Friday, September 20, 2013

Childhood SES amplifies genetic effects on adult intelligence

Timothy Bates, a professor of psychology at the University of Edinburgh, and an occasional commenter on this blog, has a new paper out, which looks quite interesting. [See comments for references to additional literature and an overview from Tim!]
Childhood Socioeconomic Status Amplifies Genetic Effects on Adult Intelligence

Studies of intelligence in children reveal significantly higher heritability among groups with high socioeconomic status (SES) than among groups with low SES. These interaction effects, however, have not been examined in adults, when between-families environmental effects are reduced. Using 1,702 adult twins (aged 24–84) for whom intelligence assessment data were available, we tested for interactions between childhood SES and genetic effects, between-families environmental effects, and unique environmental effects. Higher SES was associated with higher mean intelligence scores. Moreover, the magnitude of genetic influences on intelligence was proportional to SES. By contrast, environmental influences were constant. These results suggest that rather than setting lower and upper bounds on intelligence, genes multiply environmental inputs that support intellectual growth. This mechanism implies that increasing SES may raise average intelligence but also magnifies individual differences in intelligence.
See also WSJ coverage by Alison Gopnik:
... When psychologists first started studying twins, they found identical twins much more likely to have similar IQs than fraternal ones. They concluded that IQ was highly "heritable"—that is, due to genetic differences. But those were all high SES twins. Erik Turkheimer of the University of Virginia and his colleagues discovered that the picture was very different for poor, low-SES twins. For these children, there was very little difference between identical and fraternal twins: IQ was hardly heritable at all. Differences in the environment, like whether you lucked out with a good teacher, seemed to be much more important.

In the new study, the Bates team found this was even true when those children grew up. IQ was much less heritable for people who had grown up poor. This might seem paradoxical: After all, your DNA stays the same no matter how you are raised. The explanation is that IQ is influenced by education. Historically, absolute IQ scores have risen substantially as we've changed our environment so that more people go to school longer.

Richer children have similarly good educational opportunities, so genetic differences among them become more apparent. And since richer children have more educational choice, they (or their parents) can choose environments that accentuate and amplify their particular skills. A child who has genetic abilities that make her just slightly better at math may be more likely to take a math class, so she becomes even better at math.

But for poor children, haphazard differences in educational opportunity swamp genetic differences. Ending up in a terrible school or one a bit better can make a big difference. And poor children have fewer opportunities to tailor their education to their particular strengths. ...

24 comments:

tractal said...

Wow.

botti said...

Interesting to see those results with adult twins. There was a paper early last year by Plomin which looked at the moderating effect of SES on children's intelligence at ages 2, 3, 4, 7, 9, 10, 12 and 14:

"Results

We found greater variance in intelligence in low-SES families, but minimal evidence of GxE interaction across the eight ages. A power calculation indicated that a sample size of about 5000 twin pairs is required to detect moderation of the genetic component of intelligence as small as 0.25, with about 80% power - a difference of 11% to 53% in heritability, in low- (−2 standard deviations, SD) and high-SES (+2 SD) families. With samples at each age of about this size, the present study found no moderation of the genetic effect on intelligence. However, we found the greater variance in low-SES families is due to moderation of the environmental effect – an environment-environment interaction.

Conclusions

In a UK-representative sample, the genetic effect on intelligence is similar in low- and high-SES families. Children's shared experiences appear to explain the greater variation in intelligence in lower SES.
The environment can moderate the effect of genes - a phenomenon called gene-environment (GxE) interaction. Several studies have found that socioeconomic status (SES) modifies the heritability of children's intelligence. Among low-SES families, genetic factors have been reported to explain less of the variance in intelligence; the reverse is found for high-SES families. The evidence however is inconsistent. Other studies have reported an effect in the opposite direction (higher heritability in lower SES), or no moderation of the genetic effect on intelligence.

http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0030320

Diogenes said...

"The environment can moderate the effect of genes"


NO!

pnin1957 said...

