The Unz Review - Mobile
A Collection of Interesting, Important, and Controversial Perspectives Largely Excluded from the American Mainstream Media
Email This Page to Someone

 Remember My Information

Authors Filter?
Razib Khan
Nothing found
 TeasersGene Expression Blog

Bookmark Toggle AllToCAdd to LibraryRemove from Library • BShow CommentNext New CommentNext New Reply
🔊 Listen RSS

Sir Francis Galton

Modern evolutionary genetics owes its origins to a series of intellectual debates around the turn of the 20th century. Much of this is outlined in Will Provines’ The Origins of Theoretical Population Genetics, though a biography of Francis Galton will do just as well. In short what happened is that during this period there were conflicts between the heirs of Charles Darwin as to the nature of inheritance (an issue Darwin left muddled from what I can tell). On the one side you had a young coterie around William Bateson, the champion of Gregor Mendel’s ideas about discrete and particulate inheritance via the abstraction of genes. Arrayed against them were the acolytes of Charles Darwin’s cousin Francis Galton, led by the mathematician Karl Pearson, and the biologist Walter Weldon. This school of “biometricians” focused on continuous characteristics and Darwinian gradualism, and are arguably the forerunners of quantitative genetics. There is some irony in their espousal of a “Galtonian” view, because Galton was himself not without sympathy for a discrete model of inheritance!

William Bateson

In the end science and truth won out. Young scholars trained in the biometric tradition repeatedly defected to the Mendelian camp (e.g. Charles Davenport). Eventually, R. A. Fisher, one of the founders of modern statistics and evolutionary biology, merged both traditions in his seminal paper The Correlation between Relatives on the Supposition of Mendelian Inheritance. The intuition for why Mendelism does not undermine classical Darwinian theory is simple (granted, some of the original Mendelians did seem to believe that it was a violation!). Many discrete genes of moderate to small effect upon a trait can produce a continuous distribution via the central limit theorem. In fact classical genetic methods often had difficulty perceiving traits with more than half dozen significant loci as anything but quantitative and continuous (consider pigmentation, which we know through genomic methods to vary across populations mostly due to half a dozen segregating genes or so).

Notice here I have not said a word about DNA. That is because 40 years before the understanding that DNA was the substrate of genetic inheritance scientists had a good grasp of the nature of inheritance through Mendelian processes. The gene is fundamentally an abstract unit, an analytic element subject to manipulation which allows us to intelligibly trace and predict patterns of variation across the generations. It so happens that the gene is instantiated in a material sense through sequences of the biomolecule DNA. This is very important. Because we know the material basis of modern genetics it is a much more fundamental science than economics (economics remains mired in its “biometric age!”).

The “post-genomic era” is predicated on industrial scale analysis of the material basis of genetics in the form of DNA sequence and structure. But we shouldn’t confuse DNA, concrete bases, with classical Mendelism. A focus on the material and concrete is not limited to genetics. In the mid-2000s there was a fad for cognitive neuroscience fMRI studies, which were perceived to be more scientific and convincing than classical cognitive scientific understandings of “how the mind works.” In the wake of the recession of fMRI “science” due to serious methodological problems we’re left to fall back on less sexy psychological abstractions, which may not be as simply reduced to material comprehension, but which have the redeeming quality of being informative nonetheless.

This brings me to the recent paper on SNPs associated with education in a massive cohort, GWAS of 126,559 Individuals Identifies Genetic Variants Associated with Educational Attainment. You should also read the accompanying FAQ. The bottom line is that the authors have convincingly identified three SNPs to explain 0.02% of the variation in educational attainment across their massive data set. Pooling all of the SNPs with some association they get ~2% of the variation explained. This is not particularly surprising. A few years back one of the authors on this paper wrote Most Reported Genetic Associations with General Intelligence Are Probably False Positives. Those with longer memories in human genetics warned me of this issue in the early 2000s. More statistically savvy friends began to warn me in 2007. At that point I began to caution people who assumed that genomics would reveal the variants which are responsible for normal variation on intelligence, because it seemed likely that we might have to wait a lot longer than I had anticipated. As suggested in the paper above previous work strongly implied that the genetic architecture of intelligence is one where the variation on the trait in the normal range is controlled by innumerable alleles of small effect segregating in the population. Otherwise classical genetic techniques may have been able to detect the number of loci with more surety. If you read Genetics of Human Populations you will note that using classical crossing techniques and pedigrees geneticists did in fact converge upon approximately the right number of loci segregating to explain the variation between European and African pigmentation 60 years ago!

Some of my friends have been arguing that the small effect sizes here validate the position that intelligence variation is mostly a function of environment. This is a complicated issue, and first I want to constrain the discussion to developed Western nations. It is an ironic aspect that arguably intelligence is most heritable among the most privileged. By heritable I mean the component of variation of the trait controlled by genes. When you remove environmental variation (i.e. deprivation) you are left with genetic variation. Within families there is a great deal of I.Q. difference across siblings. The correlation is about 0.5. Not bad, but not that high. Of course some of you may think that I’m going to talk about twin studies now. Not at all! Though contrary to what science journalists who seem to enjoy engaging in malpractice like Brian Palmer of Slate seem to think classical techniques have been to a great extent validated by genomics, it is by looking at unrelated individuals that some of the most persuasive evidence for the heritability of intelligence has been established. It is no coincidence that one of the major authors of the above study also is an author on the previous link. There is no contradiction in acknowledging difficulties of assessing the concrete material loci of a trait’s variation even if one can confidently infer that association. There was genetics before DNA. And there is heritability even without specific SNPs.

Additionally, I want to add one caveat into the “environmental” component of variation. For technical reasons this environmental component may actually include relatively fixed biological variables. Gene-gene interactions, or developmental stochasticity come to mind. Though these are difficult or impossible to predict from parent to offspring correlations they are not as simple as removing lead from the environment of deprived children. My own suspicion is that the large variation in intelligence across full siblings tell us a lot about the difficult to control and channel nature of “environmental” variation.

Finally, I want to point out that even small effect loci are not trivial. The authors mention this in their FAQ, but I want to be more clear, Small genetic effects do not preclude drug development:

Consider a trait like, say, cholesterol levels. Massive genome-wide association studies have been performed on this trait, identifying a large number of loci of small effect. One of these loci is HMGCR, coding for HMG-CoA reductase, an important molecule in cholesterol synthesis. The allele identified increases cholesterol levels by 0.1 standard deviations, meaning a genetic test would have essentially no ability to predict cholesterol levels. By the logic of the Newsweek piece, any drug targeted at HMGCR would have no chance of becoming a blockbuster.

Any doctor knows where I’m going with this: one of the best-selling groups of drugs in the world currently are statins, which inhibit the activity of (the gene product of) HMGCR. Of course, statins have already been invented, so this is something of a cherry-picked example, but my guess is that there are tens of additional examples like this waiting to be discovered in the wealth of genome-wide association study data. Figuring out which GWAS hits are promising drug targets will take time, effort, and a good deal of luck; in my opinion, this is the major lesson from Decode (which is not all that surprising a lesson)–drug development is really hard

Addendum: Most of my friends, who have undergraduate backgrounds in biology, and have taken at some quantitative genetics, seem to guess the heritability of I.Q. to be 0.0 to 0.20. This is just way too low. But is it even important to know this? I happen to think an accurate picture of genetic inheritance is probably useful when assessing prospects of mates….

Citation: Rietveld, Cornelius A., et al. “GWAS of 126,559 Individuals Identifies Genetic Variants Associated with Educational Attainment.” Science (New York, NY) (2013).

(Republished from Discover/GNXP by permission of author or representative)
🔊 Listen RSS

Prompted by my post Ta-Nehisi Coates reached out to Neil Risch for clarification on the nature (or lack thereof) of human races. All for the good. The interview is wide ranging, and I recommend you check it out. Read the comments too! Very enlightening (take that however you want).

