Mike White finally left a comment on my post Complex traits & evolution:
I'm trying to make a distinction between what geneticists call complex or quantitative traits (traits affected by different alleles of many different genes, with a quantitative range of phenotypes), and something I would call a physiologically complex (or complicated) trait.
Complex or quantitative traits include both height and intelligence. But I'm arguing that something like height is not physiologically complex the way intelligence is.
...
So, for example, in the case of height, you can imagine that it is easy for a single allele of large effect to reach high frequency in a given population, resulting in a fairly tall (or short) population.I don't think that such a thing is very likely for intelligence, because, unlike what I think is the case for height, single alleles of large effect on intelligence probably also have large deleterious effects (like neurological disorders) - something like intelligence is so physiologically complex that it is much easier to 'break' with a large-effect allele than something like stature, where an allele that extends how long growth plates in your leg bones are active (for example) is unlikely to also have major deleterious effects.
And to get race-specific differences, given current human genetic variation, you need single alleles with large effects - thus you have genetic differences between populations in height and skin color, but not IQ.
I disagree that alleles of large effect are necessary. We already have a large range of extant variation across human populations. From what I can tell about the Breeder's Equation QTLs of large effect are not needed.
- Log in to post comments
No, single genes aren't needed. A quantitative trait will diverge through drift like a single locus (something a lot of us have overlooked when working on this problem), regardless of the number of loci. You don't even need to invoke selection.
http://www.scientificblogging.com/adaptive_complexity/what_our_genes_te…
I tried to raise this point on Michael's blog a while ago, but eventually had to get back to work, without much progress. You can probably save time in this discussion by looking at the arguments Michael et al used there.
Selection is always happening, whether you invoke it or not. Drift can only make substantial changes in a trait if that trait is biologically near neutral: not gonna happen.
1. I apologize for not responding to your last blog post earlier. We've had some bugs with the traffic stats functionality on my blog, so I wasn't able to see your link until today.
2. The main issue I've had in conversations on my blog with Steve Hsu and Utilitarian and other, is not over what's possible in theory, but what is likely to be the case regarding genetic variation and population differences in intelligence among humans.
Of course, given enough time, you can in principle have significant phenotypic differences between populations dues to multiple alleles of small effect, either through drift or selection. It could, in principle, be the case that a large amount of between-population human genetic variation could be due to natural selection. You could come up with any number of hypothetical scenarios.
But what is likely to be the case? I'll admit that we don't have enough studies of the genetic basis of quantitative traits between human populations to definitively settle the question, contrary to what some psychologists have been claiming for decades.
So I'm offering an opinion regarding what I think is likely - an opinion shared to some degree by real population geneticists (I'll also admit I'm not a population geneticist), some of whom I cited back over at my blog, and others who live down the hall from my desk.
Look at recent studies of classic traits that distinguish populations, like skin pigmentation. The differences between populations appear to be due primarily to single alleles of large effect driven most likely by selection to high frequency in various populations. Why? Because that's much easier to produce over 50k years of small or moderate levels of selection. For the reasons quoted in razib's original post, I don't believe that intelligence is able to be modified so rapidly by large-effect alleles with out some extremely deleterious side effects.
Could there be significant population differences in IQ caused by multiple low-effect alleles whose population frequencies are due to drift? Again it's possible, but given that intelligence in the human species overall is under at least negative selection (and some have suggested stabilizing selection), and given how little between-population variation there is, I find it implausible.
So again, the question is not what's possible in theory with a given quantitative trait and 15% between-population genetic variation - I'm interested in what is likely to be the case in human populations, given human population history, the physiological complexity of cognitive ability, and the likely selection pressures over human history.
This whole issue cropped up on my blog because of a ridiculous statement by John Derbyshire. He basically said that roughly half of the black-white disparity in IQ scores (in the US) is due to genetics. Even if there are real genetic differences in average IQ between populations (and I don't buy the claim that IQ measures among ethnic groups in the US are representative of ancestral population IQ scores), there is no way any sort of drift or selection has produced a difference of that magnitude between Africans and Europeans - it's not physiologically plausible, and it's not plausible that selection pressures for intelligence were that different between Africans and Europeans.