Thanks for the reply. The mean age in your sample is about 54 with an SD of about 12 years. If the age distribution is normal, this means that the majority of subjects are not in their 50s but younger or older. A substantial percentage of the sample therefore consists of pairs of twin pairs that are several decades apart in age. If the distribution is non-normal and most of the subjects are in fact in their 50s, then there must be a rather large number of subjects who are extreme outliers age-wise.


What I am concerned with is that your SES variable must be highly correlated with the subjects' age, which in turn might be correlated with the variances of the IQ scores and the parameters of the biometric model. The variance of SES must also be much larger in this sample compared to if the subjects were of similar age. Your analysis would be much more credible had you reported results just for that "great bulk" of twins who are in their mid-50s or so.

You write in the paper:

In this final, preferred model, the main effect of unique environment on variance in IQ (expressed in IQ units) was 9.56, 95% CI = [8.64, 10.61], and the magnitude of the genetic interaction effect with SES was 3.19, 95% CI = [2.85, 3.51].

This seems to suggest that heredity has no main effect on IQ, and that the unshared environment is the main source of individual differences. Needless to say, this is a very radical claim that flies in the face of decades of research. I would have liked to see discussion in the paper about why you think all the previous studies in the USA using diverse methods (classic twin design, MZA, adoption, extended kinships) somehow missed the fact that IQ differences are overwhelmingly due to unsystematic influences ("noise") unrelated to either nature or nurture as traditionally understood.

TimothyBates said...

Whether or not the average heritability of the sample is attributed to a main effect, or included in the moderated effect (as it was here) is a statistical artifact of how SES is scaled: So the main effect of genes on IQ was zero but that doesn't mean the average h2 was zero, as can be seen: it's not.

The results also don't fly in the face of prior reports. In classic twin models (reported in the paper) we get the same (very high) heritabilities as have been reported by Bouchard and others in other samples. But you are right, there is a theoretical claim: it's about the mechanism by which many genes affect IQ. It is that "low IQ" alleles and "high IQ alleles" differ in how much information they extract from the environment and incorporate into the internal representations of the world which we manipulate when we reason.

So the alleles can't generate much variance between people in informationally impoverished environments (where there is little complexity to extract), but can generate large differences in attained IQ in environments with high information availability. I think this fits well with the finding that bright children have delayed cortical pruning and prolonged sensitivity to the environment (as reflected in a later loss of shared environmental effects among initially brighter children. Not sure that we're as much in disagreement as you might imagine.

TimothyBates said...

Thanks for the comment Diogenes. We've recently been looking at whether education raises g or raises task-specific skill. In one paper that's out, we found that education raised test scores, but left basic processing speed unchanged. In another that we're having a hard time placing, we found that school doesn't raise g, but rather raises test scores in a practice-specific fashion.

But I think there's a big difference between these mean-raising effects of math class, versus the genetic variance raising (heritability raising) effects of SES.

Robert Plomin highlights gene-environment covariation: Where genes cause people to actively evoke a more complex environment: Either evoking complex stimulation from existing environments (getting parents to use more complex language) or choosing complex niches (picking smarter friends, reading wikipedia articles). A powerful way to consider this is to think of environmental complexity as part of the extended phenotype of the gene.

If choices are unavailable (imagine everyone around you is dangerous and unthinking), then this kind of GxE covariation would turn into the sort of GxE interaction we observed, where the gene has a bigger effect in a better environment... Elliot Tucker Drob and Paige Harden have done work on trying to test this. For instance:

Tucker-Drob, E. M., & Harden, K. P. (2012). Intellectual interest mediates gene x socioeconomic status interaction on adolescent academic achievement. Child Development, 83(2), 743-757. doi: 10.1111/j.1467-8624.2011.01721.x

You might enjoy this (2-page) article by Professor Plomin this year.

Plomin, R. (2013). Commentary: Missing heritability, polygenic scores, and gene-environment correlation. Journal of Child Psychology and Psychiatry and Allied Disciplines, 54(10), 1147-1149. doi: 10.1111/jcpp.12128

There's a non-firewall copy on our journal club here http://differentialclub.wikidot.com/gxe

pnin1957 said...