When it comes to this debate I have focused on the issue of population substructure, or race. The reason is simple. Due to Lewontin’s Fallacy it is widely understood among the “well informed general public” that “biology has disproved race.” Actually, this is a disputable assertion. For a non-crank evolutionary biologist who is willing to defend the race concept for humans, see Jerry Coyne. When you move away from the term “race,” then you obtain even more support from biologists for the proposition that population structure matters. For example, a paper in PLoS GENETICS which came out last week: Analysis of the Genetic Basis of Disease in the Context of Worldwide Human Relationships and Migration. In other words, it is useful to understand the genetic relationships of populations, and individual population identity, because traits correlate with population history. Barring total omniscience population history will always probably matter to some extent, because population history influences suites of traits. If nothing in evolutionary biology makes sense except in light of phylogeny, much of human biology is illuminated by phylogeny.

But that doesn’t speak to the real third rail, intelligence. Very few people are offended by the idea of the correlation between lactase persistence and particular populations. Neil Risch says in the interview with Coates:

One last question. Your paper on assessing genetic contributions to phenotype, seemed skeptical that we would ever tease out a group-wide genetic component when looking at things like cognitive skills or personality disposition. Am I reading that right? Are “intelligence” and “disposition” just too complicated?

Joanna Mountain and I tried to explain this in our Nature Genetics paper on group differences. It is very challenging to assign causes to group differences. As far as genetics goes, if you have identified a particular gene which clearly influences a trait, and the frequency of that gene differs between populations, that would be pretty good evidence. But traits like “intelligence” or other behaviors (at least in the normal range), to the extent they are genetic, are “polygenic.” That means no single genes have large effects — there are many genes involved, each with a very small effect. Such gene effects are difficult if not impossible to find. The problem in assessing group differences is the confounding between genetic and social/cultural factors. If you had individuals who are genetically one thing but socially another, you might be able to tease it apart, but that is generally not the case.

In our paper, we tried to show that a trait can appear to have high “genetic heritability” in any particular population, but the explanation for a group difference for that trait could be either entirely genetic or entirely environmental or some combination in between.

So, in my view, at this point, any comment about the etiology of group differences, for “intelligence” or anything else, in the absence of specific identified genes (or environmental factors, for that matter), is speculation.

In response to this commenter Biologist states (note, I know who this is, and they are a biologist!):

Risch writes: “…the explanation for a group difference for that trait could be either entirely genetic or entirely environmental or some combination in between. … So, in my view, at this point, any comment about the etiology of group differences, for “intelligence” or anything else, in the absence of specific identified genes (or environmental factors, for that matter), is speculation.”

This is essentially correct. The quality of available evidence on which to estimate the contribution of genetic versus environmental factors to group differences in cognitive ability scores is quite poor by biomedical research standards — maybe more in line with standards for social science (I’m only half joking).

In light of that, one is forced fall back on to one’s priors. Without trying to speak for Risch, it is generally considered appropriate to adopt a uniform prior in the absence of other evidence. Under a uniform prior, “…the explanation for a group difference for that trait could be either entirely genetic or entirely environmental or some combination in between.” Maybe Risch would propose a different prior.

In fact, the uniform prior says that there’s a 25% chance that the explanation is 0% to 25% genetic, a 50% change that the explanation is 25% to 75% genetic, and a 25% chance that the explanation is 75% to 100% genetic. Obviously many people who write about this topic do not adopt a uniform prior. [my emphasis -Razib]

As Risch observes above intelligence is highly polygenic. There’s a fair amount of genomic evidence for this now. In other words the likelihood is not high that we will be able to account for the differential distribution of IQ between any two populations by differences in allele frequencies. Even if we do find the allelic differences, they’ll account for far too little of the variation in the trait. But there is another way we can get at the issues. Others have pointed out exactly how we can get more clarity on the race and IQ question before, so I’m not being original. And since I suspect that within the next decade this sort of analysis will likely be performed at some point somewhere because the methods are so simple, I might as well be explicit about it.

Let’s focus on the black-white case in the American context. On intelligence tests the average black American scores a bit less than 1 standard deviation below the average white American. As I’ve observed before the average black American is ~20% European, but there is variation around this value. Because the admixture is relatively recent (median ~150 years before the present) there is a wide range across the population of ancestry. In fact, the admixture is recent enough that siblings may even differ in the amount of European ancestry on a genomic level. An additional issue which is of relevance is that the correlation between ancestry and physical appearance in mixed populations is modest. By this, I mean that there are many individuals who are more European in ancestry in the African American population who have darker skins and more African features than those who have less European ancestry. Obviously on average more European ancestry predicts a more European appearance, but this is true only on average. There are many exceptions to this trend.

At this point many of you should have anticipated where I’m going. If the gap between blacks and whites on psychometric tests is totally driven by genetic differences between Africans and Europeans, then the gap should be obvious between pools of individuals of varying levels of European ancestry within the African American population. It seems unlikely that it would be that simple (i.e., all driven by genes without any sensitivity to environmental inputs or context). Therefore I suspect some design where you compare siblings would be more informative.

In a model where all of the between group differences are due to environmental inputs then genomic ancestry by geography within family should add little in terms of prediction of the phenotype. More plainly, when accounting for other variables which might correlate with ancestry (e.g., skin color), how African or European a sibling is should not influence outcomes on psychometric tests when looking at large cohorts of sibling pairs if those differences track nothing more than social construction/perception of race. If on the other hand there are many alleles of small effect distributed throughout the genome correlated with geographic ancestry which affect the final phenotype then adding ancestry as an independent variable into the model should be informative. This sort of indirect inference has already been performed with a character similar in genetic architecture to intelligence: height. Researchers have found that African Pygmies with more non-Pygmy ancestry are taller.

Ultimately I say that this issue might semi-resolve, because I think a hereditarian position in terms of group differences is not going to be tenable if the correlations with ancestry do not run in the direction expected within admixed populations. This sort of model is relatively straightforward in its predictions, and appeals to parsimony. Trying to salvage it with non-additive genetic variance is going to complicate matters. In contrast, those who champion the opposite position often dispute the very characterization of intelligence as a trait in the first place, so I presume that they would still exhibit skepticism if there was a correlation between genomic ancestry and the trait.

Addendum: I want to be clear: with the widespread availability of data sets and crappy security of said data sets this analysis is probably a few SQL joins away in 10 years.

(Republished from Discover/GNXP by permission of author or representative)
• Category: Science • Tags: Genetics, I.Q., Intelligence, Psychology, Race 
🔊 Listen RSS

In light of the previous post I was curious about the literature on inbreeding depression of IQ. A literature search led me to conclude two things:

- This is not a sexy field. A lot of the results are old.

- The range in depression for first cousin marriages seems to be on the order of 2.5 to 10 IQ points. In other words ~0.15 to ~0.65 standard deviation units of decline in intelligence.

The most extreme case was this paper from 1993, Inbreeding depression and intelligence quotient among north Indian children. The authors compared the children of first cousin marriages, and non-bred in individuals, from a sample of Muslims in Uttar Pradesh of comparable socioeconomic status (though the authors note that inbreeding has a positive correlation with socioeconomic status in this community). A table with results speaks for itself:

(Republished from Discover/GNXP by permission of author or representative)
• Category: Science • Tags: I.Q., Psychology 
🔊 Listen RSS

In light of my previous posts on GRE scores and educational interests (by the way, Education Realist points out that the low GRE verbal scores are only marginally affected by international students) I was amused to see this write-up at LiveScience, Low IQ & Conservative Beliefs Linked to Prejudice. Naturally over at Jezebel there is a respectful treatment of this research. This is rather like the fact that people who would otherwise be skeptical of the predictive power of I.Q. tests become convinced of their precision of measurement when it comes to assessing whether a criminal facing the death penalty is mentally retarded or not! (also see this thread over at DailyKos). You can see some of the conservative response too.

