There’s recently been a lot of research on the genetic basis for academic accomplshments (i.e. grades, education level).  One might ask why people are studying school grades when they could be studying IQ.  Well, probably for the same reason that scientists have studied head size when they could have been studying brain size:  The former is more accessible.

Just as it’s easier to put a tape measure around someone’s head than it is to remove their brain from their skull (or now days, shove them under an MRI), it’s easier to ask someone for their GPA or highest degree than it is to give them an IQ test.

As a result, HUGE sample sizes have been emerging linking genes to scholastic success (hat-tip to Steve Hsu):

FT.com: Genetic scoring predicts how children do at school

Professor Robert Plomin, senior author, called the study “a tipping point for predicting individuals’ educational strengths and weaknesses from their DNA”. It is published in the journal Molecular Psychiatry.

An individual’s “polygenic score” is based on the presence or absence of 20,000 common DNA variants across many different genes. Each has a tiny effect on its own but together they explain 10 per cent of the variation in children’s educational attainment at the age of 16.

Taking the square root of 10%, we learn there’s a 0.32 correlation between genetic grades and phenotypic grades (i.e. actual grades).  That’s not a low correlation.  What it suggests is that if you have an embryo that is +5 standard deviations in genetic grades, you can predict with 50% certainty that it will grow up to have grades that are at least +5 SD(0.32) = +1.6 SD (top 5% of an average class), and such an embryo would likely grow to get a graduate degree (top 5% in years of education which I assume would enjoy the same correlation) which is more than even I have! 🙂

Of course as commenter Mugabe/Videla would point out, we don’t know if these are independent genetic effects.  That is, the 0.32 correlation between genetic grades and grades may only hold in specific cultures and thus needs to be tested internationally.

But is even a within culture 0.32 correlation a victory for HBD?  It depends to what extent grades are caused by mental traits (as identified by psychology) as opposed to random factors.  If grades are caused by mental traits then HBD predicts a high correlation between grades and genes because HBD believes mental traits are EXTREMELY genetic by (later) adulthood.

However if grades are caused by random factors (the school or teachers one has, etc), then even a moderate correlation between grades and genes is a win for HBD, because even with all those random factors, a genetic signal is breaking through the random noise.

My sense is that grades are caused in part by two putative mental traits:  IQ (intelligence quotient) and CQ (conscientiousness quotient).

In the general, non-selected U.S. population, IQ correlates about 0.55 which academic success, whether you measure that success by grades in classes where the full range of ability exists, years of education, or prestige of college attended (if any).

Let’s say CQ also correlates 0.55 with academic success, and let’s say there’s a zero correlation between IQ and CQ.  Then we might expect a composite measure of both traits (IQCQ) would explain 61% of the variation in grades

0.55(0.55) + 0.55(0.55) = 0.61

And yet genes only explain 10% of the variation in grades

If we assume that all 10% of the variation in grades that is genetic, is variation in IQCQ, it implies that 16% of IQCQ is genetic, does it not? (10/61 = 16)

That sounds pretty low considering mental traits are often claimed to have heritabilities approaching 80% (of course heritability is not precisely the same thing as phenotypic variation that is genetic, though the term is often used that way).

But maybe the study suffered from range restriction or maybe they only looked at a subset of genes, or maybe I’m making an error in my reasoning or assumptions.

Update July 31, 2016 (evening): On the other hand, finding genes that explain 16% of the variation in mental traits might be considered a HUGE victory for HBD, because we know from Genomewide Complex-Trait Analysis (GCTA), that the amount of  variance explained by specific genes that are found for a particular trait are often AN ORDER OF MAGNITUDE less than the amount of genetic variation in said trait that is shown to exist (at least within countries)

Many non-scientists have a great interest in heritability, but lack the science education and/or cognitive ability to understand modern techniques like Genome-wide Complex Trait Analysis (GCTA), so this post is a quick attempt to explain it. Full Disclosure: I have virtually no formal science training beyond high school but this is just an oversimplified explanation.

GCTA gives a measure of the squared correlation between additive genotype and phenotype.  The reason it’s so confusing is that you can’t directly correlate a phenotype with a genotype if you haven’t found the genes that code for that phenotype, and thus you can’t determine if someone is genetically high on a given trait.

So for example, you can’t determine if someone’s genetic IQ matches their actual IQ, if you don’t know if they have the genes for IQ.  Since a correlation, by definition, is how close the rank order of two variables (i.e. genetic IQ and actual IQ) agree, it can’t be directly calculated if one of said variables (i.e. genetic IQ) can’t be ranked.  It would be like trying to calculate the correlation between height and weight, but all the weights were reported in a language you didn’t speak.

