They're obviously wrong and engaged in puffery when they claim to be SOTA. Look at http://www.nature.com/mp/journal/v20/n2/full/mp2014188a.html , for example - that was 3 years before and had a similar PGS of 1.2%. Or, if that's too obscure, look at the SSGAC EA2 / Okbay/Selzam et al 2016 (no excuse for not knowing that one), 2 years before, and more than double, at 3.5%.
> 3.5% genetic association with intelligence means that 96.5% is not genetic.
No. It does not mean that. As I literally just explained to you: a specific PGS variance != SNP heritability != heritability. (How could you possibly think that when I also just explained to you in my other comment that the current IQ SOTA is 11% and you replied acknowledging as much?) The best estimate of intelligence heritability in adulthood at the peak, using measurement-error-corrected estimates, is 70-80%. Very far from 3.5%. Or 11%, or 30%, or 50% for that matter.
"Heritability" includes non-genetic factors, and combined with genetic factors accounts for ~50% of variation in intelligence.
For genetic factors alone, the consensus so far is that there is very weak correlation of genetic variation with intelligence, unless a not quite-yet-established polygenic score is constructed. To wit:
>From the 1990s until 2017 no replicable associations were found. GPS from these early GWAS, which we refer to as ‘IQ1’, predicted only 1% of the variance of intelligence in independent samples. It became clear that the problem was power: the largest effect sizes of associations between individual single nucleotide polymorphisms (SNPs) and intelligence were extremely small, accounting for less than 0.05% of the variance of intelligence.
However, things got "better":
>in 2016, a second meta-analytic GWAS analysis with a sample size of 294,000 identified 74 significant loci30. This analysis produced a GPS, EA2, that predicted 3% of the variance in years of education on average in independent samples30.
It stretches the imagination how one would go from these numbers to claiming that "The best estimate of intelligence heritability in adulthood at the peak, using measurement-error-corrected estimates, is 70-80%.
> "Heritability" includes non-genetic factors, and combined with genetic factors accounts for ~50% of variation in intelligence.
No, it doesn't. Heritability is intended and defined to include genetic factors only, unless you are referring to indirect genetic effects which are mediated through the environment but are still genetic in origin.
> For genetic factors alone, the consensus so far is that there is very weak correlation of genetic variation with intelligence, unless a not quite-yet-established polygenic score is constructed. To wit:
You are again being extremely misleading and selective in your quote-mining, and you omit all of the qualifiers that Plomin & Stumm include, eg "Recent genome-wide association studies have successfully identified inherited genome sequence differences" - 'identified genome sequence differences' does not imply that they are the only genes involved. Merely that some have been identified, and more will be. As they then go on to point out in the parts of the review you ignore, future GWASes will surely make much more progress in identifying SNPs, and approach the SNP heritability ceiling of ~30% given enough data.
> It stretches the imagination how one would go from these numbers to claiming that "The best estimate of intelligence heritability in adulthood at the peak, using measurement-error-corrected estimates, is 70-80%.
Er, because I'm not talking about SNP heritability estimates. I am talking about estimates from twin studies, adoption studies, and MZA studies, which estimate total genetic contributions. The lower-bound is 50%, you don't get anything lower from any twin studies, and this is deflated by the fact that many twin studies are looking at young kids and the Wilson Effect has not yet kicked in; this is then further deflated by the fact that they use noisy IQ measurements, which inflate the error variance component and hide the shared-environment/heritability components. Correct for those, and you get up to 70-80%. All of this is covered in the Plomin textbook cited there, starting pg171, where they cover things like how MZA designs deliver 72% heritability estimates and they explicitly mention that measurement error biases downwards ('Corrected for unreliability of measurement, heritability estimates would be higher...'). They are being highly conservative by using a lower bound of 50%, and they are wrong to, because as I already mentioned family-GCTA/GREML-KIN indicates that semi-rare variants alone get you up to 50% heritability, so we can be sure that 50% is a considerable underestimate (which is consistent with their expanded textbook discussion noting that the real heritability is much higher).
No, I do not. Linking to Shalizi, who systematically misrepresents the issue and ignores all the evidence against his claims ( https://humanvarieties.org/2013/04/03/is-psychometric-g-a-my... ) is not an impressive thing to do, nor does it defend in any way pcrh's persistent errors and misquotations and misrepresentations of SNP heritability & PGSes (which rhetorical strategies and arguments bear a remarkable resemblance to Shalizi's own attempts to FUD intelligence & heritability, which have failed so spectacularly he hasn't dared address the issue since Rietveld et al 2013 came out...).
Let me get this straight, you contend that we should believe that Shalizi is an unreliable source and your justification is a post on a "human bio-diversity" blog run by white supremacist and anti-semite John Fuerst amongst others.
It would be bad enough if you were linking to a rebuttal of the post on heritability linked to by the parent, but this is a blog post attacking a different post by Shalizi.
