As Borsboom notes, intelligence and height are pretty different. Intelligence is measured and defined as a between subjects latent variable. Height, on the other hand, is not defined as in terms of between subject differences, but in terms of an empirical concatenation operation.
They are pretty clearly different! It is crazy to see many prominent intelligence researchers think this is even remotely controversial. A lot of them have convinced themselves that any sort of criticism of intelligence research is ideologically motivated or something, which is pretty unfortunate.
I am really glad that you are popularizing the criticisms of hereditarian/race science research that have always been met with vitriol and negativity by those specific communities so it has been hard to actually openly argue about them. You are really doing a great public service.
Thanks! Yes, what has surprised me is how much internal conceptual debate there has already been between the various IQ schools and the psyshometricians, which is then completely ignored in the public discussion.
And yeah I didn't get a chance to work in my favorite point from Borsboom et al. 2003 (https://pubmed.ncbi.nlm.nih.gov/12747522/) which, as you note, draws a really important distinction between the measurement of height and the latent variable analysis of IQ:
"But what about variables like height? Is it not unreasonable to say, “If Einstein had been taller, he would have been able to reach the upper shelves in the library”? No, this is not unreasonable, but it is unreasonable to assume a priori that intelligence, as a between- subjects latent variable, applies in the same way as height does. The concept of height is not defined in terms of between-subjects differences, but in terms of an empirical concatenation operation (Krantz, Luce, Suppes, & Tversky, 1971; Michell, 1999). Roughly, this means that we know how to move Einstein around in the height dimension (for example by giving him platform shoes) and that the effect of doing this is tractable (namely, wearing platform shoes will enable Einstein to reach the upper shelves). Moreover, it can be assumed that the height dimension applies to within- subject differences in the same way that it applies to between- subjects differences. This is to say that the statements, “If Einstein had been taller, he would have been able to reach the upper shelves in the library” and “If we had replaced Einstein with a taller person, this person would have been able to reach the upper shelves in the library” are equivalent with respect to the dimension under consideration. They are equivalent in this sense, exactly because the dimensions pertaining to within- and between-subjects variability are qualitatively the same: If we give Einstein platform shoes that make him taller, he is, in all relevant respects, exchange-able with the taller person in the example. We do not object to introducing height in a causal account of this kind, because variations in height have demonstrably the same effect within and between subjects. But it remains to be shown that the same holds true for psychological variables like intelligence."
Yes! There has been significant conceptual debate within psychometrics. There is a really good presentation by Borsboom on the philosophy of psychometrics (https://vimeo.com/256145513) that touches on those debates and provides a really good introduction of the field. I really recommend it to anyone that is interested in psychological measurement or even measurement theory in general.
I think part of the reason those things are ignored in the public discussions is because they are just too dry and esoteric, and they are just too negative. People tend to want to be positive about their research when they communicate it to the public or even think about it at all. As Borsboom jokingly says in the presentation above, if you raise those questions in a conference, you won't be invited to any party. There is also the fact that there is a pretty significant disconnect between psychologists that study those areas and the psychometricians. I think the field has been set back quite a bit because of that. There is an interesting dissertation to be written about this for a sociologist/philosopher of science and try to trace it back to the controversies of behavioral genetics and intelligence research and the legacy of scientific racism and/or the sociobiology wars.
In general though my impression is that the field has started, slowly but surely, moving away from the dominance of latent variable models and stuff like network models are growing in popularity (doi.org/10.1146/annurev-clinpsy-050212-185608). As a biologist, I find network models way more realistic and reasonable than latent variable models for sure!
Great article! Hope you [or some affiliate that has the time] expands this into a book with extensive lay person support. It's nightmare arguing about this stuff at the barbeque !
"We know this because we have estimated a parameter called molecular heritability, which tells us the upper bound on what a genetic predictor could ever achieve."
Could you talk more about how we know that molecular heritability represents the upper bound of heritability? My impression was that not all SNPs have been discovered, and these have polygenic effects that are hard to define. So how can you know that the maximum possible genetic effect on IQ and other traits has been discovered?
TLDR: Molecular heritability is an upper bound on what you can predict *from* all the molecular data you have typed (regardless of whether you identified the specific mechanisms). But in general the typed molecular genetic variation appears to explain most (>80%) of the total genetic variation.