The main issue I've had in conversations on my blog with Steve Hsu and Utilitarian and other, is not over what's possible in theory, but what is likely to be the case regarding genetic variation and population differences in intelligence among humans.
ok, i actually thought you thought that steve was making stupid theoretical claims. your claims seemed unfounded from any a priori theoretical grounds, ergo, i responded.
Look at recent studies of classic traits that distinguish populations, like skin pigmentation. The differences between populations appear to be due primarily to single alleles of large effect driven most likely by selection to high frequency in various populations. Why? Because that's much easier to produce over 50k years of small or moderate levels of selection.
yes, but as a point of fact many of these alleles which result in skin color variation aren't "small" or "moderate" levels of selection, and are very recent. e.g., SLC24A5 & OCA2 started sweeping in > 10,000 years BP. SLC45A2 on the order of 10,000 years. so the 50k scaling misleads as an empirical matter for this trait. also, skin color popped up on tests which try and detect haplotype blocks because you have relatively recent powerful sweeps. it stands to reason that there a finite number of these big blocks responsible for extant variation. and GWAs and linkage studies would pick up these big affects. the current tests haven't offered much insight into traits controlled by a lot of diffused and numerous QTLs because they don't have the power too. that doesn't mean that selection isn't operating there just because we don't have much of a clue as the genetic architecture of quantitative traits.
Again it's possible, but given that intelligence in the human species overall is under at least negative selection (and some have suggested stabilizing selection), and given how little between-population variation there is, I find it implausible.
it just isn't true that there is little between-population variation. e.g., the black-white IQ difference is on the order of 1 standard deviation. IOW, the average black american is on the 15th percentile of white IQ. the average ashkenazi jews has an IQ about 2/3 of a standard deviation above that of a white gentile. IOW, 75th percentile of white IQ. you may hold that these between group differences are not due to genetics, but it is just false to contend that there aren't significant between group phenotypic differences .
there are a priori reasons to be skeptical of strong unidirectional selection for IQ. but, thousands of years is many generations, and all sorts of frequency dependent and environmentally mediated selection regimes may have operated differentially. or not. also the g-matrix.
it's not physiologically plausible, and it's not plausible that selection pressures for intelligence were that different between Africans and Europeans.
you keep neglecting the fact that within populations there is an enormous physiologically plausible range of IQ. we know there's a non-trivial heritability of IQ on the order of 0.50. you don't need new mutant variations of large effect, you have extant genetic variation. you can keep asserting that there isn't any selective regime, historical condition, etc. that's fine, and a matter of opinion. but i reject this idea that we'd need new mutations. there are plenty of non-pathological humans with IQs on the order of 70 walking around. hundreds of millions in fact. there are also plenty of humans with IQs on the order of 130. you walk down the hall and talk to plenty of them, and are almost certainly one yourself.
one can assert that the between group differences are due to non-genetic factors. plenty of population geneticists do. i'll table that. but there is a great deal of extant genetic variation within human populations, and no one denies that a substantial proportion of variation in intelligence is heritable. (even leon kamin does not hold to 0 heritability anymore) for selection to operate you need:
1) differential reproduction correlated with variation in phenotypes
2) a correlation between phenotypic and genotypic variation
physiologically plausibility is irrelevant if you're talking about a quantitative trait where we simply want to shift the mean value by selecting off of extant variation. people with high and low IQs in the normal range are equally fertile, even if they have a hard time communicating with each other.
ok, and just to clear to everyone....
1) estimates for heritability of IQ are between 0.2 and 0.8
2) that means that between 20% and 80% of the variation in IQ is due to variation in genes
3) we haven't found any genes which definitively explain this variation (yes, there are genes of big effect that cause pathological retardation, that's not what i'm talking about)
4) the inference is then made that the genes responsible for variation in IQ are innumerable and of very small effect size. perhaps on the order of tens of thousands of genes.
5) this means that there are many, many, many, genetic variants out there today in human populations which are likely to be responsible for changes in IQ, and, these variants are not implicated in any sort of extreme pathology
this is why i see the point of mike's comments, but i think they're off base. i agree that a new gene of large effect is likely to fuck things up. that's why the cochran-harpending "overclocking" model for ashkenazi IQ fixates on cognitive disorders. but we have a big range of variation to select from now. quantitative genetic studies in fact have shown that you can obtain a population mean trait values above any value which was manifest in generation 1 of breeding of the population because as you change allele frequencies novel combinations which were not normally extant in the population become common.
