Dan the man's post on Race & IQ generated a lot of feedback. A lot. Those of you who are familiar with my weblog oeuvre know I used to be more interested in psychometrics. No more. Rather, if you don't want to believe in IQ or general intelligence, fine. My own experience is that very intelligent people (e.g, Mark, PhD physiology, now getting his MD, undergraduate background in physics) often are the most robust and cogent objectors to IQ or psychometric testing as a relatively useful reflection of intelligence. Dumb people know very well they're dumb, and they're not too coherent or interested in such abstract topics as whether the Raven's Matrices have instrumental value in predicting someone's ability to crunch through tasks requiring rational-abstraction heft.
Let's think like an economist; assume that intelligence tests are useful in measuring intelligence, that intelligence is around 50% heritable, so it has an underlying genetic component. With these assumptions it is theoretically possible that scientists might be able to map from genotype-to-phenotype with some level of plausibility. That an allele, X, might have an effect on the trait value of Y, within a population.
As it happens IQ genes haven't really been found. The behavior genetic data imply that loci which effect population level variation are going to be small, perhaps far less than 1% of the variation in IQ being due to any given polymorphic locus. That means there are perhaps thousands of extremely small effect alleles scattered through the genome which effect IQ. There might be levels of nuance within the genetic architecture, perhaps there are larger effect alleles which are found only among the extremely intelligent, 2 or more standard deviations above the norm (top 2-3% in IQ). So far the standard genetic methodologies haven't yielded anything promising, linkage studies which are better are looking for larger effect if rare variants didn't find anything, and newer association studies which are better at picking out smaller effect but more common variants within populations haven't picked up anything of a high likelihood (yes, IQ genes pop up every year, but do enough studies and you'll start digging up spurious associations here and there).
Obviously catching IQ genes is not a trivial task. They might be scattered and relatively numerous, but if their effect size is vanishingly small how will you distinguish the faint signal from the noise? In other words, those who tremble at the thought of differences in intelligence being adduced to genotypic variation probably shouldn't stay up at night with worry.
But times change. Here are some issues:
1) Genome sequencing is getting hella cheap, and way more thorough. What might it be like in 10 years? We're talking exponential increases in bang for the buck.
2) The computational power to munge its way through the data sets is getting more robust year after year. Again, exponential increase in bang for the buck.
3) It seems likely that widespread sequencing will be a habitual feature of the landscape in around 10 years due to considerations of preventive health. Personal genomics for this reason might be overhyped, but if the costs come down low enough people will likely do it even for marginal benefit.
But even then you might need really large sample sizes to get enough statistical power to pick up the IQ QTLs. This is why I think #3 will be important, if most people have a thumb drive (or whatever cool thing we have by then) with their whole genome the key issue becomes....
4) Getting people to do the tests.
This is where the concerns of the likes of Mr. Rose become important. There happens to be several large scale institutions which do a great deal of psychometric testing: the armed forces of various nations. Nations which have conscription have been some of the best sources for detailed longitudinal data because of this. If in 10-20 years most of these soldiers-to-be already have a personal genome hanging around, perhaps one of the Nordic countries who have socialized medicine where the government might keep a central database, it might be easy to cross-reference with the psychometric data.
As it is I suspect there will be a lot of resistance to this through lobbying. But I don't think it will matter in the long run, someone will figure out a way to get a large enough sample size, or synthesize the two disparate data sets. It's a big world, and I don't think this will be Big Science of the type which will require enormous outlays of funds. Perhaps for various reasons Western nations will manage to tamp down on exploration of IQ related loci. Well, we always have the Chinese, with a 10 million or more man army to order around.
Perhaps the technical hurdles will be too high, perhaps the QTLs are infinitesimal in a manner that Charles Darwin or R. A. Fisher would appreciate. But let's say that the Chinese state find many candidate loci which sum up to a substantial fraction of the presumed heritable variation. Can they keep this a state secret? In the age of the telegraph they could, but I doubt they could in 2020. There will naturally be objections. Epistasis? Or, perhaps people will discover the importance of "genetic background" (this is code for another word!) and claim that effects of allele A in the Chinese background don't apply to other populations. Perhaps there'll be no difference, and everyone can breath easy.
Thinking like an economist, presuming that the QTLs for IQ are discovered, neoeugenics, gene therapy or perhaps some sort of magical drugs (I have no idea how this would work, so "magical") which modulate the regulation of the QTLs in question will emerge. A race for intelligence might ensue, up to the physiological limit. People in the 2-4 standard deviation range are sometimes weird, but trust me, not that weird. At that point questions about differences within the population will be rendered moot as the ethical implications of differences between generations, and the extent to which parents will go to give their kids "an edge," will come to the fore.
