Sure there’s no “right” heritability, but equally one mustn’t ignore the differences between what they measure - which you know, but many do not. Under ideal (theoretical!) conditions, twin studies capture all genetic differences, which array chips can never do. OTOH, both twin and PGS have different biases in real world conditions. (Is that right?) I would say this paper doesn’t mean the original twin studies were misleading. It does suggest that PGS don’t provide much causal “oomph”. Ie it raises more question marks over behavioural/social PGS specifically, than over social genomics as a whole.
And it is a bit of a step back from “the end of social genomics” to “still not very useful”! Based on this paper, I would still be trying to understand the different sources of heritability statistics.
Twin studies are based on false assumptions. Critics have been saying that for 90 years, and molecular genetic failures and low h2 estimates have finally proven that the critics were right all along. The missing heritability problem is, in reality, a "twin study misinterpretation problem."
That’s possible but not proven. In particular, to me the most glaringly false assumption of twin studies - random mating - would lead them to underestimate heritability.
Great points, Eric. I’d point out, though, that some of the public enthusiasm for this was generated by scientists’ overly optimistic claims, which date well before the GWAS era (coupled of course with movie and soap opera twin plots and genetic themes in science fiction). Moreover, taking at face value the paltry .05 heritability you discuss, ignores the possibility that even more confounding will be recognized on the horizon as well as the possibility that some some of this tiny heritability has more to do with physical traits (height, attractiveness, skin color, etc.), which would likely have a small influence on any “behavioral” trait. I don’t know if it is the end of social science genomics, but it should certainly be the end of attributing significant genetic influence to behavioral traits (despite the recent scientist-generated cartoons touting genes for “income”).
I’m not a biologist: could someone clarify the broader implication? Is it that we now believe behavior isn’t genetic, or that the influence of genes on behavior is too complicated to model? If I’m reading it right, Table 5 seems to be saying that height is only 33% heritable - how is that possible? We don’t see adopted children growing to the height of their adoptive families, for example - what am I missing?
Heritability is not the same as something being genetic. For example, having eyes is genetic, but the heritability is low. It’s a different concept. Heritability refers to the proportion of variation in a trait within a population that can be attributed to genetic differences. It is expressed as a coefficient between 0 and 1. So, a heritability estimate of 0.33 means that 33% of the variation in a trait—such as height—across individuals in the population is associated with genetic variation.
So if another paper finds that height is 80% heritable ( https://pmc.ncbi.nlm.nih.gov/articles/PMC10368389/ ), is the discrepancy due to plausibly heritable factors that are not modeled here, or does this analysis rule those out?
Biggest possible "no duh" moment ever. Time for everyone to go back and relearn the Central Limit Theorem: a normal distribution is the result of an effect being caused by many, small, uncorrelated, random influences. Now, if intelligence follows a normal distribution, that must mean it is caused by...think carefully...
There's no doubt that this reported findings have dealt a fatal blow to my conviction that behavioral traits are pre-eminently heritable. But I'd like to think that the right conclusion from these findings isn't that genetics is inconsequential in traits distribution within a population and between families.
This is a remarkable example of an objective statistical fact mercilessly crushing the more subjective experiential sense of "A looks and acts more like B than C because A and B have the same parents." This subjective evidence is almost unshakable and universal in its application as a tried and tested psychosocial heuristic. And yet, here we are.
I posit that for such a stark gap to exist between what we know and observe subjectively versus what science and statistics is reporting, something must be off about the conceptual and empirical paradigm being currently used in the field of behavioral genetics. I want to trust the findings of Tan et al as summarized in this article but I also can't shake the sense that genetics is indeed far more important than this study and others like it suggests such that even if Tan et al and Eric are right that the causal influence of genes are negligible with respect to behavioral traits, it still doesn't provide a superior and "simple" substitute for predicting behavioral outcomes in real life situations.
I find it rare that the people in families don’t have very different personalities and interests. I would argue that the perceived similarities (leaving out physical similarities and similar rearing) are largely projections, perpetuated by these endless scientific claims.
The puzzle isn't that family members differ in personality and interest, rather it is that they look more similar than random others. And mind you, rearing environment is also partly genetic. I think it's wrong to start with the assumption that family members should look alike beyond certain probability cutoff point, say 50%, given the statistical randomness of the law of genetic inheritance. To then dismiss or minimize genetics as the most consequential basis of manifest phenotypes is fundamentally misguided. This is the reason why I argued that we need to reexamine our starting assumptions about genes transmission, combination, and expression when it comes to how they impact behavioral phenotypes.