The paper itself is Bright Minds and Dark Attitudes: Lower Cognitive Ability Predicts Greater Prejudice Through Right-Wing Ideology and Low Intergroup Contact:

Despite their important implications for interpersonal behaviors and relations, cognitive abilities have been largely ignored as explanations of prejudice. We proposed and tested mediation models in which lower cognitive ability predicts greater prejudice, an effect mediated through the endorsement of right-wing ideologies (social conservatism, right-wing authoritarianism) and low levels of contact with out-groups. In an analysis of two large-scale, nationally representative United Kingdom data sets (N = 15,874), we found that lower general intelligence (g) in childhood predicts greater racism in adulthood, and this effect was largely mediated via conservative ideology. A secondary analysis of a U.S. data set confirmed a predictive effect of poor abstract-reasoning skills on antihomosexual prejudice, a relation partially mediated by both authoritarianism and low levels of intergroup contact. All analyses controlled for education and socioeconomic status. Our results suggest that cognitive abilities play a critical, albeit underappreciated, role in prejudice. Consequently, we recommend a heightened focus on cognitive ability in research on prejudice and a better integration of cognitive ability into prejudice models.

I emphasized sections that I assume will answer some immediate questions, as not everyone has access to Psychological Science. Yes, they used different types of intelligence tests; verbal and spatial. Yes, they corrected for socioeconomic background. Their replication was in the UK and USA. Importantly, they focused on a few characteristics, attitudes toward homosexuals and race. It doesn’t seem like they explored an enormous range of opinions. And as noted in the paper they were looking at the social dimension of political ideology.

There is plenty of work on cognitive styles and political orientation. Recently it is moral foundations from Jon Haidt. Earlier you had George Lakoff’s models. Neither of these focused on general intelligence, the raw CPU power of the mind. Rather they surveyed moral intuition and personality profiles (for example, there is some evidence that those with a greater bias toward “openness” are more socially liberal).

Looking at the General Social Survey I too have found at a correlation between higher intelligence and social liberalism. On the other hand a good objection to this is that my estimator of intelligence, WORDSUM, was verbal, and liberals and conservatives may exhibit different cognitive profiles. This study takes that into account, adding spatial I.Q. tests to the mix.

It is important to emphasize that the authors do not posit an independent direct causal connection between low I.Q. and more reactionary attitudes towards race and homosexuality. Rather, they start out with a model where low cognitive ability people are drawn (or remain in) to conservative orientation, and this is further correlated with these specific racial and sexual attitudes. Like almost all psychology you can’t get the causation airtight (if you are a hardcore Humean you could probably say this for everything), but the correlation is suggestive in light of political and psychological models. The problem is the second. As Jonathan Haidth has articulated most recently most academic political scientists and psychologists have strongly social liberal views, and so they consciously or unconsciously tend to caricature and misrepresent the views of half their study population (notice that the authors assume that these socially conservative positions are ‘Dark Attitudes’; most people today would agree, but shouldn’t intellectuals avoid this sort of thing?). So though I have some confidence in the correlations, I’m a lot more skeptical of the explanatory models (though I don’t reject them out of had). There are so many models sitting around that how you chose models can be shaped by bias rather easily.

First, let’s hit the results.

The table above represents the results for the British cohorts and race, and the diagram to the left illustrates the outcome for the American sample and homosexuality. The primary point is that as per their hypothesis the effect of lower cognitive ability on prejudice toward other races and homosexuality is mediated more or less through ideology. Coarsely, stupid people aren’t racist, stupid people are more likely to be socially conservative, and more socially conservative people are more likely to be racist. How these join together though is something one can subject to more critical examination. The authors allude to this when they note that there is a finding that those who know people of other races tend to be less prejudiced, with the inference being that contact makes one less racist. But this is not an established causality. Rather, it could be that people with less prejudiced tendencies put themselves into situations where they are likely to meet other races. This tendency could be correlated with higher I.Q. through a mediation of a “cosmopolitanism index.” Who knows? There are many stories one could tell.

I do want to emphasize though that this is a coarse measure of ‘conservatism.’ In the early to mid aughts Paul Wolfowitz was a hated figure on the American political Left because of his critical role in buttressing the intellectual armamentarium favoring the invasion of Iraq. But it is well known that Wolfowitz was and is a social liberal, like a subset of neoconservatives who focus on foreign policy. On the above measure Wolfowitz, who has undergraduate degrees in mathematics and chemistry from Cornell and a graduate degree in political science from University of Chicago, would come out as a high I.Q. social liberal. Is that right? As far as it goes it is right, but on some level the results would be misleading in the more complex terrain of coalitional politics. A substantial number of Americans shake out as social conservatives and fiscal moderates/liberals. And yet this faction is totally unrepresented in modern politics. In contrast, their inverse, libertarians, do have some representation, albeit a marginalized one. Why? Because the latter position has modest high I.Q./elite support, while the former position has far less. If you changed the question to attitudes toward global free trade there would be a correlation between lower I.Q. and the ‘more liberal’ (at last in American politics) position.

This qualification also dovetails with the broader point about styles of cognitive thinking, and reliance on traditional norms as opposed to think a priori. Ironically it makes intuitive sense that higher I.Q. people would be less reliant on intuition, impulse, and collective wisdom. But there are limits to this. For example, see the reaction to the proposition of sex between consenting adults who happen to be siblings on an atheism forum (assume they use birth control). But some moral philosophers posit that this is not harmful or immoral, and should be socially accepted. It’s an interesting illustration of the boundary condition of the power of disgust and emotion, as only the hyper-rational feel comfortable even entertaining the moral legitimacy of this proposition. More relevantly, educated liberals also make use of ‘stereotypes’ constantly. It’s just that those stereotypes are of conservatives. I know this because almost all my friends are educated liberals, and they often forget that I’m a conservative. So I hear a lot about conservatives are this and that without qualification, to great merriment and laughter (also, conservatives are genuinely evil and malevolent apparently!). The tendency toward generalization doesn’t bother me in an of itself, rather, I’m focused on whether the proposition is true. But the hypocrisy gets tiresome sometimes, as people will fluidly switch from a cognitive style which accepts generalization to one which rejects it. A stereotype is often a generalization whose robustness you don’t want to accept. Negative generalities need context when they’re unpalatable, but no qualification is necessary when their truth is congenial. Sometimes this veers into moderately politically incorrect territory. I was once an observer on a conversation between liberal white academics who were mulling over the unfortunate reality that their Asian American students were far more likely to cheat to obtain better grades. I suspect that this is actually true for various reasons. But I also suspect that these academics forgot that I was privy to the conversation, and wouldn’t have aired this truth in a more racially diverse social context.

More broadly what is the takeaway from this sort of research? Should we conclude that because the more intelligent tend to be socially liberal that socially liberal propositions are true? I think one should be skeptical of this position. There are two immediate rejoinders. First, politics is a matter of values. The reliance of reason vs. emotion, individual ratiocination vs. historical or social wisdom, may vary. But that does not speak to the truth of any given value judgement, as those judgments are embedded in a system of norms, as well as individual self-interest (e.g., the higher I.Q. tendency to favorable attitudes toward free trade may have less to do with an understanding of comparative advantage, than an implicit understanding that globalization favors them as opposed to less intelligent lower classes). Second, the moral arc of history is not always unidirectional. The ‘progressive’ position is sometimes reversed. In Better for All the World there is a broad history of the rise of a consensus among economic and intellectual elites about the wisdom of coercive eugenics as an instrument of progressive social engineering in the late 19th century. Religious conservatives, whether evangelical Protestant or Roman Catholic, were two of the greatest bulwarks against this force for progress. Arguably these two elements were more efficacious in resisting the spread of eugenics legislation than the Left critics, judging by the outcomes Southern Europe and the American South, as opposed to the more ‘forward thinking’ nation-states of Northern Europe and the American North. This fact is unknown to most of my friends and acquaintances, judging by repeated assumptions that any utilization of personal genomics for eugenic purposes will occur first in politically conservative jurisdictions.