Razib,
Nicely summarized. I'd like to ask Mike which assumption he disagrees with -- if any, since it should now be clear that the way that between group difference would arise is not necessarily what he had in mind:
0) intelligence is partially heritable
1) there is plenty of extant genetic variation, probably due to a large number of genes of individually small effect -- no additional mutations are required
2) selection can act if reproductive rates are impacted by these genes (i.e., by intelligence)
3) simple estimates suggest that 50,000 *could have* been enough time to produce .5 SD (genetic) group differences
See here for some elaboration on (2) and (3) using Greg Clark's (economist at UC Davis) data on English inheritances and family size:
http://infoproc.blogspot.com/2008/12/recent-natural-selection-in-humans…
The simplest model would be that time since development of agriculture varies between groups, and this variation leads to different levels of selection for traits which might be more useful for agriculturalists than for hunter gatherers. As Razib pointed out, there is a lot of evidence for group differences; the question is only how much is due to genetic factors.
Finally, everyone here is (I think) reasonable and realizes this matter is not yet *resolved* scientifically. The question is whether one can *exclude* possibilities based on current evidence. Mike is (almost) asserting that you can exclude the possibility of (genetic) group differences in intelligence -- e.g., he might be 95% or 99% confident that they cannot exist. We are really only arguing about what the appropriate probability estimate (confidence level) should be. For Mike to be so sure he must be able to exclude some of the assumptions (0)-(3) with relatively high confidence. Care to elaborate?
Re: fucking things up, there are disease candidates that hit Europeans* more than other racial groups -- Parkinson's and haemochromatosis.
Ferrous iron is a necessary cofactor for conversion of tyrosine to L-DOPA by tyrosine hydroxylase (the rate-limiting step in catecholamine synthesis), but also a pro-oxidant. In a genetic background where melanin-production knockdown mutants have made tyrosine more available for catecholamine synthesis, a mutant with increased iron uptake would be able to synthesize more catecholamines. The catch is that this has a tendency to result in greater accumulation of intracellular junk via oxidation of proteins -- Lewy bodies, for instance. (Maybe amyloid plaques, too.)
Oh and BTW, Parsis have an unusually high rate of Parkinson's. Certainly made me go "hmmm".
Whoops -- the asterisk in the above comment was because mutations in East Asians are very probably different; diseases that hit them more often could be informative.
"there is no way any sort of drift or selection has produced a difference of that magnitude between Africans and Europeans - it's not physiologically plausible, and it's not plausible that selection pressures for intelligence were that different between Africans and Europeans."
This is just silly. We _know_ that a major part of the variance in IQ is genetic in modern circumstances - which means that causative physiological differences are a large part of why some people have IQs of 80 while others have IQs of 150. Which is why identical twins are especially similar in IQ, why smarts runs in families, etc. Yet we don't know much about those physiological differences - other than brain size. Groups could vary, on average, in those same poorly understood factors, so how can anyone say that such differences are physiologically implausible? By the way, brain size varies between groups.
As for the implausibility of differences in selection factors - that's equally silly. You can have big responses to selection for IQ in a thousand years if the parents of the next generation average a measly point higher than the population from which they're drawn, a difference that might be barely noticeable over a human lifetime. Wealth selection alone could drive that. But wealth selection would be strongest in a Malthusian situation where the poor starve: that wouldn't be the case in a situation where populations were primarily limited by disease pressure.
Or consider pleiotropy: at least half, probably more, of all genes are expressed in the brain. What if population A experienced major selective pressures (from infectious disease, say) on a large set of genes that was not experienced by population B? Then there might well be cognitive side effects - negative ones, of course - in population A.
I would guess that this guy has never read any quantitative genetics. Most biologists have not. I would also guess that he is not familiar with the results of psychometrics - few people are.
Mike White wrote: "single alleles of large effect on intelligence probably also have large deleterious effects (like neurological disorders) "
How about astigmatism--that is pretty highly correlated with IQ.
