Thanks, will take a look! I completely agree that the Flynn effect is one of the most interesting findings in modern behavioral genetics and underscores how the estimates that come out of twin studies -- even if we do land on a model that's "true" -- are not telling us anything about how "innate" a given trait is.
My article is partly about the Flynn Effect, but it's also about Flynn's theory that there are many types of environmental influence that aren't measured by twin studies and end up getting lumped into genetic influence. There's the individual multiplier, which are the effects that individuals themselves have on their environment, and the social multiplier, which are environmental differences that exist between different societies. Also, my article is about the way research based on twin studies contributes to racism here in the USA.
It's a really interesting perspective, thanks for writing it. Yeah, twin studies will lump in all of these multipliers as genetics. I also agree that the hereditarian movement has pretty much ignored decades of data from genomics because it is not convenient to the narrative.
Excellent! All this makes me wonder… is this common knowledge now in the genetics community, especially in the area of behavioral genetics? Or are most people still pretending that these complexities don’t exist?
Thank you! I think the general view in the genetics community is to treat twin-based estimates "seriously but not literally", in that they tell us something about traits being heritable but you shouldn't interpret their point estimates to the letter. Credible geneticists routinely publish papers speculating that twin studies are inflated (e.g. https://www.nature.com/articles/ng.3912 or https://pubmed.ncbi.nlm.nih.gov/30104764/) and don't get cancelled. Within behavioral genetics I think there is more credulity towards this design because it has been such a fundamental part of the field for so long. Most twin studies will include a few sentences about environmental assumptions in their Discussion, but there is not a lot of interrogation beyond that (in part because it is just difficult). The development and application of extended twin studies and attempts to bridge family / twin analyses is quite new and, I think, a very welcome shift. The other shoe that has yet to drop is how we think about genetic relationships across traits in the context of cross-trait assortative mating (https://www.science.org/doi/10.1126/science.abo2059).
Can you perhaps try writing in a manner where you much more clearly explain all of the terms and leaps of logic. It is very hard to tell even what you are concretely claiming, let alone how it interacts with the specific claims in Plomin's book, for example.
You throw around words like the path dependent model, and estimates without showing all of the work to get from point A to point B.
I know everyone does this. I know that books by people who you disagree with, such as Plomin's specifically, are full of numbers and words being tossed out without being clearly explained. People defending every type of claim frequently write long impenetrable paragraphs of scientific jargon, and then finish it by triumphantly explaining, 'and this is why my political and social views are correct, and the people who disagree just aren't paying attention to the evidence.'
But, I think you are trying to move the conversation by saying something that is well supported by the academic evidence, and that contradicts what we've all been told the state of the science is for the last seven or eight years. If you actually want to change minds, you should work a bit harder on bridging the inferential gap so that people who are willing to listen to an argument that twin studies overestimate genetic effects can understand what the argument is without spending three days googling terms, rereading paragraphs, and staring at lines of equations and wondering how the second line claimed to be equivalent is derived from the first.
I'm happy to work on it but it would be helpful to know which specific parts were confusing [for example, I didn't use the words "path dependent model" anywhere :) ]. There's a tough balance between summarizing findings from many papers and going into sufficient detail on each one.
This is a great question. Unfortunately due challenges of collecting and defining cases at large scale, much less is known about confounding for mental health phenotypes. It is the case that the molecular/twin heritability gap is much greater for many behaviors than for more conventional measurements: https://www.nature.com/articles/s41398-017-0046-x . But I don't think we know yet why that is as well as we do for something like educational attainment.
A point I sometimes see regarding the equal environments assumption is that, for correlations between siblings it doesn't matter much whether they are same-sex or opposite sex. For IQ, same-sex pairs might correlate .45 and opposite pairs at .42 or something like that (while mz twins are often above .80). It seems plausible that your gender should substantially influence how you experience your environment, and that same-sex pairs should have more similar environments. Particularly in the past, if gender roles were more strongly enforced then. A priori, I'm not sure if I would think the EEA is more wrong for same-sex vs opposite or for mz vs dz. Yet, if EEA is supposed to explain why mz correlations are so high, it must be much much more wrong in the latter case than the former.
To me, this point counts against the idea that a substantial part of the reason why mz twins are much more strongly correlated than dz twins is that the EEA is wrong. What do you think?