The whole point there is that I don't think we're talking Large Hadron Collider or Human Genome Project scale big science. There won't be any issue of a huge debate about the ethical implication of the NIH throwing money in this direction: it simply won't happen, no way that any politically associated agency would fund such a project. Rather, it will happen when the technology makes it cheap enough that people can synthesize data sets from projects that are going on for other reasons (obviously lots of genomes + powerful computers are going to be an area where medical genetics is going to slouch), or, a rich individual funds it. If anything is discovered, it will spread at the speed of light. Of course, if many of the QTLs are discovered, and they differ between populations in a manner which suggest that allele differences map onto phenotype, preimplantation screenings could perhaps mitigate the difference within a generation. Well, unless all the white liberals turn Nazi because they snap after finding out that their socially constructed population has a somewhat different mix of alleles resulting in higher IQ, and all us colored folk get sent to the gas chambers because now they know we're retarded because of our genes (or in the case of East Asians, smarter because of their genes, rather than the evilness of white people).
Note More comfort, economists are idiots, so thinking like them might mean the whole project fails even in thought!
- Log in to post comments
There will naturally be objections. Epistasis? Or, perhaps people will discover the importance of "genetic background" (this is code for another word!) and claim that effects of allele A in the Chinese background don't apply to other populations. Perhaps there'll be no difference, and everyone can breath easy.
This is what really interests me. Will a growing portion of the public come to accept unegalitarian findings from this area as they gather? What do you think will be the trends in public opinion as the information gathers and is analyzed?
"The behavior genetic data imply that loci which effect population level variation are going to be small, perhaps far less than 1% of the variation in IQ being due to any given polymorphic locus. That means there are perhaps thousands of extremely small effect alleles scattered through the genome which effect IQ."
The data on MZ and DZ twins shows that there are enough variants with large effect that two children of the same parents can get substantially different genetic draws. If IQ variation was the result of very many alleles with extremely small effect, then the law of large numbers would effectively eliminate genetic variance among siblings. Instead, that genetic variance looks like uncommon to rare variants with moderate to large effects, maybe deleterious ones in mutation-selection equilibria (or pushed to higher frequency when swept along with something under positive selection).
Also, many variants with infinitesimal effects couldn't be very strongly (individually) selected for or against, and would tend to diffuse evenly through the population (which would prevent high measured heritability).
The data on MZ and DZ twins shows that there are enough variants with large effect that two children of the same parents can get substantially different genetic draws. If IQ variation was the result of very many alleles with extremely small effect, then the law of large numbers would effectively eliminate genetic variance among siblings. Instead, that genetic variance looks like uncommon to rare variants with moderate to large effects, maybe deleterious ones in mutation-selection equilibria (or pushed to higher frequency when swept along with something under positive selection).
yeah, this is fair. so why didn't linkage studies pick these up? cavalli-sforza used classical techniques to predict around 6 QTLs for skin color in the 1960s. the genomic data confirmed this prediction. i also presume you have the same objections to the idea that height is dispersed across small effect loci, since sibling variation is rather high as well, but the largest effect sizes are still rather modest.
Also, many variants with infinitesimal effects couldn't be very strongly (individually) selected for or against, and would tend to diffuse evenly through the population (which would prevent high measured heritability).
can you say more here. i think i understand what you're getting at, but i want you to be clear.
Will a growing portion of the public come to accept unegalitarian findings from this area as they gather? What do you think will be the trends in public opinion as the information gathers and is analyzed?
the public is irrelevant, the elites are all that matters, when it comes to the theoretical issues of knowledge. there'll be a lot of talk. as noted above, there's somewhat around 1 standard deviation difference across siblings. that means that there's likely variation within families due to segregating alleles. there will be some people willing to play the "natural experiment" with pre-implantation screens. if it works out, if the kids are in fact smarter (or for superbright parents the children do not regress toward the mean), then the talk will end and the cranks will start rolling for more practical applications.
If IQ variation was the result of very many alleles with extremely small effect, then the law of large numbers would effectively eliminate genetic variance among siblings.
i don't think so... afaik that math doesn't work out, but moreover chromosomes have a finite genetic map length, so siblings will share long stretches of their chromosomes IBD.
but moreover chromosomes have a finite genetic map length, so siblings will share long stretches of their chromosomes IBD.
iz great when you always got the basic physical genetics in the back of your pocket ;-)
"yeah, this is fair. so why didn't linkage studies pick these up?"
First, I'd say that linkage studies have good power to pick up quite large effects (I would call 10% very large), and that they set a bound from the other direction. Eventually, if we keep squeezing the space of possibilities from above and below, we'll hit something solid. Second, a very polygenic trait (brain development certainly fits the bill) might have relevant alleles roughly evenly distributed across the genome, with particular families having rare alleles at idiosyncratic places, such that when you look across heterogenous families it's harder to localize.
"i also presume you have the same objections to the idea that height is dispersed across small effect loci,"
Yes, for sufficient values of 'small.'
