Still thinking about Peter Visscher's essay and reply. The last point in his reply accuses GWAS skeptics of moving goalposts. This takes some chutzpah on the part of the GWAS community. A thread: /1
You may be able to accuse someone of moving goalposts, but not me. I was kicking field goals on this subject in 1998, when GWAS was a twinkle in human genetics' eye. /2
I wrote two long chapters referencing Visscher's work in 2011. Given the date, they were pretty accurate about where we were headed. /3
Self-citation aside, the point is that GWAS-world has been moving goalposts since the inception of what was originally called "molecular" behavioral genetics. It has been conveniently forgotten that the point was supposed to be finding the genes that explained heritability. /4
Linkage analysis was supposed to find the genes for human intelligence, but it didn't happen. /5
Then candidate gene studies were supposed to find them, but that didn't happen either: /6
Then GWAS was supposed to find them. How is that going? Quick, what were the top five SNP hits in EA4? What genes are they near? If you don't know, it is because the field has quietly stopped caring about actually identifying genes. /7
Remember all the ballyhoo about "biological annotation" in EA3? That is the next thing to quietly disappear. Once we got to, "education appears to have something to do with the brain," there wasn't much else to say. /8
So now it comes down to prediction, which is reaching an asymptote, and Mendelian randomization. These may yet be useful for social science. That's fine, but it is difficult for investigators to forswear Plominesque boasting about the sweeping importance of genes. /9
And that's the problem when Visscher, who (it goes without saying) is a brilliant statistical geneticist, argues that genomics is going to inform social policy or be applied to education. His own science shows that it won't, and shouldn't, except in very limited ways. /10
And unfortunately, in this field in particular, overenthusiastic claims about the importance of our own work- something we all do- gets picked up by the wrong crowd. Anyway, it is GWAS-world that has moved the goalpost, not the critics. /end
• • •
Missing some Tweet in this thread? You can try to
force a refresh
Peter Visscher left a comment on my blogpost from last week, here: turkheimer.com/peter-visscher…. First things first, I got the journal where he published his article wrong. I fixed it. /1
Visscher doesn't like that I used the word "eugenic" in connection with his essay. But, broadly, eugenics refers to the use of genetics to explain existing racial, social and economic disparities. Uncritical application of between family GWAS does just that. /2
He doesn't grapple with my central point: once we know about unconfounded (really less confounded) within-family estimates, those are the ones we need to interpret and base our applications on. Good intentions don't rescue between family effects from eugenic implications. /3
So EA4 is finally out. It's a massive project, and I am not here to question its scientific validity, but rather to ask a tougher question: Has GWAS of complex human behavior turned out to be a disappointment? /1 nature.com/articles/s4158…
We are now over 3 million participants. A 3x increase in sample size produced a 25% increase in between-family R2. Remember when we were told that bigger samples would inevitably lead to scientifically or practically actionable results? Are we still waiting for that? /2
Maybe it is going to come in subsequent papers, but many of the old EA goals have been quietly abandoned. Not much interpretation of top hits, not much biological annotation. In its place just raw prediction, and it is still less than impressive. /3
Yesterday @kph3k noted that I "loathe" decile analysis as a way of describing the results of a PGS analysis. The subsequent discussion clarified why I loathe it-- it's a misleading way of reporting results, a systematic sleight of hand to disguise the import of a small effect. /1
To illustrate, I created some simple data for 10,000 observations. PGS has a mean of 0 and a SD of 1; IQ has a mean of 100 and a standard deviation of 15. They are correlated around .205, so the PGS accounts for about 5% of the variance in IQ. /2
PGS analyses are very simple-- it's just a simple regression. How to illustrate the result? The obvious way is to just draw the scatterplot with the regression line through it. It is what it is. /3
From time to time you hear it said that executive function, a cognitive ability involving mental control, is 100% heritable. It seems like a compelling basis for hereditarianism- under all that complexity there is a solid gold nugget of pure genetic variation! [wonky thread]
The @kph3k book, for example, says "First,it is nearly 100 percent heritable. That is, within a group of children who are all in school, nearly all of the differences in general EF between them are estimated to be due to the genetic differences between them." /2
This statement isn't wrong, exactly, but it is misleading if you don't understand what it means. The important thing is that the 100% h2 is estimated for a *latent* variable, an abstract statistical concept that "explains" why multiple measures of EF are positively correlated. /3
When I find myself besieged by the kind of people in my mentions this morning, it is tempting to be goaded into sounding as though I think genetic endowment places 0 limits on people, or that GWAS has made no meaningful contributions to science, or that the modest predictions /1
of behavioral PGS are completely useless. I don't think any of those things and I never have. I am, after all, a working empirical behavior geneticist. What I do think is that the hardcore genetic world the right is forever envisioning shows no signs of becoming reality. /2
The RDR h2 of EA, which controls for indirect effects, is .17. (nature.com/articles/s4158…) That is an UPPER LIMIT on the performance of PGS. And even that includes a completely unknown amount of red hair effects. /3
1/ Arguing with myself: I say that "race science" is meaningless because it assigns causal value to heritability coefficients whose causal implications are not understood and generally exaggerated.
2/ But by impugning the motives of the race scientists, aren't I shutting down the very research that might produce that knowledge? No. Let's say you believe that Group X carries a gene or some polygenic mechanism that makes them exceptionally good violin players.
3/ You want to do science to show you are right. How should you proceed? In fact you have a two-part hypothesis. The first is that there exists a genetic mechanism that reliably produces violin talent, one person at a time.