I said "might". Does it have to be "the above study" when over 50% studies are non-replicating, and are shown under later examination to have crude errors, while most of "peer review" is a joke?
And of course, when the empirical evidence grossly suggests otherwise, the burden on proof is on the other side, and takes extraordinary evidence, not merely a 2024 study.
Given the above study, I do not think empirical evidence grossly suggests otherwise. Especially considering that it uses actual genetic data, and not expectations based on models of segregation.
Previous models based on twin/families w/o genetic variables cannot separate shared environment--which can lead to massive biases. The above paper uses measured genetic data in families to separate those components and demonstrate that heritabilities are very very low for behavioral traits.
With that said, it's clear you're not an expert, and are grasping for things by quoting vague generalities that don't really hold when it comes to modern genetic studies.
Can someone comment on the following implication? Tan et al. shows that even family-based GWAS, which are designed to strip out much of the confounding that plagues population-level studies, still produce unstable and highly variable polygenic scores depending on modeling choices. That strikes me as more than just a technical challenge—it seems to call into question the very stability of how researchers currently estimate genetic influence on traits in general.
If our most “unconfounded” models of direct genetic effect still produce results that vary widely depending on how we parameterize them, then can we meaningfully maintain a distinction between traits that are “more genetic” vs. “more environmental” at all—at least with our current methods and data? Is this a premature conclusion, or a sign that the field is due for a significant conceptual and methodological reset?
They should take the same samples and do a twin EWAS. Determine variance explained with DNA methylation on sites influenced by SNPs and not influenced by SNPs. But the same problem exists in epigenetics as they try to assume all the time that epigenetic variation between twins is measuring environment.
A paper from the decode group a few years back showed that indeed not all twins have identical sequence but indeed show sequence variation to a degree. So there’s also this. Interesting post!
Well, my impression is we can now extend the statement that looking for the Genetics of Schizophrenia as Looking for the Yeti, can be made for all things Mental and Genetics:
--Looking for Genetics of the Mental is like Looking for the Yeti, beyond Schizophrenia, which is the dearly beloved Poster Gal of Psychiatry.
I think I can explain why there will never be Genetical Correlations for Mental things, specially Mental Disorders, unlike for True Brain Diseases where research has shown genetic causes for many, despite those True Diseases, not Imaginary Disorders, are extremely rare, many are called Orphan Diseases precisely because their Prevalence or Incidence, their frequency is anywhere from 1 in 10,000 to 1 in a 100,000, at the maximum. If Genetic Causes where found there with such rare diseases, why not why Mental Disorders that are vastly and prominently more frequent, so they say...
It will be forever looking for another Yeti because the Mind is not Real:
I'm not a biologist or anything—I'm still new to all of this. So what are the actual implications of this study? Is it suggesting that genetics doesn’t really play a role in complex behavior or mental health? I know Eric Turkheimer isn’t an environmentalist, but this study seems to lean in that direction. Or maybe I'm misunderstanding?
The heritabilities in Supp Table 5 are SNP-heritabilities, ie they only consider the variation found in SNP array data, not all genetic variation. Everyone in the field knows this, but outsiders might not. It is surely misleading not to mention this.
Just out of curiosity, did you read the post? It is clear that I am talking about SNP heritabilities from the beginning to the end. I explicitly contrast them with twin heritabilities. Let’s go on the record here. I take it your opinion is that the close to zero heritabilities and the absolute zero polygenic score performance will both improve with better SNP technology? On what time frame do you expect that to happen?
Yes, I did read the post. The relevant quote is “The median DGE heritability for behavioral phenotypes is .048. Let that sink in for a second. How different would the modern history of behavior genetics be if back in the 80s one study after another had shown that the heritability of behavior was around .05? When Arthur Jensen wrote about IQ, he usually used a figure of .8 for the heritability of intelligence. I know that the relationship between twin heritabilities and SNP heritabilities is complicated…” That gives the misleading impression that Jensen (or, say, the more commonly used figure of 50% ish heritability for educational attainment) is wildly wrong because of the much lower SNP figure. But this is an apples-to-oranges comparison. People in the field know that, but most people won’t. You only acknowledge it by saying the relationship “is complicated”.
Actually if this argument held good, you wouldn’t even need to mention Tan et al. The population SNP-heritabilities on the x axis of Figure 3 are already mostly below 0.1! So by this logic, GWAS has already doomed behaviour genetics. I don’t find either argument convincing, and I don’t think the field does either - though sure, maybe they have grants to protect.
I don’t have the expertise to predict future performance of PGS from whole-genome data, but I think the solution to missing heritability is to look for it, not give up. But my point is simply that your argument is stated is likely to mislead the general reader.