How is this in any way relevant? More importantly, why would this convince anyone that Shalizi gets heritability wrong?
The link does nothing to refute Shalizi's point. It basically amounts to saying:
-Other models of intelligence as wrong and g is real because tests results correlate with each other, which is completely orthogonal to Shalizi's point. Because of the very nature of testing the results are going to correlate with each other; the most trivial example is, using Taleb's example, if dead people take the test. Dead people are dead and have an IQ of zero, and this is enough to increase the correlation (https://twitter.com/nntaleb/status/1114977580053147648).
-Apparently confirmatory factor analysis means g is relevant because...confirmatory factor analysis means g is relevant? Nothing addresses the fact that g could be an artifact that arises in any set of measures that correlate with each other regardless of the origin. Adding to that, there is scant evidence of any link between g and the underlying neurobiology and/or actual causal genetic mechanisms, beyond very trivial cases such as obvious malformations etc. On the contrary, neurological findings are pointing to the opposite direction (https://www.ncbi.nlm.nih.gov/pubmed/23259956).
-Even if we have no idea how g works it's still useful for predictions for stuff like educational attainment so it's relevant. Well duh, if you're good at taking tests you're going to do well in an environment where your attainment depends on your doing well in tests. Again, there is enormous circularity in reasoning. And as for the other outcomes, look at it this way: you could combine many measurements from your body (say, various analyzes from blood samples, but even trivial things like number of legs or amount of blood in your body could do), find they all correlate with each other (because when your body is breaking down it is likely damage to one organ that's spreading damage to other organs etc.), and from all that stuff you could draw a general "health quotient", or HQ. It would be highly heritable, there would be literal millions of SNPs that are associated with it. It would have immense predictive value for a variety of social outcomes. Does this mean that HQ reflects a unidimensional, physical reality in your body? Does this mean medical practitioners would have any use of it? And more crucially, does this mean that attempting to optimize for it (as opposed to, you know, optimize for actual real things that we can readily see) wouldn't be an utterly fruitless endeavor?
One last thing, since people seem to love brandishing GWASes around like a magic wand: people in the community are becoming increasingly disillusioned with them these days. (The head of EMBO, Ewan Birney, recently spoke about this. Note that I'm talking about academic geneticists, not the small clique of psychologists/"independents" from institutes with shady funding whose names invariably pop up when dealing with the less savory elements of the debate.) Again, for a geneticist, you need to find an actual genetic mechanism to prove something is "genetic", otherwise you end up with ludicrous associations (https://twitter.com/SbotGwa/status/1151964472921219073, or a GWAS for type of fat used in cooking). This has not deterred people from less rigorous fields such as psychology to attempt to GWAS for literally anything. A charitable interpretation is that people conducting such studies are unusually naive, a less charitable one would be to draw connections among the aforementioned small clique and correlate them with the views people in said clique have expressed, and reasonably infer that a motivated agenda might be at play here.
Of course I'm not looking to convince you since you appear to have read a truly gigantic amount on the subject and probably have a list of well-prepared rebuttals for literally any argument so that anyone with less time for online posting would never have the last word, so this is more of a post for any passer-by who'd think that IQ has any relevance beyond some niches of psychology. It's certainly not a thing in mainstream popgen or any field of genomics and if someone presented an abstract at any conference dealing with the kind of stuff posted on humanvarieties they'd be laughed out of the room.
> -Other models of intelligence as wrong and g is real because tests results correlate with each other, which is completely orthogonal to Shalizi's point. Because of the very nature of testing the results are going to correlate with each other; the most trivial example is, using Taleb's example, if dead people take the test. Dead people are dead and have an IQ of zero, and this is enough to increase the correlation (https://twitter.com/nntaleb/status/1114977580053147648).
No, it's not. There is no reason whatsoever that there has to be a positive manifold, attempts to construct batteries without it have failed, this is one of the most important empirical reasons for g, and you leading with this point demonstrates you didn't even bother to read the rebuttals to Shalizi.
> -Even if we have no idea how g works it's still useful for predictions for stuff like educational attainment so it's relevant. Well duh, if you're good at taking tests you're going to do well in an environment where your attainment depends on your doing well in tests. Again, there is enormous circularity in reasoning
This ignores the enormous number of correlations (and causality, via PGSes in within-family settings) with everything from brain volume to longevity, and also the fact that adding education as a mediation does not eliminate those correlations. The only circularity here are the strawmans you are erecting.
> It would be highly heritable, there would be literal millions of SNPs that are associated with it. It would have immense predictive value for a variety of social outcomes.
The heritability of longevity is actually quite low. You should think about why your thought experiment didn't work.
> The head of EMBO, Ewan Birney, recently spoke about this.