Whether or not the average heritability of the sample is attributed to a main effect, or included in the moderated effect (as it was here) is a statistical artifact of how SES is scaled

In Figure 3, what does the SES scale mean? How does it relate to the Duncan scale? What percentage of the population within the same cohort are at each level on the scale? There should've been information in the paper on what the SES index means in qualitative terms and how it is related to age.

Your h2 estimate is on the low side for an adult sample. (The ACE parameters appear to be misreported in the paper. You should square them, e.g., h2 should be about .59, not .77.). For example, this recent study put h2 at 92%. They estimated it at the level of latent factors, so measurement error should be minimal. One of the things missing in your study is information on the reliability of the IQ measure used, i.e., how much of the e2 component is due to unreliability.

So the alleles can't generate much variance between people in informationally impoverished environments (where there is little complexity to extract), but can generate large differences in attained IQ in environments with high information availability.

I don't see how any reasonable person could disagree with that theory, but I think, for reasons I've enumerated, that your study fails to convincingly show that SES differences in the US represent environmental differences that are relevant in this respect.

Henry Harpending said...

Are there differences in the extent of assortative mating by SES?

TimothyBates said...

Regarding SES, I agree that with more space we could give more information, such as qualitative descriptions of what life is like at various levels of SES. FYI, Duncan SES levels in this sample are reasonably normal with peaks that one gets given qualitative gaps in status: for instance college create a step-up in status, and narrow bands of status with more people in them (like teachers and nursing often) generate peaks in SES.

As to the nature of social status: What the effective agent that family SES gives to children is, to my mind, a very under researched question: Is it parental education, or money, health, square metres of housing, social esteem... If we could identify the active ingredient, we might be able to dispense it more cheaply that by requiring everyone to generate the wealth in the first place.... It would not be surprising to me if we could raise mean IQ and the number of extreme high scorers within the current education budget, and then generate a very high return on that investment in education. I'd like more people to stop arguing over whether people differ, or being upset about differences in outcome get behind that goal of doing a better job for each child.

The h2 estimate is what it is, but not particularly "on the low side". Scandinavia has higher h2 than most. Cast an eye over data in papers like this and it's clear that h2 varies from sample to sample. We need to be able to understand these variations mechanistically.

Deary IJ, Spinath FM, Bates TC. Genetics of intelligence. Eur J Hum Genet. 2006 Jun;14(6):690-700.

A point we wished to emphasise in Psych Science paper is that one doesn't need to treat genetics a proportion of variance (h2), but as an absolute - h2 could remain constant or even decrease, while genes caused more differences in the world.

The ace parameters aren't "misreported": They're clearly labeled as to what they are. It's like saying a correlation is misreported because it's not an R2... One just needs to be consistent and clear, which we were.

When speaking of h2 for IQ as high as 92%, you note "They estimated it at the level of latent factors, so measurement error should be minimal". That's actually saying "this a percentage of smaller amount of variance, as they subtracted both error and all the valid but non-general variance, so it should be a quite a bit larger than what you report", which of course it is. A significant amount of e2 in the models is likely due to unreliability and if you took our h2 as a proportion of reliable variance it would of course be higher. Happily we seem to have moved from the theory being dramatic to one which no reasonable person could disagree with. I would seldom be convinced by any single study, and ours is no exception. The answer lies in the coherent outcome of multiple studies. To that end, colleagues and I are about two-thirds of the way through a meta-analysis of the world's literature on gene x SES interactions, taking into account age, nation, SES measurement etc. Should help put many of these issues to bed.

Diogenes said...

" I would have liked to see discussion in the paper about why you think
all the previous studies in the USA using diverse methods (classic twin
design, MZA, adoption, extended kinships) somehow missed the fact that
IQ differences are overwhelmingly due to unsystematic influences
("noise") unrelated to either nature or nurture as traditionally
understood."