I don't know. I think I was pretty clear that the difference between SNP and twin heritabilities is a challenge, a mystery. I don't think, and didn't imply, that SNP heritabilities are right, and twin heritabilities are wrong. There is no "right" heritability, it isn't a population parameter we are trying to estimate accurately. As we all know its value depends on where you look and how you measure it.
My point is that you can't have it both ways, taking high heritabilities when they occur as a basis for establishing the importance of genetics, and then just ignoring them when they are small. If the article had had one paragraph saying, uh oh, this suggests that individual-level genomics still isn't very useful in understanding normal range differences in behavior, I would have no problem.
Sure there’s no “right” heritability, but equally one mustn’t ignore the differences between what they measure - which you know, but many do not. Under ideal (theoretical!) conditions, twin studies capture all genetic differences, which array chips can never do. OTOH, both twin and PGS have different biases in real world conditions. (Is that right?) I would say this paper doesn’t mean the original twin studies were misleading. It does suggest that PGS don’t provide much causal “oomph”. Ie it raises more question marks over behavioural/social PGS specifically, than over social genomics as a whole.
And it is a bit of a step back from “the end of social genomics” to “still not very useful”! Based on this paper, I would still be trying to understand the different sources of heritability statistics.
Twin studies are based on false assumptions. Critics have been saying that for 90 years, and molecular genetic failures and low h2 estimates have finally proven that the critics were right all along. The missing heritability problem is, in reality, a "twin study misinterpretation problem."
That’s possible but not proven. In particular, to me the most glaringly false assumption of twin studies - random mating - would lead them to underestimate heritability.
Rare variants will not save PGI, h2, etc. This has been demonstrated over and over again.
He’s also not considering the Mars effect.
Great points, Eric. I’d point out, though, that some of the public enthusiasm for this was generated by scientists’ overly optimistic claims, which date well before the GWAS era (coupled of course with movie and soap opera twin plots and genetic themes in science fiction). Moreover, taking at face value the paltry .05 heritability you discuss, ignores the possibility that even more confounding will be recognized on the horizon as well as the possibility that some some of this tiny heritability has more to do with physical traits (height, attractiveness, skin color, etc.), which would likely have a small influence on any “behavioral” trait. I don’t know if it is the end of social science genomics, but it should certainly be the end of attributing significant genetic influence to behavioral traits (despite the recent scientist-generated cartoons touting genes for “income”).
I’m not a biologist: could someone clarify the broader implication? Is it that we now believe behavior isn’t genetic, or that the influence of genes on behavior is too complicated to model? If I’m reading it right, Table 5 seems to be saying that height is only 33% heritable - how is that possible? We don’t see adopted children growing to the height of their adoptive families, for example - what am I missing?
Heritability is not the same as something being genetic. For example, having eyes is genetic, but the heritability is low. It’s a different concept. Heritability refers to the proportion of variation in a trait within a population that can be attributed to genetic differences. It is expressed as a coefficient between 0 and 1. So, a heritability estimate of 0.33 means that 33% of the variation in a trait—such as height—across individuals in the population is associated with genetic variation.
So if another paper finds that height is 80% heritable ( https://pmc.ncbi.nlm.nih.gov/articles/PMC10368389/ ), is the discrepancy due to plausibly heritable factors that are not modeled here, or does this analysis rule those out?
Biggest possible "no duh" moment ever. Time for everyone to go back and relearn the Central Limit Theorem: a normal distribution is the result of an effect being caused by many, small, uncorrelated, random influences. Now, if intelligence follows a normal distribution, that must mean it is caused by...think carefully...
There's no doubt that this reported findings have dealt a fatal blow to my conviction that behavioral traits are pre-eminently heritable. But I'd like to think that the right conclusion from these findings isn't that genetics is inconsequential in traits distribution within a population and between families.
This is a remarkable example of an objective statistical fact mercilessly crushing the more subjective experiential sense of "A looks and acts more like B than C because A and B have the same parents." This subjective evidence is almost unshakable and universal in its application as a tried and tested psychosocial heuristic. And yet, here we are.
I posit that for such a stark gap to exist between what we know and observe subjectively versus what science and statistics is reporting, something must be off about the conceptual and empirical paradigm being currently used in the field of behavioral genetics. I want to trust the findings of Tan et al as summarized in this article but I also can't shake the sense that genetics is indeed far more important than this study and others like it suggests such that even if Tan et al and Eric are right that the causal influence of genes are negligible with respect to behavioral traits, it still doesn't provide a superior and "simple" substitute for predicting behavioral outcomes in real life situations.