Von
A few comments:
Razib:
That's why I think, given the likely lack of strong unidirectional selection on one population and not others, it's very unlikely that plausible levels selection drove significant changes in IQ between populations. Where are the populations that show the significantly reduced variance you would expect after the kind of strong selection (or a strong founder effect) that would be needed to explain the major differences between populations that supposedly exist?
I'm not saying they need to be new mutations. I'm saying that the most likely scenario for achieving population differences in IQ would be exactly the same sort of scenario we see for skin or hair color - single alleles of large effect reaching high frequency. Such mutations impacting IQ could very well exist in a population before major population-specific selection, but I'm arguing such mutations, when we're talking about IQ, are also likely to come along with large deleterious effects, and thus they would be unlikely to be driven to high frequency in a population.
Steve said:
I'm not so sure how believable this really is. There is certainly evidence for group differences within the cultures of developed nations. For example, you can see group differences between people of European, Asian, and African descent within the U.S. But is this really the same thing as measuring group differences between Europeans, African hunter-gatherers and rural Chinese - or between the 6-7 major ancestral populations used in the Human Diversity Panel? In other words, I really don't believe that we have accurately used a culture-independent metric to look at group differences in populations that have not been subject to the inevitable social pressures that you see when you put people of different backgrounds into a single society. And as Josh Witten pointed out in comments on my blog (the issue is also raised in the recent Nature news piece "The Case of the Missing Heritability"), estimates of heritability for complex traits have run into problems. Again, I'm not sure how confident we can be in the current estimates of population differences in IQ, nor am I that confident in estimates for heritability of IQ.
I don't doubt for a minute that there is genetic variation underlying some of the phenotypic variation in IQ. I do doubt that we have a solid handle on how much of the phenotypic variation is genetic, especially between populations. As Mountain and Risch put it in Nature Genetics:
We can move this discussion forward if anyone can provide a single example of a significant between-population difference in a quantitative trait that does not involve basically one large-effect allele at high frequency. True, such things are harder to find, but that's the type of scenario of between-group differences for which we are lacking evidence.
As for some of the lamer ad hominem attacks by some commenters:
I work in a lab that does quantitative genetics, and in a department that routinely brings in top speakers on quantitative genetics, as you would expect for the #4 ranked genetics department in the US. This issue isn't as resolved or as simple as you seem to think it is. The strength of selection to get an average of a 1 point increase in the IQ of the parents of every generation would have to be pretty damn strong - it's exactly that kind of claim that I find unrealistic.
Where are the populations that show the significantly reduced variance you would expect after the kind of strong selection (or a strong founder effect) that would be needed to explain the major differences between populations that supposedly exist?
well, you note that you know quantitative genetics, so you know the issue of reduction of genetic variance after truncation selection on quantitative traits is an empirical matter. there are cases where you can move many standard deviations from a mean through successive generations of selection, and the heritable variation has not decreased appreciably. the differences between population means re: IQ are not trivial, but, neither are the means outside of the "normal range" of any population. e.g., the black american mean IQ is well within the normal range of the white american population. the question of strong unidirectional selection is a real one, but it doesn't look like IQ is anywhere near its physiological limits on the high or low end, but rather it is probably stabilizing in the middle. but, the weight of the parameters stabilizing might differ from population to population.
I'm saying that the most likely scenario for achieving population differences in IQ would be exactly the same sort of scenario we see for skin or hair color - single alleles of large effect reaching high frequency.
i don't really buy this. you could be right. or not. but like i said, the normal range in human IQ doesn't seem to be bumping up against physiological constraints. going back to quantitative genetics, a lot of the work is just redistributing the weights of the variation you already have.
Again, I'm not sure how confident we can be in the current estimates of population differences in IQ, nor am I that confident in estimates for heritability of IQ.
it's fine if your confidence is low. the whole point of this discussion is that you seem to have taken the tone of "here goes another physicist making retarded theoretical claims which are absurd on the face of it." since this does happen, i thought i would prod you as to the details. steve obviously has different levels of confidence on the same set of propositions. he isn't ignorant, you just disagree.
