Cosma on IQ & heritability

Share on FacebookShare on Google+Email this to someoneTweet about this on Twitter

Cosma Shalizi has put up a gigantic post on IQ & heritability; he originally titled it “Duet for Leo and Razib,” implying that I, and the audience here @ Gene Expression, are the targets of his eloquence (at least in part). Now, I have to admit something, I’m not really interested in psychometrics that much anymore. It has been a while since I have been, stupid people are obviously stupid and I am not interesting in debating that fact. I take my own opinions in this area as background assumptions, so I’m not going to respond to Cosma. In fact, I won’t read the post right now, there’s some interesting stuff on HLA & heterozygosity that I want to check out! But, I do invite readers to digest what Cosma is saying, because I guarantee you that you’ll see it replicated by lesser minds elsewhere.

Labels: ,


  1. let’s start with this… 
    a small excerpt from a large text 
    Bouchard and McGue (2003) 
    An informative place to begin the discussion of studies of genetic influence on g is a recent reanalysis, by Devlin et al. (1997a), of a slight update of kinship correlations for IQ originally summarized and published by Bouchard and McGue (1981). Devlin has written critically of the behavioral genetic literature (Devlin et al., 1995, 1997b), so that we may expect him to report a conservative estimate of genetic influence. The most important findings in their article are that: the broad heritability of IQ is about 50% (additive variance = .34, nonadditive genetic variance = .15); twin maternal (i.e., in utero) effects account for 20%, and sibling maternal effects account for 5% of IQ variance; and shared environmental factors account for 17% of IQ variance. As McGue (1997) notes in an accompanying commentary, the result of this article is to center the debate on whether IQ is 50 or 70% heritable. This is a remarkable shift from the previous view, asserted by numerous critics, that the heritability of IQ is near zero. 
    While we regularly use modeling in our own work, we would like to repeat the caveat we placed at the end of our original presentation. ?Although the data clearly suggest the operation of environmental effects, we found no evidence for two factors sometimes thought to be important?sex-role effects and maternal effects. That the data support the inference of partial genetic determination for IQ is indisputable: that they are informative about the precise strength of this effect is dubious. Certainly the large amount of unexplained variability within degrees of relationship, while not precluding attempts to model the data, suggests that such models should be interpreted cautiously? (Bouchard and McGue, 1981, p 1058). 
    Given the need for cautious interpretation, it is especially noteworthy that the Devlin et al. (1997a) estimate of 49% for the heritability of IQ is in close agreement with the estimate of 51% reported by Chipuer et al. (1990) and the estimates of 47 and 58% reported by Loehlin (1989), who fit different although clearly converging models to the IQ correlations. Where the different analyses disagree is in terms of apportioning environmental, and not genetic, effects. Devlin et al. (1997a) conclude that the prenatal envi- ronment exerts a significant influence on IQ, increasing the IQ similarity of twins over other relative pairings. In contrast, Chipuer et al. and Loehlin conclude that the postnatal rather than the prenatal environment is most important. 
    The Devlin et al. (1997a) conclusion that the prenatal environment contributes to twin IQ similarity is especially remarkable given the existence of an extensive empirical literature on prenatal effects. Price (1950), in a comprehensive review published over 50 years ago, argued that almost all MZ twin prenatal effects produced differences rather than similarities. As of 1950 the literature on the topic was so large that the entire bibliography was not published. It was finally published in 1978 with an additional 260 references. At that time Price reiterated his earlier conclusion (Price, 1978). Research subsequent to the 1978 review largely reinforces Price?s hypothesis (Bryan, 1993; Macdonald et al., 1993; Hall and Lopez-Rangel, 1996; see also Martin et al., 1997, box 2; Machin, 1996). 
    Consideration of features of kinship similarity for IQ not incorporated into the analyses reported by Devlin et al. and the other modelers can help to further elucidate the nature of environmental influences on IQ. In particular, kinship correlations for IQ vary with age and failure to take this into account may have resulted in an overestimate of maternal environmental effects. For example, the kinship that provides the most direct test for postnatal environmental effects is the correlation between nonbiologically related, reared-together (i.e., adoptive) siblings (unrelated together or URTs). Devlin et al. did not include this kinship in the analysis they report ?because the observed correlations are extremely variable? (p. 469). In fact, and as shown in Figure 3, a major contributor to the heterogeneity in the adoptive sibling correlation is the age of the sample. The childhood data are from Burks (1928), Freeman et al. (1928), Leahy (1935), Skodak (1950), Scarr and Weinberg (1977), and Horn et al. (1979). The adult data are from Scarr and Weinberg (1978), Teasdale and Owen (1984), Scarr et al. (1993), Loehlin et al. (1997), and Segal (2000). The adoptive siblings assessed in childhood or adolescence, when they were presumably still living together, had an average IQ correlation of .26, suggesting that common rearing accounts for 26% of IQ variance. 