I dont understand how IQ is this nebulous? Environmental factors can unilaterally wash away any genetic link? Then despite environemntal factors the Chinese and Europeans (living all over the world) have a similar IQ as they do to their people living in their native homelands. Wouldnt that alone prove genetics? Who is convinced that parents with advanced degrees and difficult, highly technical jobs wont on average have more intelligent children than someone with low skilled felon parents? Education and lead free painted homes dont explain this
This is an interesting and important finding, but boy is it hard to follow along with exactly what all these different statistical measures actually mean.
I mean, I still believe that conclusion, but I have to concede its more because of the authoritativeness of the source than my being able to reconstruct the argument myself. And I'm pretty smart. How would one communicate these findings to people of average smartness and more suspicious of scientists as a source of unbiased information?
That's a good question, what do you think would help? So far I've tried to (a) simplify the intuition as much as possible; (b) explicitly quote from other papers in areas that might be seen as controversial. Would it be useful to try to get rid of any jargon entirely? Or provide a shorter overall summary? Something else? There are only ~3 core concepts involved here (population heritability, direct heritability, genetic correlation, factor analysis) but each one is unfortunately hard to be precise about without getting into jargon.
Great article. Let's try to make things more intuitive. Height is the combined length of many bones, some people have long legs, other long torsos, some both. Yet it can be measured very simply. But the underlying structures are complex. My 185cm height number does not say whether my shin are short or long. In some sense, "height" is not a thing, not an irreducible essence at least, it is an aggregate of many things. It is the combined outcome of shin growth, spine growth etc. etc. when we think of a measurement as a thing, it is called reification.
IQ measurement is very complex. The underlying structures hence must be really complex and thus intelligence is even less of an irreducible essential thing as height is. It is a measurement of the combined effect of many things. Many abilities.
The reason IQ is useful is that it correlates with success with many tasks in modern life. g-loaded tasks. Not all tasks are g-loaded (e.g. running is not), but many are, those that contribute to life outcomes the most (except for those people whose life outcome is e.g. Olympic gold in running). IQ is basically a measurement of adaptation to modern life. Balancing checkbooks etc. it is not a single unified essential irreducible "intelligence thing", but an aggregate measurement.
In 1930 a European hunter went hunting in what would later become Kenya. And he was amazed. He spent his entire life with European hunters and no one of the could read tracks so well as the native African villagers. They would look at a track and say "young pregnant female elephant, no, not the same one as yesterday, a different one". OTOH some of them spent years serving in the British military and could not speak a word of English. Were they high IQ or not? I think this is not a sensible question. It depends whether you put elephant tracks on the test or English words. Why were they very good at learning reading tracks and not good at learning English? No one really knows.
No. Why? Have you seen a Raven's Progressive Matrices? I think that is where the shape rotator meme comes from. It is an engineering skill test for people who never studied engineering.
The heritability estimates indicate that population heritability for height is 37%, while IQ is at 23% and educational attainment is only 12%. However, educational attainment shows a high genetic correlation with participation (r² = 0.72), and IQ is not far behind (r² = 0.39). This suggests that the observed heritability for IQ might be lower due to participant bias. If we adjust the heritability of IQ to account for this genetic correlation, it could potentially increase to about 35.4%. What do you think?
In fact, the selection bias in GWAS is also important for all characteristics, as mental illness, for example, also correlates with IQ. Speaking of IQ: it is more or less the best measure we have in psychology ...
This is good intuition, but Schoeler et al. (https://www.nature.com/articles/s41562-023-01579-9) included a comparison of the raw population heritabilities with heritabilities estimated after re-weighting for participation and the results were only slightly different: "While heritability estimates were less impacted by weighting (maximum change in h2, 5%) ..."
Fascinating, but I wouldn't dismiss it just because it's low.
What do you think of this?
GWAS and twin studies are both quite robust, so rather than asking which is right, we should focus on understanding the reasons for the gap and trying to bridge it. It is plausible that both twin studies and GWAS are accurate to some extent. Twin studies focus on overall genetic influence, and GWAS focus on specific common variants. Twin studies may do a better job of capturing this parent-offspring heritability, while GWAS may struggle because they look at a more genetically and environmentally diverse population. If the GWAS estimated genetic correlation of a trait is substantially lower than the twin studies estimated heritability of the trait, it may indicate that the GWAS is missing important genetic components. Why not argue that the GWAS estimate is too low?