I find it rare that the people in families don’t have very different personalities and interests. I would argue that the perceived similarities (leaving out physical similarities and similar rearing) are largely projections, perpetuated by these endless scientific claims.
The puzzle isn't that family members differ in personality and interest, rather it is that they look more similar than random others. And mind you, rearing environment is also partly genetic. I think it's wrong to start with the assumption that family members should look alike beyond certain probability cutoff point, say 50%, given the statistical randomness of the law of genetic inheritance. To then dismiss or minimize genetics as the most consequential basis of manifest phenotypes is fundamentally misguided. This is the reason why I argued that we need to reexamine our starting assumptions about genes transmission, combination, and expression when it comes to how they impact behavioral phenotypes.
"Would you believe your own lying eyes over a study that might have tons of methodolical errors, p-hacking, not replicate, etc?"
Can you expand on how the above study employs p-hacking? Or what "methodolical [sic] errors" exist?
I said "might". Does it have to be "the above study" when over 50% studies are non-replicating, and are shown under later examination to have crude errors, while most of "peer review" is a joke?
And of course, when the empirical evidence grossly suggests otherwise, the burden on proof is on the other side, and takes extraordinary evidence, not merely a 2024 study.
Given the above study, I do not think empirical evidence grossly suggests otherwise. Especially considering that it uses actual genetic data, and not expectations based on models of segregation.
Previous models based on twin/families w/o genetic variables cannot separate shared environment--which can lead to massive biases. The above paper uses measured genetic data in families to separate those components and demonstrate that heritabilities are very very low for behavioral traits.
With that said, it's clear you're not an expert, and are grasping for things by quoting vague generalities that don't really hold when it comes to modern genetic studies.
That's like being between a rock and a hard place.
Can someone comment on the following implication? Tan et al. shows that even family-based GWAS, which are designed to strip out much of the confounding that plagues population-level studies, still produce unstable and highly variable polygenic scores depending on modeling choices. That strikes me as more than just a technical challenge—it seems to call into question the very stability of how researchers currently estimate genetic influence on traits in general.
If our most “unconfounded” models of direct genetic effect still produce results that vary widely depending on how we parameterize them, then can we meaningfully maintain a distinction between traits that are “more genetic” vs. “more environmental” at all—at least with our current methods and data? Is this a premature conclusion, or a sign that the field is due for a significant conceptual and methodological reset?
The suggestion here is that the field is possibly dead.
Guess I won’t worry about 23andMe selling my genetic data anymore;)
I suppose there are other reasons to worry, but what matters is not real risk but whatever actuaries say is a risk.
They should take the same samples and do a twin EWAS. Determine variance explained with DNA methylation on sites influenced by SNPs and not influenced by SNPs. But the same problem exists in epigenetics as they try to assume all the time that epigenetic variation between twins is measuring environment.
A paper from the decode group a few years back showed that indeed not all twins have identical sequence but indeed show sequence variation to a degree. So there’s also this. Interesting post!
Well, my impression is we can now extend the statement that looking for the Genetics of Schizophrenia as Looking for the Yeti, can be made for all things Mental and Genetics:
--Looking for Genetics of the Mental is like Looking for the Yeti, beyond Schizophrenia, which is the dearly beloved Poster Gal of Psychiatry.
I think I can explain why there will never be Genetical Correlations for Mental things, specially Mental Disorders, unlike for True Brain Diseases where research has shown genetic causes for many, despite those True Diseases, not Imaginary Disorders, are extremely rare, many are called Orphan Diseases precisely because their Prevalence or Incidence, their frequency is anywhere from 1 in 10,000 to 1 in a 100,000, at the maximum. If Genetic Causes where found there with such rare diseases, why not why Mental Disorders that are vastly and prominently more frequent, so they say...
It will be forever looking for another Yeti because the Mind is not Real:
https://federicosotodelalba.substack.com/p/beauty?r=4up0lp
I'm not a biologist or anything—I'm still new to all of this. So what are the actual implications of this study? Is it suggesting that genetics doesn’t really play a role in complex behavior or mental health? I know Eric Turkheimer isn’t an environmentalist, but this study seems to lean in that direction. Or maybe I'm misunderstanding?
Heritability of depression is actually .05 and the remainder is attributable to environment or gene-environment correlations?
Does anyone actually believe that could possibly be true?
I have not delved into the details but this strikes me at first glance as likely overcontrol in the analysis.
Geneticist and Sociologist Dalton Conley just released a book about Social Genomics and it's possible future.
Is it just time to go back to the drawing board!?
How about move on from it?