    The adoptive sibling pairs assessed in adulthood, however, had an average IQ correlation of only .04, suggesting that common rearing effects do not endure once the siblings no longer live together. Importantly, failure to observe significant IQ similarity in adult adoptive sibling pairs does not appear to be a consequence of biased sample selection. Teasdale and Owen (1984) reported an IQ correlation of .02 for 24 pairs of adopted, adult brothers obtained through Danish conscription board files. Because evaluation for conscription is mandatory for Danish males (regardless of medical status), and because the researchers had access to the complete Danish adoption register, this sample can be considered one of the most, if not the most, representative adoption study in the literature. While the representativeness of the sample does not solve the problem of restriction of range, it certainly makes it less salient. 
    Twin studies also suggest that genetic and environmental contributions to IQ vary with age. Wilson (1978) was one of the first to explore changes in kinship correlations for IQ in a longitudinal study, and his findings are shown in Figure 4. Prior to age 2, the phenotypic assessments used in this study are best characterized as indicies of mental development, and not IQ. The content of these mental development assessments is quite different from the, primarily verbal, content of the IQ tests used in the later years. In any case, if we use the Falconer formula 2(rmz rdz) as an estimate of genetic influence we see that in the early months there is minimal genetic influence but that by the age of 1 genetic factors begin to express themselves and they get much larger from 4 years of age and on. The same influences are expressing themselves in the sib-twin and midparent-offspring correlations. These longitudinal data thus suggest that with age, genetic factors increase while environmental factors decrease in importance. 
    Building on the work by Wilson, McGue et al. (1993) plotted twin IQ correlations by age. The IQ variance estimates derived from comparing the agespecific MZ and DZ correlations are shown in Figure 5. 
    Again we see the growing expression of genetic influence and decreasing manifestation of shared environmental influence. Not shown in Figure 5 is the extreme paucity of adult twins in studies of IQ?the younger twins swamp the data base. It is far easier to recruit twins in school settings than it is to recruit adult twins and bring them to a laboratory. Nevertheless, these data are highly consistent with the URT data and also suggest that heritability is contingent on age. 
    Recent longitudinal family and adoption data from the Texas Adoption Project (TAP) and Colorado Adoption Project (CAP) confirm these findings. In the TAP, researchers reported that estimates of IQ heritability increased (from .38 to .78) while estimates of shared environmental influence decreased (from .19 to .00) as the adopted children in the families aged from adolescence to young adulthood (Loehlin et al., 1997). A notable feature of TAP is that test reliability was incorporated into the model so that parameter estimates refer to true score rather than observed score variance. In CAP (Plomin et al., 1997), parent-offspring IQ correlations (weighted average for mothers and fathers) for adoptive and control (matched biological) families were assessed at 1, 2, 3, 4, 7, 12, and 16 years of age. These findings are shown in Figure 6. The correlations are modest and in about the same range for both types of families until about age 5, after which they diverge dramatically, with the adoptive family correlations reaching an asymptote of zero at age 12. Model fitting to the data yielded a heritability estimate of .56, an environmental transmission value of .01, an assortative mating value of .21, and a genotype-environment correlation of .01. A very similar trend, of adopted children becoming more similar to their biological than their adoptive parents over time, was reported by Honzik (1957). 
    Boomsma et al. (1999) have recently published estimates of heritability and shared environmental influence for IQ by age (5, 7, 10, 16, 18, 27 years of age) from a sample of Dutch twins. To these estimates we add data extending the Dutch sample to age 50 that were kindly provided to us by Prof. Boomsma (see also Posthuma et al., 2002a, Figure 12.1). The results are shown in Figure 7. 
    Interestingly, the heritability of general cognitive ability may decline in late life. McClearn et al. (1997) reported estimates of heritability and shared environmental influence in a sample of 117 twins age 80 years or older. For the first principal component of the seven cognitive tests, an index of g, heritability was estimated at .62 (95% CI, .29?.73), and shared environment was estimated at .11 (95% CI, .00?.47). If a short form of the Weschler Adult Intelligence Scale was used to estimate g, heritability was estimated at .55 (95% CI, .19?.76), and shared environment was estimated at .20 (95% CI, .00?.47). The influence of shared environment could have been dropped from the model in both instances as indicated by the confidence intervals. McGue and Christensen (2001) recently replicated McClearn et al.?s findings by reporting a heritability estimate of .54 (95% CI, .27?.63) for a general cognitive ability measure, in a sample of Danish twins 75 years and older. These heritability estimates are a bit lower than in younger adult data (Plomin et al., 1994), and suggest that heritability decreases in older cohorts. This conclusion is also supported by longitudinal studies of older twins (Finkel et al., 1995, 1998). 
    In summary, twin, adoption, and longitudinal family studies of IQ all converge on the conclusion that genetic factors increase while shared environmental factors decrease in importance with age, at least until middle age. Summary estimates of heritability from Devlin et al. (1997a), Chipuer et al. (1990), and Loehlin (1989) all fail to take these age effects into account.