Another take: The assumption that twin studies and GWAS measure "different aspects of the same underlying phenomenon" may be too simplistic. In reality, they may measure fundamentally different things, with GWAS focusing on additive genetic variance and twin studies capturing a broader range of genetic influences, including dominance and epistasis, which could make direct comparisons problematic.
Alternative Hypothesis: Since GWAS do not sequence the entire genome but focus primarily on SNPs, they may underestimate the true genetic heritability of IQ compared to full genome sequencing. I find it hard to believe that the genetic heritability for educational attainment is genuinely so low. It’s not uncommon to see a “not-so-smart” child in a gifted family, which suggests that IQ and educational attainment may be influenced more by genetic factors than environmental ones. However, there seems to be greater variance between parents and children than we might expect based on traits like height, possibly because intelligence is inherently more complex. What are your thoughts?
It's certainly possible that rare variants will explain some of the additional variance, though the initial studies of genes/exomes (which is where we would expect most of this extra variance to lie) are showing very little (summarized here: http://gusevlab.org/projects/hsq/#h.375v1ho0nj4).
Regarding seeing patterns in families, I would caution against that as heritability behaves very counterintuitively. For example, for a trait with 10% prevalence in the population (say, something like being noticeably "smart") even with a high heritability of 0.8 and *both* parents exhibiting the trait, the probability that the offspring exhibits it is just 55% (for a heritability of 0.2, this is 18%). Which is to say that even for highly heritable traits you would expect to see *many* families where both parents are in the top 10% and the child is not. Can you distinguish between 55% of families and 18% of families just from anecdotal evidence? I don't think I could.
This may be a juvenile question, but aren't the majority of "IQ" tests in major biobanks pretty barebones? I believe IQ tests generally correlate with each other around .8. I'm not sure how well the biobank questionnaires would correlate with a full-fledged IQ test.. but if they correlate less with a standard IQ test, say .4, would this mean that the estimated direct heritability would be less than if all the GWAS participants were given standard tests? Thanks- great article!
This is a good question, the primary IQ test in the UK Biobank is a fairly bare bones ~30 minute questionnaire. However, the test-retest reliability is actually pretty good; this paper (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0231627) quantified at at 0.82 for the general factor, which is only slightly lower than what you see with fancy 90 minute tests. Separately, this paper (https://pubmed.ncbi.nlm.nih.gov/36378351/) used a larger test battery across a subset of individuals to derive a more accurate general factor; they put in a ton of work and the resulting measure had ... basically the same heritability as the 30 minute test and a genetic correlation of ~1.0 with it. It just doesn't seem like IQ test complexity is all that related to the genetic component.
Intelligence is a much more complex phenomenon than height and so its genetics is much more complicated than the genetics of height. The most likely explanation of the 'missing heritability problem' is that as more information comes in the gaps will become smaller and converge with population heritability. This is the latest iteration of 'egalitarianism of the gaps' argument and the safe bet is that it will end the same way the others ended.
The trend so far has been that as more information comes in, we learn that the genetic studies have confounding we didn't initially know about, and the gap with twin studies grows. I think the most likely explanation of the missing heritability problem is that twin studies are inflated by environmental differences, as has been known since the 70's (https://theinfinitesimal.substack.com/p/twin-heritability-models-can-tell) and also shown with modern quantitative methods (https://pubmed.ncbi.nlm.nih.gov/30104764/).
But at the same time, more information has come in supporting the hereditarian hypothesis. For example, it is now possible to see if higher white ancestry correlates with higher IQ among African Americans and it does. Conversely, there is no additional evidence in favour of any of the proposed environmental mechanisms for racial differences in IQ. The missing heritability problem is interesting, but the smart move is to wait and see before overturning decades of cumulative research all pointing in one direction.
Admixture analysis is just another way of measuring environmental correlations, it neither supports nor refutes either hypothesis (see examples here: https://x.com/SashaGusevPosts/status/1713626146963599712). But I can think of quite a few genetic predictions from hereditarians that have not held up:
1. That IQ is just genetics and non-shared environment with negligible cultural influence: indirect effects in genomic studies clearly show there are vertical cultural influences.
2. That outcomes like educational attainment are highly heritable and will be easy to build genomic predictors for: one of the largest genetic analyses of any trait is of educational attainment (n=3M) and showed <5% of the effects are actually causal families. In general IQ/edu is one of the least heritable traits of all tested.