To sidestep this problem, GCTA was invented by a scientist of East Asian heritage.  In GCTA, instead of ranking everyone in your sample from highest to lowest on each trait, you simply randomly assign people to pairs, and for each pair, calculate the genetic distance and the phenotype distance.  So for example, if the people who differ by 100 single nucleotide polymorphisms (SNPs), on average, differ by one standard deviation in IQ, and if people who differ by one standard deviation in IQ differ, on average, by 39 SNPs, then perhaps it can be inferred that (in this sample) the correlation between genetic IQ and actual IQ is whatever number when squared and multiplied by 100, equals 39.

That number is 0.62

This is because in a bivariate normal distribution, the slope of the standardized regression line equals the correlation between two variables, so if a genetic difference of 100 SNPs regresses to a one standard deviation difference in IQ, then one standard deviation must be only 62% as extreme as 100 SNPs and if a one standard deviation difference in IQ regresses to a 39 SNP difference, then 39 must be only 62% as extreme as one standard deviation.

Once we have the correlation of say 0.62 between additive genotype and phenotype , we square it to get the amount of variation explained which in this example would be 0.38 (the real number is probably much higher, and even higher still for broad-sense heritability).

Of course what very few people realize is that heritability is technically NOT the percentage of the phenotypic variation explained by genes, it’s the percentage explained by genes when environment is held constant or allowed to vary randomly.

I’m almost done my post on Angela Merkel, but I thought I would do a quick post about genetics and IQ.  Commenter Afrosapiens asked how we know there are 10,000 genes (actually genetic variants) for intelligence, instead of say 13,000, 15,000 or 20,000.  I found an article by Dr. Steve Hsu where he answered this very question.

Apparently, when scientists do a Genome Wide Complex Trait Analysis (GCTA), which is a technique where a sample of people are assigned to pairs, and then the phenotype-genotype difference in all the pairs is correlated, they can determine the number of  single nucleotide polymorphisms (SNPs) that are associated with a one standard deviation difference in phenotype.  On the IQ scale, a one standard deviation difference is about 15 IQ points.  Once they know how many  SNPs are associated with a 15 point IQ difference, they square that number of SNPs to estimate the total number of cognitive variants.  So apparently, a difference of 100 SNPs has been associated with a 15 point IQ difference and since 100² is 10,000, they estimate IQ is associated with 10,000 genetic variants.

Now you may be wondering why they square the number of variants that “cause” a one standard deviation IQ difference.  I assume it’s because the standard deviation squared is known as the variance, thus you have to square the number of SNPs that cause one SD change in IQ, to estimate how many cause the total variance in IQ.

Now I’m confused because in this source Hsu writes:

Average pairwise genetic distance changes with mean IQ and IQ difference: ∼ 39 SNPs per population SD

So in one source he’s saying that 100 SNPs are associated with a 1 SD change in IQ, but in another source he’s saying 39 SNPs.  Perhaps this is just an example of regression working both ways?  In other words people with a 100 SNP difference average a 15 IQ point difference, and people with a 15 IQ point difference average a 39 SNP difference.  Kind of like parents with an IQ of 125 average kids with an IQ of 115 but kids with an IQ of 115 average parents with an IQ of 109.

But this is where it gets really interesting.  Hsu writes:

Given that there are many thousands of potential positive variants, the implication is clear: If a human being could be engineered to have the positive version of each causal variant, they might exhibit cognitive ability which is roughly 100 standard deviations above average. This corresponds to more than 1,000 IQ points.

I don’t follow the math here.  If there are only 10,000 variants, and 100 variants causes a one SD increase in IQ,  then to be 100 standard deviations above average, wouldn’t you need 10,000 more positive variants than the average person?  But I thought there were only 10,000 variants in total so wouldn’t that imply the average person has no positive variants?

Can someone please explain.

In any event, if they ever were successful genetically engineering someone with an IQ 100 standard deviations above average, it would probably immediately die from an over-sized brain.

Got to love the internet.  You can get a free university education without even leaving your bed.  Just listened to a nice discussion about IQ and genes (hat-tip to Steve Hsu):

Among the interesting points:

IQ is probably determined by as many as 10,000 genes, and each gene may only influence IQ by as little as a fifth or a tenth of a point.

Once they get a million high quality genotype-phenotype pairs, they will be able to create a useful formula estimating IQ from genes, though commenter chartreuse would argue that reaction norms would make this formula only useful in a narrow range of environments.

Steve Hsu imagines police looking at the DNA left at the crime scene and knowing the suspect was a 6’1″ Asian with a 160 IQ, at which point someone on the panel joked that Hsu was talking about himself.

Apparently height can already be predicted from DNA to some degree.  Professor Manfred Kayser states:

We were able to predict extreme height, which is those in the upper 3 percent, with an accuracy of 0.75, where 0.5 is random and 1 is a perfect indicator.