  2. Ok, I admit I just jumped down to the end of Cosma’s post and saw where he asks an archtypical heredeterian: why it is so important to you that IQ be heritable and unchangeable?[question mark added] 
    Yes, that’s it. Speaking for myself, I know that I do so badly want massive numbers of people to be born doomed to economically unproductive lives; to be forever dependent on or resentful of those fortunate enough to be born with more abstract-thinking minds. I’d personally hate it if we could just throw money at early childhood education and housing improvements to eliminate inner-city strife and widespread poverty. I wouldn’t be happy unless an entire continent were mired in a moribund economic state, unable to contribute significantly to the world’s economy. And of course I just love to read year-in and year-out about the ‘achievement gap’ and how we must make NCLB even stronger to improve our ‘failing schools’. 
    Does he actually believe that people think intelligence is a real and hereditable phenomenon because it makes them feel better to put others down, or whatever satisfaction that’s supposed to provide? It doesn’t occur to him that maybe it just seems to make sense and explains most of what we see around us?

  3. It suddenly strikes me how very hard it must be for many laymen to cope with scientific writing. Cosma refers to very strong assumptions – which probably implies quite the wrong meaning to a layman. Yo above quotes Bouchard and McGue who refer to a “conservative estimate of genetic influence” – a description that is ambiguous even to a scientist, I’d say, and probably pretty opaque to most laymen.

  4. Ziel: This can be a political issue for either side. Socialists and other egalitarians don’t like the combination of heritable g and IQ as the major determinant of life chances. Other political tendencies like it fine, since it shows that the present class structure is natural and good, and makes it seem likely that any deliberate attempts to achieve equality will accomplish nothing and do more harm than good. 
    People here at GNXP often speak as though only the other side has a political angle, but plenty of people here do too.

  5. John,  
    What you say is true. But overall, the heritability of IQ coupled with the current global dysgenic breeding structure is leading humanity down a foreboding path. I am more than happy to admit to a strain of knee-jerk conservativism, but if there was good evidence out there that IQ is *not* highly heritable, the part of me that prefers to imagine a future world not mired in desperation and poverty would certainly trump the part of me that is resistant to government interference, and I would jump aboard the big government foreign/domestic aid bandwagon with all the other starry-eyed egalitarians.  
    In other words, I think someone would have to be really, really attached to the prevailing social order, to the point of having some pretty screwed up priorities, to root against the possibility that early state-sponsored intervention could boost IQ scores and life prospects for thsoe born into poverty. (Not that those people don’t exist.)