With all these qualifications, I believe this sort of research is essential and insightful. We need to understand the patterns of cognitive variation, whether it be intelligence or personality, which may result in differences of opinion. At the end of the day no opinions may change, but one may be able to construct a crisper argument when taking into account the genuine roots of one’s political opponents viewpoints, rather than your own ill-informed caricature.

Addendum: I did not address the issue of revealed vs. avowed preferences and attitudes. But I think that this difference will not change the sign of correlation. For example, for various reasons I assume that the gap between white liberals and white conservatives when it comes to race is smaller in terms of the preference revealed in their choices, rather than the survey responses they give, but I don’t think it reverses the rank order of the correlation.

Citation: Bright Minds and Dark Attitudes: Lower Cognitive Ability Predicts Greater Prejudice Through Right-Wing Ideology and Low Intergroup Contact, Psychol Sci. 2012 Jan 5.

(Republished from Discover/GNXP by permission of author or representative)
• Category: Ideology, Science • Tags: Culture, I.Q., Intelligence, Liberalism, Politics 
🔊 Listen RSS

A questioner below was curious if vocabulary test differences by ethnic and region persist across income. There’s a problem with this. First, the INCOME variable isn’t very fine-grained (there is a catchall $30,000 or greater category). Second, it doesn’t seem to control for inflation. But, there is a variable, DEGREE, which asks the highest level of education attained. I used this to create a “college” and “non-college” category (i.e., do you have a bachelor’s degree or not). Because of sample size considerations I removed some of the ethnic groups, but replicated the earlier analysis.

Below are two tables. One shows the mean vocab score for region and ethnicity (for whites) for those without college educations, and another shows those with college educations. I decided to generate a correlation over the two rows, even though it sure isn’t useful as a quantitative statistical measure because of the small number of data points. Rather, I just wanted a summary of the qualitative result. The short answer is that the average vocabulary difference seems to persist across educational levels (the exception here is the “German” ethnicity).

Mean WORDSUM Score by Ethnicity and Region
No college education




German 6.05 5.81 5.79 6.11
Eastern Europe 6.17 6.16 6.18 6.29
Scandinavian 6.35 5.97 6.23 6.35
British 6.6 6.21 6.02 6.57
Irish 6.66 5.83 5.69 6.58
Italian 6 5.85 5.8 6.18
College educated




German 8.03 7.48 7.63 7.33
Eastern Europe 7.7 7.37 7.5 8.09
Scandinavian 8.5 7.82 7.86 7.92
British 8.44 8.06 7.76 7.95
Irish 8.03 7.79 7.39 7.59
Italian 7.45 7.75 7.6 7.87
Correlation of college and non-college
German 0.08
Eastern Europe 0.92
Scandinavian 0.57
British 0.70
Irish 0.57
Italian 0.40
(Republished from Discover/GNXP by permission of author or representative)
• Category: Science • Tags: Data, Data Analysis, GSS, I.Q., Regionalism 
🔊 Listen RSS

The title says it all, and I yanked it from a paper that is now online (and free). It’s of interest because of its relevance to the future genetic understanding of complex cognitive and behavioral traits. Here’s the abstract:

General intelligence (g) and virtually all other behavioral traits are heritable. Associations between g and specific single-nucleotide polymorphisms (SNPs) in several candidate genes involved in brain function have been reported. We sought to replicate published associations between 12 specific genetic variants and g using three independent, well-characterized, longitudinal datasets of 5571, 1759, and 2441 individuals. Of 32 independent tests across all three datasets, only one was nominally significant at the p ~ .05 level. By contrast, power analyses showed that we should have expected 10–15 significant associations, given reasonable assumptions for genotype effect sizes. As positive controls, we confirmed accepted genetic associations for Alzheimer disease and body mass index, and we used SNP-based relatedness calculations to replicate estimates that about half of the variance in g is accounted for by common genetic variation among individuals. We conclude that different approaches than candidate genes are needed in the molecular genetics of psychology and social science.

My hunch is that these results will be unsatisfying to many people. The authors confirm and reassert the heritability of general intelligence, both by reiterating classical results, and utilizing novel genomic techniques. But, they also suggest that the candidate gene literature is nearly worthless because of the lack of power of most of the earlier studies. The latter is probably due to the genetic architecture of the trait. Intelligence may be determined by numerous genes of very small effect (e.g., 0.01% of the variance effected by one particular SNP), or, “rare, perhaps structural, genetic variants with modest to large effect sizes.” The former case is pretty obvious, but what about the latter? I’m mildly skeptical of this because I’m curious why modest-to-large effect variants didn’t show up in family-based studies (presumably within the family the same variants would localize to sections of the genetic map)? But I’m not fluent enough in the literature to know if there was a lot of work in this area with families previously.

Related: Here’s the first author’s article in Commentary from the late 1990s, IQ Since “The Bell Curve”.

(Republished from Discover/GNXP by permission of author or representative)
🔊 Listen RSS

I saw this link posted on twitter, IQ and Human Intelligence:

An interesting finding from genetic research, which Mackintosh mentions, only in passing, as posing a problem in the estimation of the heritability of g, is that there is greater assortative mating for g than for any other behavioral trait; that is, spouse correlations are only ∼.1 for personality and only ∼.2 for height or weight, but the correlation for assortative mating for g is ∼.4. In addition to indicating that people are able to make judgments about g in real life, this finding suggests that assortative mating may contribute to the substantial additive genetic variance for g, because positive assortative mating for a character can increase its additive genetic variance.

I’ve seen these sort of results before. The review is from 1999. In general I always wonder if quantitative values for personality are not to be trusted because of issues with the measurement of personality types. But this is clearly not an issue with height or weight. And in the case of height the overwhelming causal explanation for variation in the West is genetic variation. Overall I’m rather surprised by the rather low correlations for some of these traits, such as height and intelligence. I wonder if beauty, perhaps measured by an index of facial symmetry, might exhibit higher correlation values?

(Republished from Discover/GNXP by permission of author or representative)
• Category: Science • Tags: I.Q., Marriage, Psychology, Sociology 
🔊 Listen RSS

A new paper in Molecular Psychiatry has been reported on extensively in the media, and readers have mentioned it several times in the comments. I read it. It’s titled Genome-wide association studies establish that human intelligence is highly heritable and polygenic. But the fact is that I read this paper last year. Back then it was titled Common SNPs explain a large proportion of the heritability for human height. I kid, but you get the picture. The new paper establishes for intelligence what we already suspected: most of the genetic variation in this heritable trait is accounted for by numerous genes of small effect. You inherit variants of these numerous genes from your two parents, and your own trait value is to a large extent a combination of the parental values. The issue is not if intelligence is heritable, but the extent of that heritability.

The standard way to estimate human heritability was to track similarities across individuals with varying degrees of relatedness. For example, compare identical twin correlations on a trait with fraternal twin correlations. The main objection to these methods is that one could argue that environmental factors were correlated with particular genetic relationships (e.g., you treat individuals who are presumed identical twins more similarly). There are many reasons that I’m skeptical of extreme objections in this vein, but there are out there. This particular experiment design sidesteps that issue by looking at unrelated individuals. Not just notionally unrelated individuals, but actually those who were not genomically related. That’s a key difference between quantitative genetics and quantitative genomics. The former takes biological relatedness at face value, translating from ideal categories. The relatedness between full siblings is 0.50 for example. But when you look at the genomic level you can account and correct for the variation of that relatedness amongst siblings (e.g., two of my siblings exhibit a relatedness of only 0.42)! In this study they focused on numerous widely dispersed single nucelotide polymorphisms (SNPS), specific variants within genes, and used these to infer the nature of the genetic architecture of intelligence. More specifically, the genetic variation of two forms of intelligence, crystallized and fluid. The former seems to correspond to knowledge and the latter to raw problem solving abilities. Perhaps the difference between having an excellent operative system and applications vs. top of the line hardware?