I assume he means that if they find some DNA, his prediction of whether the person is extremely tall will be right 75% of the time?

Steve Hsu notes that about 16% of the variation in height can be traced to genes that have actually been identified which means that a formula predicting height from genes would correlate 0.4 with actual height.

0.4 is about the correlation between height and weight.  In other words, scientists are now at the point that can predict someone’s height from their genes as well as they can predict their height from their weight.

As more height genes are discovered, that correlation could roughly double.

Also mentioned in the above video is DNA editing CRISPR, but this will have little relevance to IQ for a long time to come, because if 10,000 genes are linked to IQ, and each one only affects it by a tenth of a point or so, then knowing where to edit to get a noticeable effect is currently impossible, and potentially dangerous, because humans have only 20,000 genes, so each one has many functions.

Scientist Steve Hsu writes:

GCTA (Genome-wide Complex Trait Analysis)…allows an estimation of heritability due to common SNPs using relatively small sample sizes (e.g., a few thousand genotype-phenotype pairs). The new method is independent of, but delivers results consistent with, “classical” methods such as twin and adoption studies. To oversimplify, it examines pairs of unrelated individuals and computes the correlation between pairwise phenotype similarity and genotype similarity (relatedness). It has been applied to height,intelligence, and many medical and psychiatric conditions.

Apparently this technique has been used to show that (fluid) IQ has a narrow sense of heritability of at least 51%.  Not quite the 80% heritability found in twin studies but keep in mind, the latter measures broad sense heritability.

Of course as commenter chartreuse has noted, all of these studies are based on local populations so they don’t prove that high IQ folks are truly genetically smarter than low IQ folks; rather they may only show that high IQ folks have genes that are compatible with a specific environment.  If their genes were planted in a different time and place,  low IQ people might have higher IQs than high IQ people.

In order to show independent genetic effects (geneotypes that are smarter in virtually every environment), charteruse feels you would need a study of people, preferably of similar genetic background, living in different countries.  He recomends a sample taken from the entire developed World.

On the other hand, I would recommend doing a study comparing unmixed African Americans (whose ancestors had been in the United States for centuries) with unmixed West Africans living in the same part of West Africa.  Despite having the same genetic background, the latter score much lower on IQ tests than the former, presumably because of the extreme deprivation of the Third World compared to the First World.

If you had a 1000 pairs of such unrelated individuals (with one member of each pair being an unmixed African American, and the other being a West African) and you found a high correlation between IQ similarity and genetic similarity among coethnics living in such radically different environments, then this should be convincing evidence of independent genetic effects on IQ

Anytime HBDers show commenter “Charteruse” evidence that IQ is substantially heritable, he dismisses it because it doesn’t show independent genetic effects.  In other words, twin studies and adoption research has more or less proven that that genes cause IQ within Western countries, but they have not yet proven that genes cause IQ in a wide range of environments.

In other words, if you score low on an IQ test, and then you read that IQ is highly genetic, you do not have to feel genetically inferior, because it could be that your genes only impair IQ in Western societies.  It could be that those same genes, if reared in the rain forest or the Arctic, with different foods and experiences, might make you incredibly brilliant.

Now if you have Down’s Syndrome, you would likely have an extremely low IQ in any environment, regardless of whether you were raised in the Arctic, the United States, or the jungles of Africa, but if you’re just a regular low IQ person, we just don’t know, because all the heritability studies have been confined to a small range of environments.  For example, when psychologists study identical twins raised apart, the twins and their co-twins were raised in the same country, often in the same town, so we don’t really know how highly their IQs would have correlated if they had been separated into radically different environments.

The father-child IQ correlation

It’s well known that the IQ correlation between fathers and sons is 0.45.  Part of the reason the correlation is this high is because assortative mating, where men mate with women of similar IQ, which maximizes the genetic similarity between parent and child.  I calculate that if men mated with women at random, the parent-child IQ correlation would drop to 0.3.

We know from adoption studies that how you were raised has almost no measurable effect on adult IQ within  the range of most American homes, so if men just donated their sperm to random women, and did not even meet their kids, the father-child IQ correlation would still be about 0.3.  A 0.3 correlation implies that the sperm of a man with an IQ of 145 will likely produce a kid that is 21 IQ points higher than the sperm of a man with IQ 75, even though the mother is chosen at random and neither she or the kid have any contact with him.

But what if we had a random sample of 1000 white American men donate their sperm and had their IQs tested, and then we paid a 1000 female San women living a hunter-gatherer life style in Africa to have their babies and raise them to adulthood.  If at adulthood, we tested the offspring using a version of the WAIS adapted to San culture, would we still find a 0.3 correlation between the offsprings’ IQs and the IQs of the American fathers they never knew, or would we find that genes that enhance IQ in America do not correlate with high IQ in this radically different environment, or worse still, correlate negatively?  Would the offspring of IQ 145 men still be 21 IQ points higher than the offspring of IQ 75 men?