  6. Yo: 
    This is a remarkable shift from the previous view, asserted by numerous critics, that the heritability of IQ is near zero. 
    Who said this? When? In what context? 
    Does he actually believe that people think intelligence is a real and hereditable phenomenon because it makes them feel better to put others down, or whatever satisfaction that’s supposed to provide? It doesn’t occur to him that maybe it just seems to make sense and explains most of what we see around us? 
    Well, people thought much the same things about the wave of immigrants in the nineteenth century: That they were obviously inferior and that not too much effort should be wasted on trying to raise them up. What could possibly have motivated these people? It seemed like common sense to them, didn’t it? But it wasn’t. 
    It might well be so with you: What seems to you to be common sense is actually prejudice. Prejudice almost always seems to be common sense to he who holds it. That, I believe, was one of the fundamental observations of the enlightenment. 
    Also, Yo’s material, while interesting, doesn’t seem to answer Cosma’s bigger points–he seems more to me to be questioning whether “heritability” calculations are really terribly significant things, rather than disputing  
    what those calculations ought to be.

    Who said this? When? In what context? 
    you know someone named leon kamin right? if you don’t, you aren’t worth talking to.

  8. I can’t speak for Bouchard and McGue but Kamin (1974) made the claim and was favorably quoted ad nauseum after that.

  9. I think that the range of general opinion has narrowed. One guy framed it as 30 / 70 vs. 70 / 30, and 60/40 vs. 40/60 would probably capture most people. 
    This whole argument is happening while a variety of changes are restratifying the US economy, and this includes reduced educational opportunities as financing education becomes more difficult. There is a very strong political tendency that believes that stratification is good and natural and wants to preserve and increase it. Most GNXPers are meritocrats, which is quite a different thing, but the dominant right faction wants to reduce class mobility and make politics more elitist. (The Democrats are not at all exempt.)

  10. john, remember that if a meritocracy is operative then genetic stratification is pretty much inevitable via assortative mating. i described that here. it might not be a good thing, but it might be natural given stable conditions and a pretty good correlation between talent and outcome. various factors can slow it down, but unless there are big shocks toward the system or changes in the incentives (so that the “talent” gets redefined in a way that hasn’t sorted itself out genetically) that is the direction a society which rewards talent is going to shift toward. the highest mobility circumstance would be on where we are transitioning from a class society where birth matters to one where talent is prioritized. the post world war ii era is a classic case. the ivy league schools dropped their jew-quotas and a massive sea change occurred over a few years. now in 2007 people are worrying that the social mobility has decreased since the 1960s in terms of who goes to the ivy league…but the fact is that the children of lower middle class urban jews are the affluent professionals. and some of their talents are heritable. 
    in any case, it’s a complex issue. there’s a lot of prejudice and bias in this area, and the science itself is statistical so it lends itself toward confusion and misunderstanding. that being said, from a libertarian perspective a strongly hereditarian model is probably pretty not a good thing, since it makes it seem as if one’s success is unearned. ergo, HBD has made me less libertarian and more sympathetic toward arguments from the left.

  11. John – the dominant right faction wants to reduce class mobility and make politics more elitist. 
    I would agree that the wealthy today are quite aggressive and determined to retain their wealth and power and show no inclination to share their gains – ill gotten or otherwise. But not sure what the “dominant right faction” is. It seems the wealthy – corporate power brokers, say – are more inclined to support the opposite of the “gnxp argument” and support affirmative action, increasing spending on urban schools over middle-class public schools, unfettered immigration, etc. as a tactic to keep the broad white middle class – who represent their only real threat – in check. So I’m wondering who you’re thinking of as this dominant right faction.

  12. Genetic stratification is pretty much inevitable via assortative mating 
    Conceding that just for the sake of argument, there are also other independent forces leading to stratification which have traditionally been very powerful, and are today. You don’t have to believe the worst claims about President Bush to know that he didn’t get to where he is by talent. It’s by birth, and birth is somewhat a proxy for loyalty. (Rice and Rove and Gonzales are not Bushes, but they’re loyal retainers. Rove is a skilled operative, Rice is somewhat talented, and as far as I can tell, Gonzales is not talented.) 
    From what I’ve seen, it takes 3 or 4 generations to slip from the upper class (say, $100,000+ / year, maybe $300,000+) to the ordinary working class, and in many cases the successful parent of a lame kid can rescue the grandchildren. Business capital, money for education, or even just a house down payment can do that. 
    For a lot of reasons (mostly financing of school, especially granted that BS or BA is nothing these days) the upward track is harder than it was, and entry-level business is also harder.  
    Immigrant successes work against my narrative, but a lot of them were middle class when they came here (even if they dropped down a notch) and plenty of immigrants are unsuccessful.