In any case, here’s their abstract:

General intelligence is an important human quantitative trait that accounts for much of the variation in diverse cognitive abilities. Individual differences in intelligence are strongly associated with many important life outcomes, including educational and occupational attainments, income, health and lifespan. Data from twin and family studies are consistent with a high heritability of intelligence, but this inference has been controversial. We conducted a genome-wide analysis of 3511 unrelated adults with data on 549 692 single nucleotide polymorphisms (SNPs) and detailed phenotypes on cognitive traits. We estimate that 40% of the variation in crystallized-type intelligence and 51% of the variation in fluid-type intelligence between individuals is accounted for by linkage disequilibrium between genotyped common SNP markers and unknown causal variants. These estimates provide lower bounds for the narrow-sense heritability of the traits. We partitioned genetic variation on individual chromosomes and found that, on average, longer chromosomes explain more variation. Finally, using just SNP data we predicted ∼1% of the variance of crystallized and fluid cognitive phenotypes in an independent sample (P=0.009 and 0.028, respectively). Our results unequivocally confirm that a substantial proportion of individual differences in human intelligence is due to genetic variation, and are consistent with many genes of small effects underlying the additive genetic influences on intelligence.

The authors suggest that these values are a floor to heritability estimates, at least with the sorts of homogeneous populations they have here. That’s because their statistical genetic method is likely to miss a lot of true genetic variance due to its diminishing power when the causal genes are at too low of a frequency. They’re working within a framework where a given typed marker is correlated with a nearby I.Q. causal marker. At very low correlations they are going to miss the causal variant.

Some of the psychologists interviewed by the media contended that on one level these were banal findings. A value in the range they report is entirely within the mainstream of behavior genetic studies, which use pedigrees and what not. But many people don’t trust behavior genetics for whatever reason. One person’s banality is another person’s profundity.

But I think these sorts of findings should tilt us away from the proposition that large effect quantitative trait loci are common for I.Q. By this, I mean an “I.Q. gene” which is responsible for a huge difference between two people. There are some of these no doubt, especially those which result in mental retardation, but they don’t play that much of a role in all likelihood in ‘normal’ variation. Earlier linkage studies which reported such genes made huge media splashes and tended to fade because of lack of replication. Those genes may actually have been real QTLs, but the huge effect was likely to have been a random chance occurrence. Genome-wide association is better able to detect smaller effect genes within populations, but even it has been notably lacking in robust results.

Overall, this is good science. The results aren’t what those of us who were hoping that the intersection between psychometrics and genomics would yield low hanging fruit were pulling for. But reality is often likely to dash our hopes. No matter the banality of the final results, I do think the figure to the left is rather cool. It shows that the larger the chromosome the greater the proportion of variance is explained by that chromosome. This is entirely expected in theory (large chromosomes would carry more causal variants), but it is gratifying to still see it born out empirically.

(Republished from Discover/GNXP by permission of author or representative)
🔊 Listen RSS

Update: Stephen Dubner emailed me, and pointed me to this much longer segment which has a lot of Bryan Caplan. So it seems like the omission that I perceived was more of an issue with the production and editing process and constraints of the Marketplace segment than anything else.

End Update

I play a lot of podcasts during the day as I go about my business on my iPod shuffle. One of them is Marketplace, which has a regular Freakonomics Radio segment, where Stephen Dubner “freaks” you out with incredible facts and analysis, often with a helping hand from Steven Levitt. With all due respect to Dubner and Levitt, this still has very pre-Lehman feel. Economics has “solved” the workings of the explicit market, so why not move on to other areas which are ripe for conquest by the “logic of life?”

In any case this week’s episode kind of ticked me off just a little. It started off with the observation that college educated women apparently put 22 hours weekly into childcare today, vs. 13 hours in the 1980s. I guess fewer latchkey kids and more “helicopter parents?” Dubner basically indicates that the reasoning behind this is many parents are in a “red queen” arms race to polish the c.v.’s of their children for selective universities. This makes qualitative sense, but can we explain an increase of 9 hours on average for the ~25% of women who are college educated on striving to make sure that their kids have Wesleyan as the safety school?

Let’s put our quantitative “thinking-caps” on “freakonomics” style. ~25% of adults have university degrees. ~80% of these have public university degrees, which are usually not too selective. Some of the ~20% are from not particularly elite religious colleges. So the subset of Americans who graduated from elite universities is actually not too large a number. You can include these as natural aspirants for the best spots for their children. And a proportion of the large remainder, I’d estimate ~90%, who didn’t go to a university which required a great deal of stress and c.v. polishing would certainly strive and hope for better for their kids. But can this explain a 9 hour average rise among tens of millions of women? Doesn’t seem to pass the smell test for me. I suspect there’s a more general norm of shifting toward “high investment parenting” among the college educated cohorts.

A second aspect of the Dubner piece for Marketplace is that it totally doesn’t clue the listener in to the reality that there’s a huge behavior genetic literature which predates the interest of economics in the outcomes of parenting. ~10 years ago Judith Rich Harris came out with The Nurture Assumption, which reported the conventional finding that shared family environment only explains a small proportion of the variation in many behavioral outcomes within the population. The remainder is split between genes and “other environment” (which is a catchall category). More recently Bryan Caplan’s Selfish Reasons to Have More Kids is steeped in Harris’ work. It’s gotten a lot of media exposure, so I was surprised that Dubner didn’t mention Caplan. Instead he focused on Bruce Sacerdote at Dartmouth, who has done some research on outcomes for adoptive and biological children.

His research in this area seems about right, judging from what I know about findings in behavior genetics. In other words, he’s not a trail-blazer as much as a trail-tender. You can find a representative paper online, What happens when we randomly assign children to families?:

I use a new data set of Korean-American adoptees who, as infants, were randomly assigned to families in the U.S. I examine the treatment effects from being assigned to a high income family, a high education family or a family with four or more children. I calculate the transmission of income, education and health characteristics from adoptive parents to adoptees. I then compare these coefficients of transmission to the analogous coefficients for biological children in the same families, and to children raised by their biological parents in other data sets. Having a college educated mother increases an adoptee’s probability of graduating from college by 7 percentage points, but raises a biological child’s probability of graduating from college by 26 percentage points. In contrast, transmission of drinking and smoking behavior from parents to children is as strong for adoptees as for non-adoptees. For height, obesity, and income, transmission coefficients are significantly higher for non-adoptees than for adoptees. In this sample, sibling gender composition does not appear to affect adoptee outcomes nor does the mix of adoptee siblings versus biological siblings.

If you are an adopted kid there are some traits where parents matter a lot. For example, what religion you follow. There are some traits where parents don’t matter much at all. For example, how tall you’re going to turn out to be. And there are all the traits in between, like whether you’re going to finish college or are a regular church attender. Like most economics papers there’s a lot of fancy regressions. But a few figures and tables will give you the right idea.

The table below shows the proportion of the variation of adopted and biological children as explained by the variation of the parents. The key is to look at the ratio column. You probably wouldn’t be too surprised variation in parents’ heights can explain 10 times more of the variation in their biological children’s heights than their adopted children (ratio ~0.10). But variation in parents’ education explains 3.6 times more of the variation in their biological children’s outcomes than their adoptive children!

