[Still have not read the post. Feel free to just point me at a section or something.]
Hold on, just to check, do you think there’s actually much empirical evidence in favor of there being only very little linear genetic effects on various personality traits? (So I’m excluding the theories you present, not at all to say they are irrelevant, just to focus on the basics of what we know.)
The types of evidence positively showing the existence of “chaotic” heritability (by which I mean “not a nice function of a big weighted sum of genes”) that I currently understand:
The difference between MZ and DZ heritability.
Actually seeing epistases, or more generally giving a nonlinear model that predicts better than “the nearest” linear model.
The failure to find linear effects after having looked hard enough that you should have found them if they are there.
Standard narrowsense heritability estimates saying “there’s barely any”. (ETA: maybe this is basically the same as 3.)
But I’m unclear what you even think of the MZ/DZ heritability numbers you looked at, in terms of personality. Generally I’m unclear which pieces of evidence of the above form, or of other important forms, we actually have.
A suggestion from Steve Hsu: personality traits are harder to measure, so existing studies use noisy measurements; probably there’s more heritability than we know, and more data would reveal plenty of linear effects.
I don’t have any reason to believe that “chaotic” heritability exists at all, among traits that people have measured. Maybe it does, I dunno. I expect “nice function of a big weighted sum of genes” in most or all cases. That ASPD example in §4.3.3 is an especially good example, in my book.
The main observation I’m trying to explain is this Turkheimer quote:
Finally, for most behavioral traits, PGS just don’t work very well. As I already mentioned, these days they do work well for height, and pretty well for some medical conditions or metabolic traits. The best PGS for EA accounts for 15% of the variance, which is superficially impressive, but see below. For most other behavioral traits for which we might hope to have good PGS, however, such as personality or depression, they barely work at all, consistently accounting for under 5% of the variance…
Modern genomics of human behavior has had a very surprising result: it has decreased estimates of heritability from where they stood 50 years ago in the twin study era. The common assumption back then was that the heritability of pretty much everything was around 0.5, if not higher, and no matter how skeptical one becomes about the causal meaning of the second decimal place of heritability coefficients, more than half the variance is a lot to reckon with. But with SNP heritabilities, and then PGS as real-world instantiations of SNP heritabilities, and then SNP heritabilities and PGS that have been appropriately corrected for between-family effects, those numbers are a lot closer to 0.1 than they are to 0.5. For many traits (personality, most forms of psychopathology) they are closer to 0.01.
That’s my starting point, especially the last sentence. If Turkheimer is wrong (that SNP heritabilities of most personality and behavior measurements tend to account for well under 5% of the variance, and are often closer to 1%), then I’d be very interested to know that, and would probably want to revise parts of this post.
A suggestion from Steve Hsu: personality traits are harder to measure, so existing studies use noisy measurements; probably there’s more heritability than we know, and more data would reveal plenty of linear effects.
The missing heritability problem is the idea that twin and adoption studies find that every adult trait is around 50% heritable, whereas SNP heritabilities are lower than that.
Measurement noise does not by default explain missing heritability, because the twin and adoption studies have measurement noise too. Measurement noise can only explain missing heritability if the GWAS studies have more measurement noise than the twin and adoption studies. …And that can happen! GWASs have far larger sample sizes, so the same kind of measurement instrument which would be cost-prohibitive in a GWAS may be fine in a twin study. And I think that’s part of what’s happening for IQ, i.e. it’s part of why there’s more missing heritability for IQ than (say) height, as I discuss in the OP (§4.3.2).
I’m open-minded to the possibility that this same thing (noisier measurements in GWASs than twin & adoption studies) is also true for personality and behavioral measurements. But I’m skeptical that that’s a big part of the explanation, if it’s even a factor at all. For one thing, there’s way more missing heritability for personality than IQ, as I understand it, but I don’t think it’s harder to get a low-noise measurement of personality than IQ. In fact, I’d guess it’s the opposite. Like, I would strongly guess that it takes many fewer minutes / survey questions to find out whether someone has a history of major depression, or to measure where they are on the introvert-vs-extrovert scale, than to measure their IQ, with equivalent noise levels. In fact, I wonder if the measurement instruments are even different at all between the GWASs and the twin & adoption studies, in the case of personality & mental health. This could be looked up, I suppose.
[Still have not read the post. Feel free to just point me at a section or something.]
Hold on, just to check, do you think there’s actually much empirical evidence in favor of there being only very little linear genetic effects on various personality traits? (So I’m excluding the theories you present, not at all to say they are irrelevant, just to focus on the basics of what we know.)
The types of evidence positively showing the existence of “chaotic” heritability (by which I mean “not a nice function of a big weighted sum of genes”) that I currently understand:
The difference between MZ and DZ heritability.
Actually seeing epistases, or more generally giving a nonlinear model that predicts better than “the nearest” linear model.
The failure to find linear effects after having looked hard enough that you should have found them if they are there.
Standard narrowsense heritability estimates saying “there’s barely any”. (ETA: maybe this is basically the same as 3.)
In this section you mention MZ/DZ: https://www.lesswrong.com/posts/xXtDCeYLBR88QWebJ/heritability-five-battles#4_4_2_Applying_that_analysis_plan_to_different_traits_and_outcomes
But I’m unclear what you even think of the MZ/DZ heritability numbers you looked at, in terms of personality. Generally I’m unclear which pieces of evidence of the above form, or of other important forms, we actually have.
A suggestion from Steve Hsu: personality traits are harder to measure, so existing studies use noisy measurements; probably there’s more heritability than we know, and more data would reveal plenty of linear effects.
I don’t have any reason to believe that “chaotic” heritability exists at all, among traits that people have measured. Maybe it does, I dunno. I expect “nice function of a big weighted sum of genes” in most or all cases. That ASPD example in §4.3.3 is an especially good example, in my book.
The main observation I’m trying to explain is this Turkheimer quote:
That’s my starting point, especially the last sentence. If Turkheimer is wrong (that SNP heritabilities of most personality and behavior measurements tend to account for well under 5% of the variance, and are often closer to 1%), then I’d be very interested to know that, and would probably want to revise parts of this post.
The missing heritability problem is the idea that twin and adoption studies find that every adult trait is around 50% heritable, whereas SNP heritabilities are lower than that.
Measurement noise does not by default explain missing heritability, because the twin and adoption studies have measurement noise too. Measurement noise can only explain missing heritability if the GWAS studies have more measurement noise than the twin and adoption studies. …And that can happen! GWASs have far larger sample sizes, so the same kind of measurement instrument which would be cost-prohibitive in a GWAS may be fine in a twin study. And I think that’s part of what’s happening for IQ, i.e. it’s part of why there’s more missing heritability for IQ than (say) height, as I discuss in the OP (§4.3.2).
I’m open-minded to the possibility that this same thing (noisier measurements in GWASs than twin & adoption studies) is also true for personality and behavioral measurements. But I’m skeptical that that’s a big part of the explanation, if it’s even a factor at all. For one thing, there’s way more missing heritability for personality than IQ, as I understand it, but I don’t think it’s harder to get a low-noise measurement of personality than IQ. In fact, I’d guess it’s the opposite. Like, I would strongly guess that it takes many fewer minutes / survey questions to find out whether someone has a history of major depression, or to measure where they are on the introvert-vs-extrovert scale, than to measure their IQ, with equivalent noise levels. In fact, I wonder if the measurement instruments are even different at all between the GWASs and the twin & adoption studies, in the case of personality & mental health. This could be looked up, I suppose.
Gotcha, thanks!