  13. Ziel: Whoever it is that’s been backing Bush. Oil, finance, resource industries, media, maybe retailing.  
    A lot of the biggies play both sides, but until last year the Republicans had a very big advantage. 
    Affirmative action and differential spending on urban schools are minor fig leaf issues. Immigration works against upward mobility of the people already here. The big changes are in taxes, finance, higher education funding, corporate governance, etc.

  14. The easiest way to de-stratify (?) the class system is to reverse the trend of assortative mating for IQ / education / etc. 
    Now, why would realistically motivate smart, educated, successful guys to marry down? Yep, not hard to guess. Technically smart women could marry male underwear models or other types with lowish IQ but some other good trait, but I don’t think elite women are interested in that. 
    Somehow I doubt the egalitarians will push for this, even though it’s the most feasible. 
    Also, this would only work in the short-term (like hundreds of years). After awhile, you’d pool the pretty and smarty genes into the same families, and the plain and dull genes into other families. In fact, that’s just what modern South Asia looks like! And it likely resulted from the same dynamics: lower-status males couldn’t break out of their caste, but pretty girls could marry up.

  15. Er, what would realistically motivate…

  16. One guy framed it as 30 / 70 vs. 70 / 30, and 60/40 vs. 40/60 would probably capture most people. 
    The increase in h2 and decrease in c2 with age, combined the tendency to oversample younger twins, confounds any such summary statistic. The 30% values are coming from young children and the 70% values are coming from adults.

  17. I am new to this web site (found the link through the Intersection by going through the other Gene Expressions’ link). Can some explain what h2 and c2 are? This is all Greek to me. Alright, most of the site is Greek to me. 
    Thank you.

  18. h2 = h2. heritability in the narrow sense. basically measure the proportion of phenotypic variation which is due to additive genetic variation. intuitively it measures the extent to which you can predict traits of a set of offspring from a set of parents.

  19. OK. I guess I’ll start with the Wikipedia entry :-)

  20. Shalizi says: Does a trait’s heritability tells us anything about its malleability, about how easy it is to change the trait with environmental manipulations? The answer is “no, of course not”… 
    By itself, no – but we have more information than that. Adoption studies can’t directly tell us the specific causes of trait differences, but they can rule out certain possibilities. In particular, shared environmental causes (i.e. environments that make people in the same family more alike in IQ or whatever) seem to disappear in adulthood. Now, Shalizi seems to accept this – he talks about “shared environments other than the family” – which is certainly possible, but then what could they be? Wealth, social class, SES etc are family-level variables, so if not that, then what?  
    Randomized experiments, natural experiments and the Flynn Effect all show what competent regressions also suggest, namely that IQ is, indeed, responsive to purely environmental interventions. 
    Randomized experiments – really? Shalizi mentions a study published in 1972 that only looked at young children with low-IQ mothers – been a long time since then. There would obviously be tremendous interest in a smoking-gun, randomized treatment that permanently raised IQ, and lots of stuff has been tried (Head Start, nutritional supplements, etc) – so where are the things that work?