Overall, I agree with Dubner, Levitt, Sacerdote, Harris, and Caplan, that our society has convinced many parents that there are huge marginal returns in investment in quantity of time as opposed to quality. Falsely. By “our society,” I don’t mean specific people. Rather, I think the Zeitgeist changes from generation to generation, and some prominent people reflect that Zeitgeist. There was a time where nature was all dominant, and then the pendulum swung back to nurture during the era of the “frigid mother.” In the 1960s and 1970s despite the ascendant anti-hereditarian paradigm in the social sciences the rapid emergence of the “working mom” through female labor force participation resulted in less supervision in kids in households where both parents were working. But after this cultural “shock” perhaps we’ve adapted to the idea of women at work to the point where latchkey kids are no longer a culturally acceptable option? Or at least if you do have latchkey kids you’re negligent. Much of the reaction to the free-range kids movement seems to verge on moral panic, indicating to me that helicopter-parenting has less to do with individual rational action and more to do with group norm adherence. “It’s just what’s done!”

In hindsight I would have to admit that I was a de facto latchkey kid, and I had a stay at home mom! I Just mapped the route to and from the public library I regularly walked over the summers starting at the age of 8, alone, and it comes it at 0.8 miles. My dad was always at work, and my mom had less interest in books than I did. I do recall some young librarians asking if I was “OK” as I was carting stools back and forth because I was way too short to reach the top shelves in the adult stacks, as if I was lost, but after a while they got use to my presence and didn’t bug me (though I do recall one security guard who always seemed to be think I was up to no good as I lugged the huge oversized biogeography books around).

If this post piqued your interest, don’t stop. To understand what this all means you need to think and read about this more.

- Gene-environment correlation
- Gene-environment interaction
- Heritability
- Norm of reaction

For example, if you are thinking, “OK, so Razib just explained that getting a college education is mostly genetic,” you don’t get what I am trying to say here.

(Republished from Discover/GNXP by permission of author or representative)
🔊 Listen RSS

In my experience most scientists are not too clear on the details of intelligence testing, perhaps because the whole area is somewhat in ill repute (except when you want to brag about your own SAT/GRE score!). This despite the fact that the profession of science is skewed toward the right end of the intelligence bell curve. Steve Hsu, a physicist at the University of Oregon (and someone I’ve known for a while in the interests of “full disclosure”) has a nice presentation up in PDF format which summarizes the major points of interest in this area. Worth a skim if you are unfamiliar. Additionally he alludes to future directions in the study of the genetic basis of intelligence using genomics. Here’s his abstract:

I begin with a brief review of psychometric results concerning intelligence (sometimes referred to as the g factor, or IQ). The main results concern the stability, validity (predictive power) and heritability of adult IQ. Next, I discuss ongoing Genome Wide Association Studies which investigate the genetic basis of intelligence. Due mainly to the rapidly decreasing cost of sequencing (currently below $5k per genome), it is likely that within the next 5-10 years we will identify genes which account for a significant fraction of total IQ variation. Finally, I end with an analysis of possible near term genetic engineering for intelligence.

This talk is aimed at physicists and should be accessible even to those with no specialized background in psychology or biology.

Also, in case you are skeptical, Steve is quite aware of the difficulties with the enterprise which he outlines in the presentation assuming that the genetic architecture of intelligence is as he assumes. As sequencing gets cheaper and the sample size of full genomes hits the tens of thousands someone will tackle this, so he and his colleagues figured why not now?

(Republished from Discover/GNXP by permission of author or representative)
🔊 Listen RSS

WORDSUM is a variable in the General Social Survey. It is a 10 word vocabulary test. A score of 10 is perfect. A score of 0 means you didn’t know any of the vocabulary words. WORDSUM has a correlation of 0.71 with general intelligence. In other words, variation of WORDSUM can explain 50% of the variation of general intelligence. To the left is a distribution of WORDSUM results from the 2000s. As you can see, a score of 7 is modal. In the treatment below I will label 0-4 “Dumb,” 5-7 “Not Dumb,” and 8-10 “Smart.” Who says I’m not charitable? You also probably know that general intelligence has some correlation with income and wealth. But to what extent? One way you can look at this is inspecting the SEI variable in the GSS, which combines both monetary and non-monetary status and achievement, and see how it relates to WORDSUM. The correlation is 0.38. It’s there, but not that strong.

To further explore the issue I want to focus on two GSS variables, WEALTH and INCOME. WEALTH was asked in 2006, and it has a lot of categories of interest. INCOME has been asked a since 1974, but unfortunately its highest category is $25,000 and more, so there’s not much information at the non-low end of the scale (at least in current dollar values).

Below you see WEALTH crossed with WORDSUM. I’ve presented columns and rows adding up to 100%. Then you see INCOME crossed with WORDSUM. I’ve just created two categories, low, and non-low (less than $25,000 and more). Additionally, since the sample sizes were large I constrained to those 50 years and older for INCOME.

Wealth & Intelligence (2006)
Columns = 100%
Less than $40 K $40-$100 K $100-$250 K $250-$500 K More than $500 K
Dumb 22 14 12 13 5
Not Dumb 55 65 63 57 48
Smart 23 22 25 31 47
Row = 100%
Less than $40 K $40-$100 K $100-$250 K $250-$500 K More than $500 K
Dumb 50 13 18 16 4
Not Dumb 32 16 24 18 10
Smart 29 11 20 20 20
Income & intelligence (2000-2008), age 50 and above
Columns = 100%
Low Not Low
Dumb 32 11
Not Dumb 50 50
Smart 18 39
Row = 100%
Low Not Low
Dumb 58 42
Not Dumb 32 68
Smart 17 83

Of those with low income, about 1 out of 5 are smart. And of those who are smart, 1 out of 5 are poor. Remember, this is for those above the age of 50, not college students. I thought perhaps retirees might be skewing this. Constraining it to 50-64 changes the results some in a significant fashion. 1 out of 5 poor remain smart, but only 1 out of 10 of the smart are poor. As for the rich dumb, you have to look to wealth. It is notable to me that there’s a big drop off at more than $500,000 dollars in wealth. And, a large fraction of those with wealth in the $100,000 to $500,000 are dumb. I think we might be seeing the 2000s real estate boom.

In any case, I began to think of this after a recent post by the quant-blogger Audacious Epigone, Average IQ by occupation (estimated from median income). This is what he did:

…It’s not supposed to be an exact measure of IQ by profession by any means, as it is based entirely on average annual income figures. In other words, it’s an income table with the values converted to IQ scores….

…the following table estimates average IQ scores by occupation solely on the basis of the Career Cast mid-level income figures. The median salary (of a paralegal assistant) is taken to correspond to an IQ of 100. One standard deviation is assumed to be 15 IQ points….

You can see the full list at the Audacious Epigone‘s place, but here’s a selection I found of interest:

Occupation Estimated IQ from median income
Surgeon 234
Physician 161
CEO 148
Dentist 140
Attorney 128
Petroleum engineer 126
Pharmacist 126
Physicist 125
Astronomer 125
Financial planner 123
Nuclear engineer 121
Optometrist 121
Aerospace engineer 120
Mathematician 120
Economist 117
Software engineer 117
School principle 116
Electrical engineer 115
Web developer 115
Construction foreman 115
Geologist 114
Veterinarian 114
Mechanical engineer 113
Biologist 111
Statistician 111
Architect 111
Chemist 109
Stockbroker 109
Registered nurse 107
Historian 107
Philosopher 106
Accountant 106
Farmer 105
Zoologist 104
Author 103
Undertaker 103
Librarian 103
Anthropologist 103
Dietician 102
Archeologist 102
Physiologist 102
Teacher 102
Police officer 101
Actor 101
Electrician 100
Paralegal 100
Plumber 100
Clergy 98
Social worker 97
Carpenter 97
Machinist 96
Nuclear decontamination technician 96
Welder 95
Roofer 95
Bus driver 95
Agricultural scientist 95
Typist 94
Travel Agent 93
Butcher 92
Barber 90
Janitor 90
Maid 88
Dishwasher 88

Off the top of my head, I would say that the highest disjunction in the low income direction would be clergy. This is especially true for Roman Catholic and mainline Protestant denominations in the United States, which have moderately stringent educational prerequisites for their clerics. I assume that the biggest in the other direction are surgeons and medical doctors, who enter a market where there’s less and less real price signalling, where labor controls the supply of future labor, as well as well influencing the range of services that competitive professions (e.g., nurses) can provide.