It's certainly an interesting idea and seems plausibly consistent with the EEA. But I think the core issue is that behavioral genetics has identified almost no environmental factors correlated with behaviors, even for behaviors with a very large E component estimated in twins (see Turkheimer and Waldron: https://psycnet.apa.org/record/2000-03445-004). That means we don't actually know what to look for in terms of the EEA being violated. It could be some seemingly small environmental cause that has been overlooked, or a high-dimensional combination of many small causes. But if we don't know what to look for it's very hard to draw any strong conclusions from circumstantial evidence like sibling similarities.
Yes, I agree the non-shared environment is weird and unpredictable. Interesting to see Turkheimer bring up “emergenesis” also in this paper (a form of GxG in which “the phenotypic character is determined by the interaction of independently heritable traits”). It is also mentioned in a recent paper he co-authored, which I liked: https://www.cambridge.org/core/journals/development-and-psychopathology/article/simulated-nonlinear-genetic-and-environmental-dynamics-of-complex-traits/4CB85E461BF1C693BBCC0369DB4A35A6. GxG seems not to exist for physiological traits. But if it does exist for psychological traits, as Turkheimer seems to suspect, this could help explain missing heritability. And it would leave more room for shared-environmental effects in twin studies, since they could help explain why rMZ still tends to be only 2*rDZ, and not more.
Thanks for summarizing. Seems to be a bit of a mess though. The reason Rao et al. report that multiple models were consistent with the data is because one cannot compare non-nested models by just looking at their Chisq statistics. Simulations also don't make much sense when we have a half dozen studies showing substantial differences in the parameter estimate when the EEA is relaxed (including Rao et al. but also Bingley (2023) if you want analyses from the current century).
Strictly speaking you cannot, this is why these papers had to test the EEA violation explicitly in a nested model (and found it to be violated). As a crude approximation you could compute AIC/BIC statistics if the model likelihoods were available, but that also assumes that one of the models being considered is really *right*.
This is quite an interesting deep dive into the genetic complexities of estimating heritability. I just wrote about this topic from the perspective of James R. Flynn's work on individual and social multipliers, that is aspects of our environment that twin studies can't measure. What do you think on that topic? https://open.substack.com/pub/eclecticinquiries/p/on-race-racism-iq-and-heritability?r=4952v2&utm_campaign=post&utm_medium=web
Thanks, will take a look! I completely agree that the Flynn effect is one of the most interesting findings in modern behavioral genetics and underscores how the estimates that come out of twin studies -- even if we do land on a model that's "true" -- are not telling us anything about how "innate" a given trait is.
My article is partly about the Flynn Effect, but it's also about Flynn's theory that there are many types of environmental influence that aren't measured by twin studies and end up getting lumped into genetic influence. There's the individual multiplier, which are the effects that individuals themselves have on their environment, and the social multiplier, which are environmental differences that exist between different societies. Also, my article is about the way research based on twin studies contributes to racism here in the USA.
It's a really interesting perspective, thanks for writing it. Yeah, twin studies will lump in all of these multipliers as genetics. I also agree that the hereditarian movement has pretty much ignored decades of data from genomics because it is not convenient to the narrative.
Very good, hard to fully understand from scratch, but I will definitely come back for more. I‘d love to see a simplified version of the article!
Excellent article! I learned a bunch.
The formatting of the equations isn't working on my Ipad or Iphone. The "+" signs don't show up in equations
This is simply fantastic. With a few refinements, it is exactly what the field needs.
Excellent! All this makes me wonder… is this common knowledge now in the genetics community, especially in the area of behavioral genetics? Or are most people still pretending that these complexities don’t exist?
Thank you! I think the general view in the genetics community is to treat twin-based estimates "seriously but not literally", in that they tell us something about traits being heritable but you shouldn't interpret their point estimates to the letter. Credible geneticists routinely publish papers speculating that twin studies are inflated (e.g. https://www.nature.com/articles/ng.3912 or https://pubmed.ncbi.nlm.nih.gov/30104764/) and don't get cancelled. Within behavioral genetics I think there is more credulity towards this design because it has been such a fundamental part of the field for so long. Most twin studies will include a few sentences about environmental assumptions in their Discussion, but there is not a lot of interrogation beyond that (in part because it is just difficult). The development and application of extended twin studies and attempts to bridge family / twin analyses is quite new and, I think, a very welcome shift. The other shoe that has yet to drop is how we think about genetic relationships across traits in the context of cross-trait assortative mating (https://www.science.org/doi/10.1126/science.abo2059).