3. That genetic effects on common traits would differ substantially between populations: no, causal effects are nearly identical between populations (https://pubmed.ncbi.nlm.nih.gov/36941441/).
4. That cognitive function is under strong and recent selection: No, IQ exhibits one of the weakest signals of selection if not entirely neutral.
5. Recent polygenic selection is widespread and acting on many traits (e.g. Steve Hsu predicting this: https://infoproc.blogspot.com/2012/08/recent-human-evolution-european-height.html): No, evidence of polygenic selection turned out to be a false positive and if it exists at all it is so weak as to be undistinguishable from population structure.
6. There will be large rare variant effects on IQ and rare variants explain the bulk of heritability. Nope, recent exome-wide studies show tiny contributions of rare variants (https://www.nature.com/articles/s41588-023-01398-8).
The missing heritability "problem" is just a specific example of a broader problem where hereditarians overfit to observational data and now that none of their forward-looking predictions about genetics have held up, they are inventing artifacts (or, as you are doing, simply continue to claim that it "all points in one direction").
Admixture analysis was originally suggested by environmentalists because they thought it would disprove the hereditarian thesis. Then when it didn't, they changed their mind and decided that it didn't mean anything. This is a consistent pattern of environmentalists for 50 years. There remains no hypothesis that can explain, at all, racial gaps in IQ by environmental means, whereas the genetic hypothesis is parsimonious and fits the data. All attempts to increase intelligence of low-IQ racial groups, up to and including relocating children to new families have failed miserably. Maybe environmentalists struck a pot of gold this time with the missing heritability problem, but, more likely, it's just a complex puzzle that will take many decades to solve without changing the basic picture.
It turns out that the scientific method is a test of specific hypotheses, not a test of the people that proposed them. Where scientists propose parameters and then develop methods to estimate them in an unbiased manner, "race scientists" simply walk through the garden of forking paths until they find a correlation that "looks right". Regression to the mean in siblings, admixture analysis, the method of correlated vectors / Jensen effect (essentially all of The G Factor), the national IQ nonsense, are all examples of this. When it is pointed out that their analysis in fact cannot distinguish between multiple hypotheses (often trivially so), they do not attempt to improve their methods -- as an actual scientist would -- but either fall back on an appeal to some environmentalist authority or argue that they're just providing one piece of evidence in a broader narrative, as if a pile of garbage can turn into gold if you simply pile it high enough.
You will note that the 6 examples I mentioned above were not "X hereditarian said analysis Y would prove hereditarianism" but "X parameter turned out to be Z instead of Y which is not consistent with a hereditarian model". Also none of the points were about "the missing heritability problem", which seems to have become a kind of verbal tic to avoid actually having to consider new evidence. Highly correlated effect sizes across populations have nothing to do with missing heritability. Estimates of indirect/cultural transmission have nothing to do with missing heritability. You get the idea. The thing to do here is to acknowledge that your statement "decades of cumulative research all pointing in one direction" was incorrect, and grossly so.
Re: 1, what's the best way to disentangle VCT from other types of rGE/indirect effects? Do we have any priors on how prevalent one should be over the other?
It gets estimates that are ballpark consistent with (non-genetic) multigenerational studies. However, it is almost certainly the case that assortative mating is NOT just happening on the phenotype, so this is a very simple model. I think it would be helpful if behavioral geneticists proposed more explicit models of rGE besides VCT and how to test them.
"Nothing to see here. IQ heritability upper bound stems not from GWAS studies, but from twin studies. I have no idea why he would think otherwise and that makes me question his motives."
Really glad to see you cite Borsboom.
As Borsboom notes, intelligence and height are pretty different. Intelligence is measured and defined as a between subjects latent variable. Height, on the other hand, is not defined as in terms of between subject differences, but in terms of an empirical concatenation operation.
They are pretty clearly different! It is crazy to see many prominent intelligence researchers think this is even remotely controversial. A lot of them have convinced themselves that any sort of criticism of intelligence research is ideologically motivated or something, which is pretty unfortunate.
I am really glad that you are popularizing the criticisms of hereditarian/race science research that have always been met with vitriol and negativity by those specific communities so it has been hard to actually openly argue about them. You are really doing a great public service.
Thanks! Yes, what has surprised me is how much internal conceptual debate there has already been between the various IQ schools and the psyshometricians, which is then completely ignored in the public discussion.