We can move this discussion forward if anyone can provide a single example of a significant between-population difference in a quantitative trait that does not involve basically one large-effect allele at high frequency. True, such things are harder to find, but that's the type of scenario of between-group differences for which we are lacking evidence.
i'm waiting too. from what i can tell with current techniques it's like asking the wright brothers to make the transatlantic flight. if it's low frequency it better have a big effect, and if it doesn't have a big effect it better be high frequency. though where do you draw the line between large and small effect, high and non-high frequency?
"We can move this discussion forward if anyone can provide a single example of a significant between-population difference in a quantitative trait that does not involve basically one large-effect allele at high frequency."
For multiple large-effect alleles: several of the Ashkenazi diseases involve multiple alleles affecting the same locus in the same way at elevated frequencies. For multiple small-effect alleles, hypertension in African-Americans.
" True, such things are harder to find, but that's the type of scenario of between-group differences for which we are lacking evidence."
If we had massive studies with good phenotypic data instead of microarrays, so that we could point to the precise genetic architecture of highly polygenic traits where common variants play little role, then we wouldn't be having this discussion, and could just go consult the Personal Genome Project and 1000 Genomes datasets to test for genetic group IQ differences.
"The strength of selection to get an average of a 1 point increase in the IQ of the parents of every generation would have to be pretty damn strong"
A mean of parents (weighted for offspring numbers)0.1 standard deviations above the population mean would do it. Evidence from wills and the like show strong relationships between wealth and surviving offspring for centuries. IQ improves performance at both skilled and unskilled jobs, although this is more true as cognitive complexity of the job increases. Is that 'pretty damn strong' or not?
So you work in a lab that does quantitative genetics. But have _you_ read any?
Mike,
It still seems to me that you are talking your priors. That is, the arguments you just gave don't seem strong enough to warrant a 95 or 99 percent confidence level in your position. Do you agree or disagree?
You are adopting a position that I would call extreme skepticism, I suspect because you don't like the conclusion. That is, in ordinary scientific work (say we were talking about some insect species rather than humans) the level of evidence currently available on this issue would be sufficient for you to allow that some reasonable colleagues might adopt as a working hypothesis the idea of genetically-influenced group differences in phenotype. Instead, it seems you claim that this other group of researchers is clearly wrong.
Here's the summary as I see it:
0) intelligence is partially heritable
You didn't seem to disagree, although you pointed out that heritability is not well defined except within a particular environment. (If you search on the word "turkheimer" on my blog you will see I discuss this extensively in the context of IQ.)
1) there is plenty of extant genetic variation
You didn't seem to object to this observation. (You give an argument against few alleles of large effect, but no one is arguing specifically for that scenario.)
2) selection can act if reproductive rates are impacted by these genes
You didn't seem to object to this.
3) simple estimates suggest that 50,000 [years] *could have* been enough time to produce .5 SD (genetic) group differences
You rejected this as implausible, but you didn't give any quantitative reasons, and you didn't respond to Clark's data.
So, I think your argument boils down to (3), and to a question of plausibility. You reject (3) at 95 or 99 percent confidence level without even having done a rough calculation or simulation.
Do you not agree that height differs between populations and that it is likely influenced by a large number of genes of small effect? Do you see reduced variance in height despite (seeming) 1 SD variations between populations?
Finally, you quoted me:
Steve: As Razib pointed out, there is a lot of evidence for group differences; the question is only how much is due to genetic factors.
But then you didn't disagree with what I said -- I left it open as to whether differences were genetic in origin. You just fleshed out the issue. You seem to think the overwhelmingly most likely answer is that *none* of the observed differences have even partial genetic causes. I would have thought your prior should be some smooth distribution over values ranging from 0 to 100 percent.
PS Here is a link to one of the Turkheimer posts on my blog. Joy of Turkheimer.
Although he is a psychologist, in the talk I link to (given at Stanford to an audience of geneticists) he makes an analogy between statistical challenges faced by social scientists and geneticists in GWAS. One of the examples he uses is a GWAS in which genes affecting height are studied. No single gene of large effect is found even after a massive analysis, so height is most likely like intelligence in this sense, yet we seem to see group differences, no restriction of variance, etc.