(Republished from Discover/GNXP by permission of author or representative)
• Category: Science • Tags: Data Analysis, I.Q., WORDSUM 
🔊 Listen RSS

Image credit: Aleksandra Pospiech

One of the interesting and robust nuggets from behavior genetics is that heritability of psychological traits increases as one ages. Imagine for example you have a cohort of individuals you follow over their lives. At the age of 1 the heritability of I.Q. may be ~20%. This means that ~20% of the variation in the population of I.Q. explained by variation in the genes of the population. More concretely, you would only expect a weak parent-offspring correlation in I.Q. in this sample. At the age of 10 the heritability of I.Q. in the same sample may be ~40%, and in mature adulthood it may rise to ~80% (those are real numbers which I’ve borrowed from Robert Plomin). Many people find this result rather counterintuitive. How can a trait like intelligence become “more genetic”?

Remember that I’m talking about heritability here, not an ineffable “more” or “less” quantum of “genetic” aspect of a trait. In other words: does variation in genes due to different parental backgrounds matter for a trait? Second, the nature of psychological traits is somewhat slippery and plastic. As I’ve noted before the correlation between a score on a 10-world vocabulary test and general intelligence is pretty good. You can expect people with high scores on the vocabulary test to have higher I.Q.’s than those who have low scores. But if you take an individual and lock them in a room without human contact for their first 15 years, they are unlikely to exhibit any such correspondence. You don’t have to be a rocket scientist to understand why. Quantitative behavior genetic traits are complex and are subject to a host of background conditions, and express themselves in an environmental context.

So why can you explain more of the variance of a psychological trait like I.Q. at age 40 than at age 5 with genes? It has to do with environment. Specifically, intelligence isn’t something you’re born with, it’s something that you develop over time, through a complex confluence between biology and environment. The developmental process exhibits a level of contingency as well. Decision A redounds to the choice between B and C, which redounds between a further set of choices. Small initial differences in disposition and talent can compound over time through positive feedback loops. Practice may make perfect, but perfection may be a goal to which you aspire only if you have initial talent or inclination.

In other words, your genetic disposition can shape the environment you select, which can then serve to express your genetic potential in a specific manner. Children have less power in selection of their environment than adults. Over time the model is that environmental variables which differentiate children diminish in importance as they select contexts and situations which express their own preference sets as adults. This dynamic can be illustrated with a rather strange example. Consider two siblings who are pressured to be academic by their parents. One has a natural disposition toward scholarly activities, while the other does not. Their realized performance difference in youth may be small. People can respond to incentives! But at 18 the two siblings become adults, and begin to make their own decisions. At 25 one sibling may be a university drop out, and the other a graduate student. The modest differences in adolescence may start amplifying due to the positive feedback loops which consist of a set of choices which exhibit dependencies. Of course siblings would tend to be more similar than two random individuals off the street. But even within families there is genetic variance and so innate differences of disposition (the average difference in I.Q. between siblings is about the same as the average difference in I.Q. between two random people off the street, one standard deviation, or 15 points). Modeling behavior genetic phenomena in a rough & ready fashion is then a matter of keeping dynamic networks of parameters in your head. Traits aren’t constructed about of static blocks; they’re the outcomes of a set of parameters at a given moment, as well as a developmental arc shaped by a previous set of parameters (some of them the same, some of them new). Thinking like this gives you a method by which to analyze phenomena, it does not tell you in a clear and general manner how a whole range of phenomena emerged down to the last detail.

The analysis doesn’t just apply to populations over time. You can also look to different groups which are contemporary. In 2003 a paper was published, Socioeconomic Status Modifies Heritability of IQ in Young Children. The major findings are illustrated by this figure (I’ve added some clarifying labels):

On the x-axis you see socioeconomic status (SES). This variable is a compound of traits which reflect’s one’s position in the social status hierarchy. Income and wealth are clearly important, but a salesman for a fertilizer company could presumably be more economically well off than a physics professor. So other variables such as education also matter. It is clear then that as SES increases genetic variation explains much more of the variation in I.Q., while environment explains less and less. The shared environment is rather straightforward: your family. The non-shared environment is more vague, and to some extent is just the remainder from the model which predicts I.Q. In The Nurture Assumption Judith Rich Harris posited that non-shared environment was mostly peer group effects. Interestingly, by adulthood non-shared environment tends to be a more important variable than shared environment for most psychological traits.

Any guess for why genetic variance is more efficacious in prediction of I.Q. among the high status than the low status? Here’s a clue: heritability of height is much higher in developed nations than in developing nations. In other words, environment explains more of the variance in height in developing nations, while it explains almost none of the height in developed nations. There’s only so much you can eat, and there are diminishing returns on nutritional inputs. In developed nations most of the environmental variance has been removed due to adequate nutrition. When you remove the environmental variance, the genetic variance remains. Heritability is roughly the ratio of the additive genetic variance over the total variance, so its value gets larger.

The analogy to I.Q. should be relatively easy. Don’t tell Amy Chua, but there are probably diminishing marginal returns on “nurturing” environments for a child when it comes to their intellectual development. You have only a maximum of 24 hours in the day you can study and drill, and a personal library of 10,000 is probably not very different from 1,000, if all the books fall within the purview of your interest. Even in well off suburban communities there are differences of wealth and income, but on the margin vast increases in wealth and income do not allow one’s child to develop their mental faculties proportionality greater. What there remains in well off suburban communities are differences of genetic disposition and aptitude. Bill Gates’ children are probably good candidates for the Ivy League. Not because he is worth billions of dollars in relation to a professional whose net assets barely break a million. Gates got into Harvard, and reputedly did well before dropping out to pursue his business. His wife is also an overachiever.

This is I believe a fascinating topic, and needs to be explored in more detail. Some members of the same group now have a study out which shows that differences in socioeconomic status matter differently for infants at 10 months and tots are 2 years. Emergence of a Gene × Socioeconomic Status Interaction on Infant Mental Ability Between 10 Months and 2 Years:

Recent research in behavioral genetics has found evidence for a Gene × Environment interaction on cognitive ability: Individual differences in cognitive ability among children raised in socioeconomically advantaged homes are primarily due to genes, whereas environmental factors are more influential for children from disadvantaged homes. We investigated the developmental origins of this interaction in a sample of 750 pairs of twins measured on the Bayley Short Form test of infant mental ability, once at age 10 months and again at age 2 years. A Gene × Environment interaction was evident on the longitudinal change in mental ability over the study period. At age 10 months, genes accounted for negligible variation in mental ability across all levels of socioeconomic status (SES). However, genetic influences emerged over the course of development, with larger genetic influences emerging for infants raised in higher-SES homes. At age 2 years, genes accounted for nearly 50% of the variation in mental ability of children raised in high-SES homes, but genes continued to account for negligible variation in mental ability of children raised in low-SES homes.

They used a standard SEM model. I’m not going to go over that in detail, but suffice it to say that they related a set of variables to the outputs which they wanted to predict, performance on I.Q. tests for very young children. If you are curious, the demographic sample was rather diverse, and controlling for race did not impact their outcomes. So let’s outline what’s going on here.

First, predicted:

- Performance at 10 months
- Performance at 2 years

Second, putative predictors:

- Genes (A). Specifically, additive genetic variance
- Shared environment (C)
- Non-shared environment (E)

I’ve reedited some of the main results. On the Y axis you see the % of variance explainable by A, C, and E. The variance components are broken down into two levels: SES, and age. 2 SD means 2 standard deviations. In a normal distribution that’s the ~2% tail at the ends.