And yeah I didn't get a chance to work in my favorite point from Borsboom et al. 2003 (https://pubmed.ncbi.nlm.nih.gov/12747522/) which, as you note, draws a really important distinction between the measurement of height and the latent variable analysis of IQ:
"But what about variables like height? Is it not unreasonable to say, “If Einstein had been taller, he would have been able to reach the upper shelves in the library”? No, this is not unreasonable, but it is unreasonable to assume a priori that intelligence, as a between- subjects latent variable, applies in the same way as height does. The concept of height is not defined in terms of between-subjects differences, but in terms of an empirical concatenation operation (Krantz, Luce, Suppes, & Tversky, 1971; Michell, 1999). Roughly, this means that we know how to move Einstein around in the height dimension (for example by giving him platform shoes) and that the effect of doing this is tractable (namely, wearing platform shoes will enable Einstein to reach the upper shelves). Moreover, it can be assumed that the height dimension applies to within- subject differences in the same way that it applies to between- subjects differences. This is to say that the statements, “If Einstein had been taller, he would have been able to reach the upper shelves in the library” and “If we had replaced Einstein with a taller person, this person would have been able to reach the upper shelves in the library” are equivalent with respect to the dimension under consideration. They are equivalent in this sense, exactly because the dimensions pertaining to within- and between-subjects variability are qualitatively the same: If we give Einstein platform shoes that make him taller, he is, in all relevant respects, exchange-able with the taller person in the example. We do not object to introducing height in a causal account of this kind, because variations in height have demonstrably the same effect within and between subjects. But it remains to be shown that the same holds true for psychological variables like intelligence."
Yes! There has been significant conceptual debate within psychometrics. There is a really good presentation by Borsboom on the philosophy of psychometrics (https://vimeo.com/256145513) that touches on those debates and provides a really good introduction of the field. I really recommend it to anyone that is interested in psychological measurement or even measurement theory in general.
I think part of the reason those things are ignored in the public discussions is because they are just too dry and esoteric, and they are just too negative. People tend to want to be positive about their research when they communicate it to the public or even think about it at all. As Borsboom jokingly says in the presentation above, if you raise those questions in a conference, you won't be invited to any party. There is also the fact that there is a pretty significant disconnect between psychologists that study those areas and the psychometricians. I think the field has been set back quite a bit because of that. There is an interesting dissertation to be written about this for a sociologist/philosopher of science and try to trace it back to the controversies of behavioral genetics and intelligence research and the legacy of scientific racism and/or the sociobiology wars.
In general though my impression is that the field has started, slowly but surely, moving away from the dominance of latent variable models and stuff like network models are growing in popularity (doi.org/10.1146/annurev-clinpsy-050212-185608). As a biologist, I find network models way more realistic and reasonable than latent variable models for sure!
It's not nature vs nurture. You nurture nature.
This is fantastic once again Sasha. This is a minor quibble but I would refer to the race "scientists" as race "scientists".
Thank you! Good point
Why such disrespect?
Nicely done.
I have added your article to my burgeoning education graph:
https://embed.kumu.io/c3b449cc0b3a9558b8e2b1e8f7f4bf4b
Thanks for the like. If you like short stories, I post 2 or 3 a month. The next one is out on Saturday. Take care.
Great article! Hope you [or some affiliate that has the time] expands this into a book with extensive lay person support. It's nightmare arguing about this stuff at the barbeque !
interesting read, insightful too
"We know this because we have estimated a parameter called molecular heritability, which tells us the upper bound on what a genetic predictor could ever achieve."
Could you talk more about how we know that molecular heritability represents the upper bound of heritability? My impression was that not all SNPs have been discovered, and these have polygenic effects that are hard to define. So how can you know that the maximum possible genetic effect on IQ and other traits has been discovered?
Hi Ian, it took me a bit but I expanded on the differences between molecular and family-based heritability estimates here:
https://theinfinitesimal.substack.com/i/148251755/okay-but-why-do-you-think-twin-estimates-so-much-higher
TLDR: Molecular heritability is an upper bound on what you can predict *from* all the molecular data you have typed (regardless of whether you identified the specific mechanisms). But in general the typed molecular genetic variation appears to explain most (>80%) of the total genetic variation.
Thanks I'll read it with interest!