See also here for his slides, audio etc.:
http://precedings.nature.com/documents/2642/version/1
The tests for intelligence are crap. They are not reliable enough to measure the differences that are being suggested as being important.
If you look at this paper, they use two different IQ tests, the WISC and Raven's progressive matrices and find on some individuals (yes, individuals), measured IQ differences of 70 percentile points (yes, 70 percentile points).
http://www.ncbi.nlm.nih.gov/pubmed/17680932
If two tests are supposedly measuring the same thing, and come up with measures that are 70 percentile points different on the same individuals, the tests are not measuring the same thing even if they get consistent results on other individuals and on other groups of individuals. The consistency observed in some groups must be an artifact.
If the same individuals have such different test results, it is nonsense to attribute smaller differences to "genetics" or "environment", or to something else. It is likely that the whole concept of "intelligence" is crap; that neurological functions cannot be distilled down into a single measure the way that height can be.
The brain is a lot more complicated than the skeleton, and needs to do things that are a lot more complicated over an individual's lifetime. The only time a brain can be optimized for certain mental tasks is while that brain is developing in utero. The infant brain is limited in size at birth by the maternal pelvis. The infant brain is necessarily a compromise between the different tasks that brain must perform over that individual's lifetime. A big brain is such a large advantage that humans have evolved a substantial risk of death due to cephalopelvic disproportion. Until the 20th century, too large a brain at birth was certain death for the infant and probable death for the mother.
The "obvious" way to increase brain-power is by having a bigger brain that can do more stuff. That doesn't work because it kills the infant and likely the mother. Brain-power has to be "optimized", which means trade-offs. I have blogged about what I think are the trade-offs along the autism spectrum, a theory of mind vs. a theory of reality.
Just to be clear, daedalus2u is talking specifically about autistics (click through to the paper). For the non-autistic ("normal") part of the population results on well-designed IQ tests tend to be highly correlated.
I think the comment is tangential to the main question, which is whether one can from purely population genetic arguments rule out genetically linked group differences of .5 or 1 SD in traits like IQ. This of course refers to population averages and hence by definition focuses on the "normal" part of the population.
Here's a thought: http://www.longbets.org/ could be used to bet on future findings in the area, which should be forthcoming in a few years, with the proceeds going to the charity of the winner's choice.
"The "obvious" way to increase brain-power is by having a bigger brain that can do more stuff. That doesn't work because it kills the infant and likely the mother."
Um, but there *are* variations in brain size, structure, and development that correlate (more strongly, explaining half the variation, with MRI measurement of brain structure) with individual and group differences in IQ.
Also, the greater the physiological disadvantages (death in childbirth, slower maturation, increased energy consumption), the better pleiotropy can explain a reserve of alleles with positive effects on IQ that would *collectively* become more common under increased pressure for IQ (or reduced pressure for fast maturation/low energy consumption, etc).
I feel obliged to point out that a steady increase in brain size along the hominid lineage was accompanied by extremely little sign of any cultural advancement for a good couple of million years. If those bigger brains were making them smarter it sure didn't show in anything they left behind. And oddly enough, the cultural Great Leap Forward starts at roughly the same period when cranial capacity stops increasing and then starts to shrink.
Matt,
Men and women have about the same mean IQ, despite men having significantly larger brains, and people like to trot out this fact in response to the brain size-race-IQ literature. However, they also have quite different strengths and weaknesses (e.g. verbal versus visuospatial, multitasking), and the brain regions important for performance on IQ tests are radically different between men and women:
http://dx.doi.org/10.1016%2Fj.neuroimage.2004.11.019
It's plausible that past hominids were structurally optimized more for visual acuity or some other neurally demanding task rather than general intelligence. But this is a tough argument to make about differences between continental races.
If continental races have had time and evolutionary pressure (for some mix of specializations) enough to evolve differences in brain structure so that their genetic potential for IQ is the same despite the differences in size, then group IQ differences could have emerged just as easily, contra Michael's argument.
"But this is a tough argument to make about differences between continental races."
To clarify, it's tough to make in support of Michael's thesis. It does look like there are such differences between groups, e.g. Ashkenazi favor verbal over visuospatial, while Chinese samples show the reverse pattern.