  21. On meritocracy, I think what I said in the linked thread on that topic still applies: 
    Wouldn’t it be a natural impulse for any set of powerful, high status people to ensure that their own children attained to the same status and power regardless of their innate talents? 
    And wouldn’t they tend to succeed more often in passing along status and power to their children if the rewards for power and status were themselves heritable and to some extend reverse exchangable–like property is. 
    So if intelligence and industriousness are both, say, 65% heritable, and high status individuals are, say, 85% successful in conferring high status to their children regardless of IQ or industriousness. All of these numbers seem plausible to me and they would seem to argue that meritocracy is actually impossible because of natural parental impulse to favor one’s own children (as famously worried about by Plato in his Republic). 
    In which case we would expect to see no consequences of the presence of meritocracy, because we would expect no such thing to exist. 
    Parental attachment sytematically mitagates against meritocracy, particualrly in a society which allows large accumulations of wealth and a number of good methods to deliver wealth and power to offspring with no particualr merit. I think it is indisputable that it is more likely that the offspring of a rich powerful person will inherit wealth and power than they will inherit what we might call merit. 
    Also, Leo Kamin–one authority endlessly referred back to by lay commentators–doesn’t justify the implication that the field has moved a long way from the proposition of zero inheritance. Zero inheritance was always a marginal, tactical position. My read on Kamin is that he proposes zero heritability as the proper working hypothesis that others in the field should have to refute, rather than as an honest-to-goodness interpretation of the (according to him, quite insufficient) data. And I think most people with some expertise in the field recognized that. 
    I think at this point, perhaps, we can dispense with the rhetorical flourishes–either asserting zero heritability or using it as a bugaboo.

  22. You know, with Dianetics, one can increase one’s IQ by 15 points.* 
    And if you believe that, you need it. 

  23. I agree with Razib on libertarianism not looking as good under the light of hereditarianism. Libertarians often talk about how we are so much better off today than in the past because we have more technology and office jobs rather than farm or factory work and with further automation we’ll just continue to get better ones, but if it turns out there’s a limit for a great many people the only option may be the largess of the state.

  24. TGGP: 
    The surprising thing is that the one widely-studied ability–IQ–has held about the same relationship with wages ever since the 1960′s. In the U.S. at least, the market abhors a vacuum, and so the market finds a way to use low-IQ workers productively. As a result, the link between IQ and wages just isn’t that strong in the U.S.  
    And the IQ-wage relationship has been pretty stable for decades. The link between IQ and earnings just hasn’t much budged over time. According to Bowles, Gintis and Osborne, 1 IQ point has been worth about 0.5% more wages from the early 60′s through the late 90′s.  
    Figure 6 of this paper tells the story: 
    Of course, one can raise a ton of objections: What if IQ matters more (or less) in the future? What if Bowles et al. didn’t measure something right? And don’t lots of non-IQ traits matter, too?—-Traits like patience and willingness to work long hours that Greg Clark emphasizes in Farewell to Alms
    Yes, yes, lots of other things matter, non-experimental statistics is wildly inaccurate, and parameters are unstable–but the overriding story is that the rise of information technology hasn’t led to the mass redundancy of low-IQ labor. (Recall that the late 1990′s, the age of the dot-com boom, was Mickey Kaus’s golden age for low-wage labor).  
    So there’s reason to be cautiously optimistic about the ability of modern market-based economies to find useful, above-minimum-wage employment for most of the folks who find themselves on the bottom half of the bell curve…..

  25. yo: 
    Thanks for that link. Shalizi is apparently operating in “any weapon to hand” mode, so it’s good to be reminded about the facts that don’t help his “heritability is low” story–such as the rise in heritability as people age….. 
    It would seem that his entire h^2 rant could be just as easily written on the topic of height—it seems that all of his (broadly legitimate) h^2 gripes– “heteroskedasticity, gene-environment interactions, gene-environment covariance, the existence of shared environment beyond the family, and the possibility that the samples being used are not representative of the broader population”—would apply equally well to any real-world studies about height. But he’s pretty open to the idea that height differences in modern societies are partially driven by genes. I wonder why he has a such different priors for IQ and height…. 
    And it’s noteworthy that after pages of relentlessly attacking current (imperfect) h^2 estimates, Shalizi ultimately concludes that some cognitive ability differences in normal populations are probably genetically driven. The “strength of analogy to other areas of biology” just can’t be ignored…..

  26. Rob said: 
    You know, with Dianetics, one can increase one’s IQ by 15 points.* 
…eer/ income.html 
    And if you believe that, you need it.
    Speaking of Libertarianism, Alissa Rosenbaum (“Ayn Rand”) claimed that IQ can be boosted by as much as 150 points by means of rational thinking. (See her posthumous anthology “Ayn Rand Answers.”) Whee. 
    Maybe she’s right – if the rational thinking is done by scientists who invent a brain transplant.