You've answered your own question. Studies in the USA will be restricted by the range of environments in the USA.

Diogenes said...

"Scandinavia has higher h2 than most."


yet intergenerational income elasticity is almost 0.

Diogenes said...

"These findings are consistent with the hypothesis that higher SES
affords greater opportunity for children to seek out and benefit from
learning experiences that are congruent with their genetically
influenced intellectual interests."


And thus it is shown that rich cacti find the desert and rich red woods find Big Sur, but then the height of the cacti is compared to that of the redwoods and the cacti are found wanting.



“I am almost certain that complete
genome sequencing will become part of newborn
screening in the next few years. It is likely that within
a few decades people will look back on our current
circumstance with a sense of disbelief that wescreened for so few conditions”


I hope this will force the issue of eugenics. It seems every input to production is subject to "the law of supply an demand" except for the production of human beings.

Iamexpert said...

I thought the consensus was that between family environmental differences contributed NOTHING to IQ by the time one reaches middle age. Did I miss something?



Richer children have similarly good educational opportunities, so
genetic differences among them become more apparent. And since richer
children have more educational choice, they (or their parents) can
choose environments that accentuate and amplify their particular skills.
A child who has genetic abilities that make her just slightly better at
math may be more likely to take a math class, so she becomes even
better at math.


But IQ tests are largely aptitude tests not achievement tests (though achievement tests like vocabulary are often used as an excellent proxy for aptitude). If one's access to math education is influencing scores, then it's not a valid measure of intelligence for that person, and tests of fluid intelligence by definition are not measuring acquired knowledge. However socioeconomic background does influence values. People from upper class homes might be more likely to uniformly try hard on tests, so for the upper class, tests accurately reflect innate ability. By contrast, among the lower class, some may try very hard while others are so beaten down by the system that they are openly hostile to standardized testing, however I would have thought these types of factors would vanish by adulthood.. The lower class may also be more likely to suffer from head injuries, substance abuse, fetal alcohol syndrome, and other non-genetic influences on g.

Diogenes said...

The definitive twin study or study of closely related people is one where the geographic/cultural diversity is maximal between twins, but where z scores are for the respective populations. Of all those already done one might choose the pairs which grew up and live at greatest geographic distance. Perth is more like London than Dusseldorf is, perhaps, but it would still be revealing.



A study comparing related people in ND and ZA or DE and Namibia would be revealing.

Diogenes said...

"though achievement tests like vocabulary are often used as an excellent proxy for aptitude"


had to stop reading there. vocab is the most g loaded and most reliable subtest of the WISC and of the WAIS. there is ABSOLUTELY no real distiction between aptitude and achievements tests. the subject specific a-levels of english edu are as correlated with self-described iq tests as those tests are with one another, that is they are iq tests. the aptitude achievement distinction is purely nominal.

Diogenes said...

would that affect heritability?


if the parents have found their native soil, their children are likely to have the same best environment and the parents will likely give this to them.


steve's dad was a prof of aeronautical engineering. his own children have a mother who's a prof of comp lit. i expect his children have a higher liability to becoming humanists or lawyers than steve had.

pnin1957 said...

The ace parameters aren't "misreported": They're clearly labeled as to what they are. It's like saying a correlation is misreported because it's not an R2

This is what you wrote in the paper:

The model indicated that the average heritability, h2 = .77, 95% confidence interval (CI) = [.55, .82], and unique environmental effects, e2 = .63, 95% CI = [.57, .71], were high. By contrast, the between-families environmental effects were nonsignificant, c2 = .04, 95% CI = [− .46, .48].



You say that heritability, h2, is .77, when in fact it's .59, and similarly for e2 and c2. This is indeed as if you reported a correlation but claimed it was R2.


Anyway, I'm looking forward to that meta-analysis.