Birney is deeply ignorant. He makes mistakes all the time. For example, he claimed recently that embryo selection had never been done, not even in animals; actually, it's been done commercially for years. Or his most recent op-ed, where he confused SNP heritability with heritability and claimed that demonstrated that 90% of income is 'environmental' (a misunderstanding all too evident in this thread) - so embarrassingly obviously wrong even his geneticist friends had to correct him and he issued a mea culpa. You shouldn't take your beliefs about 'disillusionment' about GWASes from a guy who doesn't even understand the basics about them and spouts off claims that he didn't spend 30 seconds googling.
> Again, for a geneticist, you need to find an actual genetic mechanism to prove something is "genetic", otherwise you end up with ludicrous associations (https://twitter.com/SbotGwa/status/1151964472921219073, or a GWAS for type of fat used in cooking).
No, you don't. This is an extremely narrow conception of things. Preferring a specific type of fat in cooking is no more ludicrous than, say, preferring tea to coffee (which is heritable and the GWAS works fine for).
> Of course I'm not looking to convince you since you appear to have read a truly gigantic amount on the subject and probably have a list of well-prepared rebuttals for literally any argument so that anyone with less time for online posting would never have the last word, so this is more of a post for any passer-by who'd think that IQ has any relevance beyond some niches of psychology.
Passerbys should take into consideration the quality of evidence being cited by this throwaway account, and to what extent they appear to have not read a truly gigantic amount on the subject, when they read sweeping descriptions of what 'academic geneticists' think etc.
> It's certainly not a thing in mainstream popgen or any field of genomics and if someone presented an abstract at any conference dealing with the kind of stuff posted on humanvarieties they'd be laughed out of the room.
Results from SSGAC etc are extremely mainstream. They are published in Nature and similar outlets. Papers use IQ/EDU PGSes and the genetic correlations all the time for many purposes. Look at the reverse citations for papers like Okbay or Selzam. 'laughed out of the room' indeed!
>No, it's not. There is no reason whatsoever that there has to be a positive manifold
Yes, there is. You didn't understand Taleb and Shalizi's point obviously. Look at the tweet.
>This ignores the enormous number of correlations (and causality, via PGSes in within-family settings) with everything from brain volume to longevity, and also the fact that adding education as a mediation does not eliminate those correlations. The only circularity here are the strawmans you are erecting.
"Adding education as a mediation" does a lot of work in this sentence here.
>The heritability of longevity is actually quite low. You should think about why your thought experiment didn't work.
Where in the world did I mention longevity?
>Birney is deeply ignorant.
Lol, so the head of the most prestigious center for genetics in the world is ignorant. Is Graham Coop ignorant as well? How many genomics professors are ignorant then? Also you didn't even bother to research what he actually said. He protested in response to the disillusion about GWASes from the community (again, we're talking about the community of academic geneticists with a PhD in biology, not social scientists or psychologists) and he was in favor of some of their predictive power.
>No, you don't. This is an extremely narrow conception of things.
Yes, you do. You have obviously never stepped in a single genetics lab or done any actual academic research if you think otherwise. That's how genetics works: you've got to show a genetic mechanism to show something is genetic. It's like the "all biologists are from Missouri" thing. GWASes suggest things, they may nudge in a direction, but they do not prove. This is like basic stuff.
>Preferring a specific type of fat in cooking is no more ludicrous than, say, preferring tea to coffee (which is heritable and the GWAS works fine for).
It's absolutely ridiculous and if you don't think otherwise I don't know what to tell you. The fact that you can GWAS for literally anything should be evidence against their predictive power, not for it.
>Passerbys should take into consideration the quality of evidence being cited by this throwaway account, and to what extent they appear to have not read a truly gigantic amount on the subject, when they read sweeping descriptions of what 'academic geneticists' think etc.
I use throwaway accounts because I can't be bothered to remember my password for this site. And, in case it wasn't obvious, and anyone may feel free to believe me or not: I happen to do genomics research for a living. It's literally my job, and I interact, for better or for worse, with hundreds of similarly minded people who do similar things. I didn't bother to read psych papers from 1983 but I do know the mainstream consensus, in 2019, in the genomics community (again, not in the social sciences).
>Results from SSGAC etc are extremely mainstream. They are published in Nature and similar outlets. Papers use IQ/EDU PGSes and the genetic correlations all the time for many purposes. Look at the reverse citations for papers like Okbay or Selzam. 'laughed out of the room' indeed!
Again, SSGAC is a social science institute. And again, doing a bunch of GWASes to "predict" (the word is used loosely) social outcomes is not how actual biology works. No, rallying behind the magic card "but I controlled for things!" does not work either. Biologists now how complicated things can be and are very hesitant to put forward causal mechanisms behind even minor, uncontroversial traits unless they're absolutely sure of what they're doing, and it usually involves tremendous amounts of money and often new technologies. That a bunch of social science guys put forward such very bold conclusions (involving, once again, a very distorted meaning of the word "predict") without batting an eye should tell you more about the researchers' bias than the "insight" they purported to discover.
Again, you just try to get a talk at the Biology of Genomes in CSHL or at the EMBL about the heritability of preference for cooking oil, see how that works.