I dont understand how IQ is this nebulous? Environmental factors can unilaterally wash away any genetic link? Then despite environemntal factors the Chinese and Europeans (living all over the world) have a similar IQ as they do to their people living in their native homelands. Wouldnt that alone prove genetics? Who is convinced that parents with advanced degrees and difficult, highly technical jobs wont on average have more intelligent children than someone with low skilled felon parents? Education and lead free painted homes dont explain this
This is an interesting and important finding, but boy is it hard to follow along with exactly what all these different statistical measures actually mean.
I mean, I still believe that conclusion, but I have to concede its more because of the authoritativeness of the source than my being able to reconstruct the argument myself. And I'm pretty smart. How would one communicate these findings to people of average smartness and more suspicious of scientists as a source of unbiased information?
That's a good question, what do you think would help? So far I've tried to (a) simplify the intuition as much as possible; (b) explicitly quote from other papers in areas that might be seen as controversial. Would it be useful to try to get rid of any jargon entirely? Or provide a shorter overall summary? Something else? There are only ~3 core concepts involved here (population heritability, direct heritability, genetic correlation, factor analysis) but each one is unfortunately hard to be precise about without getting into jargon.
It might just require me to reflect on the core concepts until I understand them fully and then reread the piece.
I sympathise with science communicators like yourself. Making complex ideas accessible to laymen is really hard!
Great article. Let's try to make things more intuitive. Height is the combined length of many bones, some people have long legs, other long torsos, some both. Yet it can be measured very simply. But the underlying structures are complex. My 185cm height number does not say whether my shin are short or long. In some sense, "height" is not a thing, not an irreducible essence at least, it is an aggregate of many things. It is the combined outcome of shin growth, spine growth etc. etc. when we think of a measurement as a thing, it is called reification.
IQ measurement is very complex. The underlying structures hence must be really complex and thus intelligence is even less of an irreducible essential thing as height is. It is a measurement of the combined effect of many things. Many abilities.
The reason IQ is useful is that it correlates with success with many tasks in modern life. g-loaded tasks. Not all tasks are g-loaded (e.g. running is not), but many are, those that contribute to life outcomes the most (except for those people whose life outcome is e.g. Olympic gold in running). IQ is basically a measurement of adaptation to modern life. Balancing checkbooks etc. it is not a single unified essential irreducible "intelligence thing", but an aggregate measurement.
In 1930 a European hunter went hunting in what would later become Kenya. And he was amazed. He spent his entire life with European hunters and no one of the could read tracks so well as the native African villagers. They would look at a track and say "young pregnant female elephant, no, not the same one as yesterday, a different one". OTOH some of them spent years serving in the British military and could not speak a word of English. Were they high IQ or not? I think this is not a sensible question. It depends whether you put elephant tracks on the test or English words. Why were they very good at learning reading tracks and not good at learning English? No one really knows.
"The reason IQ is useful is that it correlates with success with many tasks in modern life."
Maybe. But it may correlation more strongly with brainwashing (imperial conditioning).
No. Why? Have you seen a Raven's Progressive Matrices? I think that is where the shape rotator meme comes from. It is an engineering skill test for people who never studied engineering.
Exactly with the RPM you dont even need to know how to read
The heritability estimates indicate that population heritability for height is 37%, while IQ is at 23% and educational attainment is only 12%. However, educational attainment shows a high genetic correlation with participation (r² = 0.72), and IQ is not far behind (r² = 0.39). This suggests that the observed heritability for IQ might be lower due to participant bias. If we adjust the heritability of IQ to account for this genetic correlation, it could potentially increase to about 35.4%. What do you think?
In fact, the selection bias in GWAS is also important for all characteristics, as mental illness, for example, also correlates with IQ. Speaking of IQ: it is more or less the best measure we have in psychology ...
This is good intuition, but Schoeler et al. (https://www.nature.com/articles/s41562-023-01579-9) included a comparison of the raw population heritabilities with heritabilities estimated after re-weighting for participation and the results were only slightly different: "While heritability estimates were less impacted by weighting (maximum change in h2, 5%) ..."
Fascinating, but I wouldn't dismiss it just because it's low.
What do you think of this?
GWAS and twin studies are both quite robust, so rather than asking which is right, we should focus on understanding the reasons for the gap and trying to bridge it. It is plausible that both twin studies and GWAS are accurate to some extent. Twin studies focus on overall genetic influence, and GWAS focus on specific common variants. Twin studies may do a better job of capturing this parent-offspring heritability, while GWAS may struggle because they look at a more genetically and environmentally diverse population. If the GWAS estimated genetic correlation of a trait is substantially lower than the twin studies estimated heritability of the trait, it may indicate that the GWAS is missing important genetic components. Why not argue that the GWAS estimate is too low?