Iamexpert said...

had to stop reading there. vocab is the most g loaded and most reliable subtest of the WISC and of the WAIS. there is ABSOLUTELY no real distiction between aptitude and achievements tests. the subject
specific a-levels of english edu are as correlated with self-described iq tests as those tests are with one another, that is they are iq tests. the aptitude achievement distinction is purely nominal.



The distinction between aptitude and achievement tests is admittedly statistically irrelevant in the general U.S. population overall, but there are individuals and samples for which it becomes absolutely crucial. Consider cross cultural testing for example. Many immigrants from China bomb on achievement tests like vocabulary only to be moved to aptitude tests like the Raven and score in the genius range. Consider someone with brain damage or dementia. They may perform well on achievement tests like vocabulary because they acquired that word knowledge when their minds were sharp, and thus achievement scores serve as fossil remains of an ability they no longer possess, while aptitude tests like the Raven assess current intelligence. Consider studies assessing the effect of schooling on intelligence. Such studies would be circular if they used a test that measured scholastic knowledge acquired in school; you want tests that measure innate ability as directly as possible. Consider a study measuring the effects of certain drugs or medication on intelligence. Once again, you need to measure current functioning but vocabulary represents a life time of cognitive functioning.


And btw vocabulary is NO LONGER the most g loaded subtest on the Wechsler scales. That honor now goes to "Figure Weights" followed by the "Arithmetic" subtest.

Diogenes said...

right. i think you may be showing the typical and understandable prejudice that non-verbal tests are less dependent on education than verbal tests.

afrikaners scored 10 pts below british south africans in the 50s. on the verbal subtests? no. the difference was all in the non-verbal. now the difference has been erased.

using so many words and the ability to define words are two different though related abilities, sort of like the backwards digit span is more g-loaded than the forward.

Iamexpert said...

I haven't researched Afrikaner intelligence, but who says their allegedly lower scores in the 1950s were caused by lower education? Perhaps they had sub-optimum nutrition. The Dutch for example saw their non-verbal IQ's jump 21 points in 30 years, an increase that was paralleled by astonishing gains in Dutch height, strongly suggesting that 20th century nutrition has dramatically boosted both the brain and the body; indeed brain weight at autopsy has been increasing since the late 19th century, and studies of identical twins where only one is low birth weight and low head circumference reveal the stunted twin to have lower non-verbal IQ at age 15 (verbal IQ is preserved).

I've seen studies where education influences non-verbal IQ but it's usually in young children and/or in culturally loaded non-verbal tests which require one to recognize cultural objects. But culture reduced tests, particularly once one is old enough to follow basic instructions and focus, are probably much less sensitive to education than vocab tests are. On the WAIS-IV, for example, 16 year old olds score as high or better than 22 year olds on culture reduced tests, even though the latter have largely been university educated and the former are still in high school. But on vocab tests, the 22 year olds are far ahead.

And yes, vocab is still the most reliable.

JayMan said...

This exemplifies another Americanism. A person, a thing with a gender
though indefinite, should always be referred to with the gender neutral
"he, his, him". It's really horrible when he and she are randomly
distributed. Newspeak.


Agreed!

JayMan said...

Even still, I'd like to see such studies use IQ tests where we can decompose the g-loadings of the various subtests. Rushton & Jensen have argued that superior environments mostly affect the less g-loaded items on the test. Reaction time might be a good "pure" measure, for example.

Diogenes said...

here's a "behind blue eyes" for you may black-chinese-white jamaican:


the one woman this pathetically still single white guy has loved was "into" black guys. can't do anything about that. and she was very smart and ingrid bergman in looks, an 11.



c'est la vie.

Diogenes said...

"But culture reduced tests, particularly once one is old enough to follow
basic instructions and focus, are probably much less sensitive to
education than vocab tests are."


yes, but the very idea that an amazonian whose language has only "one", "two", "a lot" or a hottentot, or a new guniean whose grandfather hunted heads could be tested with the raven's or "culture fair" anything is LUDICROUS.


AND vocab is one of those very g-loaded subtests which have seen almost no flynn effect iirc.

Blog Archive

Labels