What you see are two trends with age and SES:

- For infants at the age of 10 months parents matter. Genes do not. SES is not a major issue.

- For tots at the age of 2 years, SES matters quite a bit. You see a recapitulation with the earlier data, where higher SES parents seem to be providing environments which probably exhibit diminishing marginal returns (environmental variance doesn’t have much of an effect on the margin), so that genetic variance is much more important by default. The trend is clear as you move in a step-wise fashio up the class ladder. Though I have to say, the top ~2% in SES is an elite group already, so I wonder what sort of environmental variance could be found there.

The figure to the left shows the same outcome out of their model, only now the curves illustartes the variation of the effects as you modify SES in a continuous fashion. These are estimates generated out of their model, so that probably explains the > 100% values you see on the margins. The key is to focus on the broad qualitative trends. Even at 2 years of age genes start to trump shared environment ~1 standard deviation above the norm (though not aggregate “environment”). If the earlier data is correct, the heritability will continue to increase over time for higher SES individuals, as their affluent backgrounds will give them perfect freedom to take them where their dispositions lead them.

Why does all this matter? There are practical outcomes to this sort of research. I’ll quite from the paper:

These findings build on a growing body of literature that highlights the importance of early life experiences for cognitive development…Current evidence suggests that, although children maintain a great deal of neurobiological and behavioral plasticity well past infancy…the predictive validity of infant mental ability for later cognitive ability is moderate…We agree with Bornstein and Sigman…who have strongly argued against the perspective “that infancy might play little or no role in determining the eventual cognitive performance of the child and, therefore, that individuals could sustain neglect in infancy if remediation were later made available”…Heckman…has recently taken an economic perspective on this topic. He argued that prophylactic interventions for disadvantaged younger children produce much higher rates of return on what he termed “human skill formation” than later remedial interventions for older children and adults do. On the basis of this perspective, Heckman concluded that “at current levels of funding, we overinvest in most schooling and post-schooling programs and underinvest in preschool programs for disadvantaged persons”….

My understanding is that the long-term effectiveness of even Head Start is non-existent, so I don’t know what proposals could be made based on this. Preschool for 1-2 years? I find it broadly plausible that high SES parents do provide more enriching environments, but I don’t see the detailed understanding necessary for genuinely effective prescriptions. Rather, we’re doing conventional trial & error when it comes to policy.

Additionally, the authors also admit that the high and low SES populations may have been stratified for genes. That’s just a way of saying that it isn’t as if genetic variance for things like intelligence are necessarily equally distributed across the social classes. If a genuine meritocracy exists what one should rapidly see is a crystallization of hereditary class castes, as individuals marry and associate assortatively on a meritocratic basis. Remember, assortative mating should increase heritability estimates (Quantitative Genetics says so!). This is part of the irony of some peoples’ conception of how genes relate to outcomes. Equality of opportunity will almost certainly lead to a cleaner separation of outcomes by genetic variation. In a chaotic world defined by random acts many people will find themselves in positions at variance with their aptitudes or dispositions. Once you remove the environmental randomness, then from each according to their capabilities should be the outcome.

For future investigation: the hypothesis that Goldman Sachs partners are precursors to Guild Navigators!

Citation: Tucker-Drob EM, Rhemtulla M, Harden KP, Turkheimer E, & Fask D (2010). Emergence of a Gene x Socioeconomic Status Interaction on Infant Mental Ability Between 10 Months and 2 Years. Psychological science : a journal of the American Psychological Society / APS PMID: 21169524

(Republished from Discover/GNXP by permission of author or representative)
🔊 Listen RSS

Jonah Lehrer has a post up, How Preschool Changes the Brain over at Frontal Cortex. He reports on a paper, Investing in our young people, which has been around for about 5 years. The top line of it is this, an investment in a $2,500/year (inflation adjusted) pre-school program in the early 1960s seems to have been effective in improving the life outcomes of at-risk low SES young black Americans tracked over their lives up to the age of 40. Their measured I.Q.s were not initially high, 85-75, 15th to the 5th percentile (though the median black American IQ is ~85, so not so low within ethnic group). They did gain an initial I.Q. boost, but like most of these programs that boost disappeared over time. But in terms of their non-cognitive skills there remained an appreciable effect which impact their life outcomes. What were these non-cognitive skills? To me they resemble classical bourgeois values rooted in low time preference. Willing to be a “grind,” work hard and forgo short-term pleasures and not cave in to impulses with short-term gains and long-term costs.

Here’s a figure from the paper which I’ve reedited with labels:


Intuitively we understand this. Through experience we know of this. There are individuals with high intellectual aptitudes who lack self-control. Who do not succeed in life because of poor life choices. There are individuals with mediocre intellectual aptitudes who achieve a certain amount of comfort and prestige in their life because of their rock solid focus on their goals. By analogy an old under-powered computer with Ubuntu installed on it running Open Office will still perform at a higher level in achieving productivity goals than a high-powered computer which is loaded with Windows riddled with spyware and mostly running games which require a lot of computational muscle power beyond the specs of the box.

My main question is one of interpretation: is the change in non-cognitive skill portfolio due to intervention at a “critical period” in a neurobiological sense? The authors make explicit analogy to language. If children are exposed to a language before the age of 12 they generally can learn and speak it without an accent with marginal effort. Severely abused, or in rarer cases “feral children,” who are not exposed to language at all in their formative years, may remain unable to speak fluently in any language for the rest of their years after recontact with mainstream society. This is likely a function of the biological aspect of language acquisition and learning. Or at least that is the contemporary consensus.

Does this apply to non-cognitive skills? I am moderately skeptical, though my attitude here is provisional at best. Through the pre-prints the authors take a methodological individualistic perspective. Individuals invest in their skills, and the earlier they invest in their skills the more positive feedback loops can emerge so that their skills can mature, extend and sharpen. There’s clearly something to this. But the focus on family environment and such in the paper makes me a touch skeptical. There is a large behavior genetic literature which suggests that family environment, “shared environment,” is not very predictive of long term outcomes. Rather, “non-shared environment” explained about 1/2 of the outcomes for many behavioral traits (the balance is genetic variation).

In The Nurture Assumption Judith Rich Harris argued that the non-shared environment really referred to peer groups. Again, the analogy to language is illustrative. Children do not speak with the accent of their parents, they speak with the accent of their peer groups. There is an exception to this: autistic children (or, children who consciously want to have a particular affect). Though I was not explicit, this is the sort of dynamic I was indicating when I suggested that culture matters in saving. Different cultures have different norms, values, and frameworks in which you can express your personality predispositions. In genetic terminology I’m talking about a norm of reaction.

Quickly skimming through the original paper which Jonah Lehrer’s post was based on (and skipping over the guts of the economic modeling) I was unclear if there was a long-term peer group effect, as they didn’t seem to explore this possibility. Perhaps instead of a critical period in a neurobiological sense, what we’re seeing here is the emergence of specific peer groups which reinforce and buffer individuals in decision making and goal setting? Perhaps the original intervention resulted in the emergence of a new subculture within the low SES black community of Ypsilanti, Michigan?

Life outcomes can vary a great deal based simply on social norms.


In terms of the bottom line this may not change the policy conclusion that much. The operational outcome of a given policy may be the same even if the means by which the outcomes are realized differ. That being said, I probably does matter on the margins if the effect is due to individual level biological changes vs. group level norm shifts when it comes to details of policy formation.

Image Credit: CDC

(Republished from Discover/GNXP by permission of author or representative)
• Category: Economics, Science • Tags: Behavior Genetics, Genetics, I.Q. 
No Items Found
Razib Khan
About Razib Khan

"I have degrees in biology and biochemistry, a passion for genetics, history, and philosophy, and shrimp is my favorite food. If you want to know more, see the links at"