Another take: The assumption that twin studies and GWAS measure "different aspects of the same underlying phenomenon" may be too simplistic. In reality, they may measure fundamentally different things, with GWAS focusing on additive genetic variance and twin studies capturing a broader range of genetic influences, including dominance and epistasis, which could make direct comparisons problematic.
Still, your argument is pretty compelling!
Alternative Hypothesis: Since GWAS do not sequence the entire genome but focus primarily on SNPs, they may underestimate the true genetic heritability of IQ compared to full genome sequencing. I find it hard to believe that the genetic heritability for educational attainment is genuinely so low. It’s not uncommon to see a “not-so-smart” child in a gifted family, which suggests that IQ and educational attainment may be influenced more by genetic factors than environmental ones. However, there seems to be greater variance between parents and children than we might expect based on traits like height, possibly because intelligence is inherently more complex. What are your thoughts?
It's certainly possible that rare variants will explain some of the additional variance, though the initial studies of genes/exomes (which is where we would expect most of this extra variance to lie) are showing very little (summarized here: http://gusevlab.org/projects/hsq/#h.375v1ho0nj4).
Regarding seeing patterns in families, I would caution against that as heritability behaves very counterintuitively. For example, for a trait with 10% prevalence in the population (say, something like being noticeably "smart") even with a high heritability of 0.8 and *both* parents exhibiting the trait, the probability that the offspring exhibits it is just 55% (for a heritability of 0.2, this is 18%). Which is to say that even for highly heritable traits you would expect to see *many* families where both parents are in the top 10% and the child is not. Can you distinguish between 55% of families and 18% of families just from anecdotal evidence? I don't think I could.
Thank you for taking the time to respond to all my comments. I really appreciate the debate!
This may be a juvenile question, but aren't the majority of "IQ" tests in major biobanks pretty barebones? I believe IQ tests generally correlate with each other around .8. I'm not sure how well the biobank questionnaires would correlate with a full-fledged IQ test.. but if they correlate less with a standard IQ test, say .4, would this mean that the estimated direct heritability would be less than if all the GWAS participants were given standard tests? Thanks- great article!
This is a good question, the primary IQ test in the UK Biobank is a fairly bare bones ~30 minute questionnaire. However, the test-retest reliability is actually pretty good; this paper (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0231627) quantified at at 0.82 for the general factor, which is only slightly lower than what you see with fancy 90 minute tests. Separately, this paper (https://pubmed.ncbi.nlm.nih.gov/36378351/) used a larger test battery across a subset of individuals to derive a more accurate general factor; they put in a ton of work and the resulting measure had ... basically the same heritability as the 30 minute test and a genetic correlation of ~1.0 with it. It just doesn't seem like IQ test complexity is all that related to the genetic component.
Intelligence is a much more complex phenomenon than height and so its genetics is much more complicated than the genetics of height. The most likely explanation of the 'missing heritability problem' is that as more information comes in the gaps will become smaller and converge with population heritability. This is the latest iteration of 'egalitarianism of the gaps' argument and the safe bet is that it will end the same way the others ended.
The trend so far has been that as more information comes in, we learn that the genetic studies have confounding we didn't initially know about, and the gap with twin studies grows. I think the most likely explanation of the missing heritability problem is that twin studies are inflated by environmental differences, as has been known since the 70's (https://theinfinitesimal.substack.com/p/twin-heritability-models-can-tell) and also shown with modern quantitative methods (https://pubmed.ncbi.nlm.nih.gov/30104764/).
But at the same time, more information has come in supporting the hereditarian hypothesis. For example, it is now possible to see if higher white ancestry correlates with higher IQ among African Americans and it does. Conversely, there is no additional evidence in favour of any of the proposed environmental mechanisms for racial differences in IQ. The missing heritability problem is interesting, but the smart move is to wait and see before overturning decades of cumulative research all pointing in one direction.
Admixture analysis is just another way of measuring environmental correlations, it neither supports nor refutes either hypothesis (see examples here: https://x.com/SashaGusevPosts/status/1713626146963599712). But I can think of quite a few genetic predictions from hereditarians that have not held up:
1. That IQ is just genetics and non-shared environment with negligible cultural influence: indirect effects in genomic studies clearly show there are vertical cultural influences.