"Also, many variants with infinitesimal effects couldn't be very strongly (individually) selected for or against, and would tend to diffuse evenly through the population (which would prevent high measured heritability).
can you say more here. i think i understand what you're getting at, but i want you to be clear."
For terminological ease, I'd talk about a super-simplified model where a trait's value is equal to some maximum minus the number of relevant deleterious mutations (which can occur at many loci evenly distributed across the genome) multiplied by a conversion factor.
If we start with two parents with different densities of deleterious alleles evenly distributed across their chromosomes, then offspring will have trait values equal to the midpoint of parental trait values modified by genetic variance. If we multiply the number of alleles, but adjust the conversion factor to keep trait values the same, then the variance in the number of alleles in the offspring will increase less slowly than the number of alleles. With minimal genetic variance among siblings, any existing stock of population variation will decline as individuals with different trait values mate and their offspring converge on the population average.
If initially we had big population variation in trait values, then a mating between two individuals with very different values would have to produce some genetic variance for the reasons lol mentions (although each generation would further break up the initially differentiated stretches), and selection could slow the process, but it's hard to see where that variation would have come from (except perhaps the merger of two or more populations) without being smoothed out already.
First, I'd say that linkage studies have good power to pick up quite large effects (I would call 10% very large), and that they set a bound from the other direction. Eventually, if we keep squeezing the space of possibilities from above and below, we'll hit something solid. Second, a very polygenic trait (brain development certainly fits the bill) might have relevant alleles roughly evenly distributed across the genome, with particular families having rare alleles at idiosyncratic places, such that when you look across heterogenous families it's harder to localize.
ok, again, why haven't linkage studies and family based studies picked out these large-effect but rare loci?
Yes, for sufficient values of 'small.'
ok, so why are the biggest loci being detected so small? if you are saying there are bigger effect loci shouldn't they pop up first?
With minimal genetic variance among siblings, any existing stock of population variation will decline as individuals with different trait values mate and their offspring converge on the population average.
If initially we had big population variation in trait values, then a mating between two individuals with very different values would have to produce some genetic variance for the reasons lol mentions (although each generation would further break up the initially differentiated stretches), and selection could slow the process, but it's hard to see where that variation would have come from (except perhaps the merger of two or more populations) without being smoothed out already.
you're making some of the same arguments common in the late 19th century re: blending genetics. i don't see this as relevant since we have empirical data on plenty of quantitative traits which are maintained over time. there are also models where mutation replenishes variation fast enough to be at an equilibrium with other parameters. in any case, you can respond further if you want, but if i understand you correctly i don't really care what you have to say, since empirically it just isn't true. blending theories implies 50% reduction in variance per generation. that doesn't happen.
just to be clear to readers: classic blending theory implies that 50% of the genetic variance decreases per generation. so, 1/2^(n generations). converges to 0 very quickly.
"ok, again, why haven't linkage studies and family based studies picked out these large-effect but rare loci?"
They can't be very large, as Plomin and others have shown. I agreed that those studies put an upper bound on the typical effect size, and argued for a lower bound on the other end.
"but if i understand you correctly i don't really care what you have to say, since empirically it just isn't true. blending theories implies 50% reduction in variance per generation. that doesn't happen."
I was presenting a would-be reductio for the hypothesis of huge numbers of extremely small effect alleles, the whole point is that it doesn't actually happen!
"ok, so why are the biggest loci being detected so small?"
You mean in GWAS?
" if you are saying there are bigger effect loci shouldn't they pop up first?"
If you mean in GWAS, if the bigger effects are from harmful mutations, then they're not under positive selective pressure (except incidentally when swept along with something else before segregating, which would happen faster the larger the harmful effect) and won't become common enough for GWAS.
If you mean in GWAS, if the bigger effects are from harmful mutations, then they're not under positive selective pressure (except incidentally when swept along with something else before segregating, which would happen faster the larger the harmful effect) and won't become common enough for GWAS.
selection is irrelevant. allele frequency is relevant. but linkage studies haven't picked them up either. so no, i don't mean just GWAS.
Well, we always have the Chinese, with a 10 million or more man army to order around.
just a minor correction. see here.
We're in agreement that the linkage studies rule out a class of distributions of effect sizes, as I have written repeatedly. I think this may just be an argument about what quantitative values to attach to fuzzy terms like 'very small' and 'moderate to large.' You seem eager to interpret me as denying the facts I acknowledged from the start, just as you interpreted me to be asserting classical blending when I was clearly using it for modus tollens. Anyway, this conversation is neither pleasant nor very productive, so I'll hold my peace after this.
Behavioralist,
So would you advocate for more Linkage Studies of IQ among those groups with high numbers of offspring and good records, like Amish, Mennonite, Hasidim??
If there are a very high number of QTLs with very small impact on intelligence, then another reason that the high-IQ variants would not achieve fixation is that very high IQ is not a selective advantage. I'd guess the selective advantage is probably at about the middle of the IQ curve. Anyone less lazy have data on that?