Can you perhaps try writing in a manner where you much more clearly explain all of the terms and leaps of logic. It is very hard to tell even what you are concretely claiming, let alone how it interacts with the specific claims in Plomin's book, for example.
You throw around words like the path dependent model, and estimates without showing all of the work to get from point A to point B.
I know everyone does this. I know that books by people who you disagree with, such as Plomin's specifically, are full of numbers and words being tossed out without being clearly explained. People defending every type of claim frequently write long impenetrable paragraphs of scientific jargon, and then finish it by triumphantly explaining, 'and this is why my political and social views are correct, and the people who disagree just aren't paying attention to the evidence.'
But, I think you are trying to move the conversation by saying something that is well supported by the academic evidence, and that contradicts what we've all been told the state of the science is for the last seven or eight years. If you actually want to change minds, you should work a bit harder on bridging the inferential gap so that people who are willing to listen to an argument that twin studies overestimate genetic effects can understand what the argument is without spending three days googling terms, rereading paragraphs, and staring at lines of equations and wondering how the second line claimed to be equivalent is derived from the first.
I'm happy to work on it but it would be helpful to know which specific parts were confusing [for example, I didn't use the words "path dependent model" anywhere :) ]. There's a tough balance between summarizing findings from many papers and going into sufficient detail on each one.
To what extent do these methodological issues apply to estimates of heritability for mental health disorders? Thanks!
This is a great question. Unfortunately due challenges of collecting and defining cases at large scale, much less is known about confounding for mental health phenotypes. It is the case that the molecular/twin heritability gap is much greater for many behaviors than for more conventional measurements: https://www.nature.com/articles/s41398-017-0046-x . But I don't think we know yet why that is as well as we do for something like educational attainment.
A point I sometimes see regarding the equal environments assumption is that, for correlations between siblings it doesn't matter much whether they are same-sex or opposite sex. For IQ, same-sex pairs might correlate .45 and opposite pairs at .42 or something like that (while mz twins are often above .80). It seems plausible that your gender should substantially influence how you experience your environment, and that same-sex pairs should have more similar environments. Particularly in the past, if gender roles were more strongly enforced then. A priori, I'm not sure if I would think the EEA is more wrong for same-sex vs opposite or for mz vs dz. Yet, if EEA is supposed to explain why mz correlations are so high, it must be much much more wrong in the latter case than the former.
To me, this point counts against the idea that a substantial part of the reason why mz twins are much more strongly correlated than dz twins is that the EEA is wrong. What do you think?
It's certainly an interesting idea and seems plausibly consistent with the EEA. But I think the core issue is that behavioral genetics has identified almost no environmental factors correlated with behaviors, even for behaviors with a very large E component estimated in twins (see Turkheimer and Waldron: https://psycnet.apa.org/record/2000-03445-004). That means we don't actually know what to look for in terms of the EEA being violated. It could be some seemingly small environmental cause that has been overlooked, or a high-dimensional combination of many small causes. But if we don't know what to look for it's very hard to draw any strong conclusions from circumstantial evidence like sibling similarities.
Yes, I agree the non-shared environment is weird and unpredictable. Interesting to see Turkheimer bring up “emergenesis” also in this paper (a form of GxG in which “the phenotypic character is determined by the interaction of independently heritable traits”). It is also mentioned in a recent paper he co-authored, which I liked: https://www.cambridge.org/core/journals/development-and-psychopathology/article/simulated-nonlinear-genetic-and-environmental-dynamics-of-complex-traits/4CB85E461BF1C693BBCC0369DB4A35A6. GxG seems not to exist for physiological traits. But if it does exist for psychological traits, as Turkheimer seems to suspect, this could help explain missing heritability. And it would leave more room for shared-environmental effects in twin studies, since they could help explain why rMZ still tends to be only 2*rDZ, and not more.
No, what was the takeaway?
Thanks for summarizing. Seems to be a bit of a mess though. The reason Rao et al. report that multiple models were consistent with the data is because one cannot compare non-nested models by just looking at their Chisq statistics. Simulations also don't make much sense when we have a half dozen studies showing substantial differences in the parameter estimate when the EEA is relaxed (including Rao et al. but also Bingley (2023) if you want analyses from the current century).
Strictly speaking you cannot, this is why these papers had to test the EEA violation explicitly in a nested model (and found it to be violated). As a crude approximation you could compute AIC/BIC statistics if the model likelihoods were available, but that also assumes that one of the models being considered is really *right*.