2. That outcomes like educational attainment are highly heritable and will be easy to build genomic predictors for: one of the largest genetic analyses of any trait is of educational attainment (n=3M) and showed <5% of the effects are actually causal families. In general IQ/edu is one of the least heritable traits of all tested.
3. That genetic effects on common traits would differ substantially between populations: no, causal effects are nearly identical between populations (https://pubmed.ncbi.nlm.nih.gov/36941441/).
4. That cognitive function is under strong and recent selection: No, IQ exhibits one of the weakest signals of selection if not entirely neutral.
5. Recent polygenic selection is widespread and acting on many traits (e.g. Steve Hsu predicting this: https://infoproc.blogspot.com/2012/08/recent-human-evolution-european-height.html): No, evidence of polygenic selection turned out to be a false positive and if it exists at all it is so weak as to be undistinguishable from population structure.
6. There will be large rare variant effects on IQ and rare variants explain the bulk of heritability. Nope, recent exome-wide studies show tiny contributions of rare variants (https://www.nature.com/articles/s41588-023-01398-8).
The missing heritability "problem" is just a specific example of a broader problem where hereditarians overfit to observational data and now that none of their forward-looking predictions about genetics have held up, they are inventing artifacts (or, as you are doing, simply continue to claim that it "all points in one direction").
Admixture analysis was originally suggested by environmentalists because they thought it would disprove the hereditarian thesis. Then when it didn't, they changed their mind and decided that it didn't mean anything. This is a consistent pattern of environmentalists for 50 years. There remains no hypothesis that can explain, at all, racial gaps in IQ by environmental means, whereas the genetic hypothesis is parsimonious and fits the data. All attempts to increase intelligence of low-IQ racial groups, up to and including relocating children to new families have failed miserably. Maybe environmentalists struck a pot of gold this time with the missing heritability problem, but, more likely, it's just a complex puzzle that will take many decades to solve without changing the basic picture.
It turns out that the scientific method is a test of specific hypotheses, not a test of the people that proposed them. Where scientists propose parameters and then develop methods to estimate them in an unbiased manner, "race scientists" simply walk through the garden of forking paths until they find a correlation that "looks right". Regression to the mean in siblings, admixture analysis, the method of correlated vectors / Jensen effect (essentially all of The G Factor), the national IQ nonsense, are all examples of this. When it is pointed out that their analysis in fact cannot distinguish between multiple hypotheses (often trivially so), they do not attempt to improve their methods -- as an actual scientist would -- but either fall back on an appeal to some environmentalist authority or argue that they're just providing one piece of evidence in a broader narrative, as if a pile of garbage can turn into gold if you simply pile it high enough.
You will note that the 6 examples I mentioned above were not "X hereditarian said analysis Y would prove hereditarianism" but "X parameter turned out to be Z instead of Y which is not consistent with a hereditarian model". Also none of the points were about "the missing heritability problem", which seems to have become a kind of verbal tic to avoid actually having to consider new evidence. Highly correlated effect sizes across populations have nothing to do with missing heritability. Estimates of indirect/cultural transmission have nothing to do with missing heritability. You get the idea. The thing to do here is to acknowledge that your statement "decades of cumulative research all pointing in one direction" was incorrect, and grossly so.
Re: 1, what's the best way to disentangle VCT from other types of rGE/indirect effects? Do we have any priors on how prevalent one should be over the other?
I ran some simulations here assuming a model where cultural transmission is directly on the phenotype that spouses are sorting on:
http://gusevlab.org/projects/hsq/#h.b5cuc8jc2dvc
It gets estimates that are ballpark consistent with (non-genetic) multigenerational studies. However, it is almost certainly the case that assortative mating is NOT just happening on the phenotype, so this is a very simple model. I think it would be helpful if behavioral geneticists proposed more explicit models of rGE besides VCT and how to test them.
I wouldn't hold your breath on that last one. Seems like the field mostly just ignores it.
"Nothing to see here. IQ heritability upper bound stems not from GWAS studies, but from twin studies. I have no idea why he would think otherwise and that makes me question his motives."
Not really sure how this relates to any point made in the article. But twin studies cannot agree on a single estimate of heritability either: https://theinfinitesimal.substack.com/p/twin-heritability-models-can-tell