If the father-child IQ correlation could transcend such radical differences in environment between father and child, then we could say there are independent genetic effects on IQ within the normal range of IQ variation, but if not, we would be forced to conclude that the high reported heritability of IQ is misleading.

Perhaps it would be more ethical to do the study the other way around.  Have random women in America paid to have and raise the baby of random San men.  Either way, this is a study that could easily be done.

Discussion

My own view is that the cognitive difference between humans and chimps are caused by independent genetic effects.  That is, in almost any environment, humans will be smarter than chimps.  I also think the IQ differences between the three largest most well established races (Negroids, Caucasoids, and Mongoloids) are caused by independent genetic effects, because these are very ancient and have had tens of thousands of years to evolve under many different kinds of circumstances.

I am not however 100% sure that the IQ differences within macro-races (East Asians > Native Americans or  Ashkenazi Jews > Whites > Arabs) are caused by independent genetic effects, because some of these differences supposedly evolved so so recently (Ashkenazi Jews > Whites) and do not appear to be corroborated by differences in brain size, so  more evidence is needed.

I’m also skeptical that the IQ differences between individuals are caused mostly by independent genetic effects.  As commenters Charteruse and Swank have noted, if IQ was highly selected during human evolution, then there should be relatively little genetic variation among humans (since all the low IQ genes were weeded out), especially humans of the same ethnic group.  We know that in any correlation, when there is extreme range restriction, correlations tend to shrink, so the correlation between IQ and genes should be small among humans.

That’s not to deny that the 10 IQ point difference between Ashkenazem and Whites is genetic and could not have evolved in the last 800 years, however it might imply that the difference is not an independent genetic effect.  In others words, if Jews and Gentiles were raised together in a radically different environment like the arctic or the jungle, the IQ gap might vanish, or even reverse, but no such studies have ever been done so we don’t know.

But it’s worth noting that genetic mutations that enhance IQ seem surprisingly rare.  For example, we’ve heard of genetic disorders that impair IQ or genetic disorders that enhance traits like height, but the evidence for genetic disorders that enhance IQ is slim.

If IQ enhancing mutations are rare, then evolution can only raise IQ by acting on existing genetic variation, and because there’s already been so much natural selection for high IQ in humans, such variation can not be that great.

On page 522 of Arthur Jensen’s book The g Factor, he discusses a study of identical twins raised apart (Newman et al., 1937). One pair of raised apart twins known as Gladys and Helen differed 1.5 standard deviations in IQ. Converting to a scale where white Americans average 100 (SD = 15), their adult IQs were 92 and 115 (a 23 point gap).

If IQ is extremely heritable, how do we explain this huge gap? Well it’s just one data point, and the two twins were raised in extremely different environments and differed greatly in health history, and even differed in finger prints (suggesting different prenatal effects), but what really struck me, is that one twin had only a third grade education and the other had graduated from college to become a school teacher. That sounds like a difference of 13 years of schooling!

About a year ago I wrote:

In a fascinating 1968 Swedish study, a scholar named K. Harnqvist compared the IQ’s of males who were tested at both age 18 and age 13. He found that if two boys, for example, were both of the same social class, and both had the same IQ at age 13, but one boy completed all four years of high school and the other completed none of them, the latter boy scored nearly 8 points lower at age 18. In other words, each of year of missed schooling causes IQ to drop by 1.8 points. Herrnstein and Murray independently found a similar effect on page 615 of The Bell Curve (1.65 IQ points per school year).

So if schooling adds 1.8 IQ points per year, then a 13 year difference in schooling between identical twins should produce a 13(1.8) = 23 point difference in IQ, and that’s exactly what we see.

This is an example of the much maligned Phenotype = Genotype + Environment model working beautifully. Somethings really are as simple as they seem.

Of course as I also noted a year ago:

It’s very unlikely that school makes you smarter, but it is possible that school makes you more test savvy and more motivated to do well on mental tests. Many IQ tests require complex focused thinking. Someone who dropped out of school at 13 and has been working in an outdoors fun type job is likely to find such mental effort excruciatingly boring, annoying, and intimidating, while someone who stayed in school her whole life and works in academia is likely to be used to sitting still and focusing and may enjoy the challenge and have the confidence and intellectual ego to stay motivated.

So one twin was probably educated beyond her ability and over-performed on the IQ test. The other twin was educated beneath her ability and under-performed on the IQ test. In this case it would make sense to subtract IQ points from the educated twin and given them to the uneducated twin.

Does that mean we should subtract IQ points from all college grads and give them to grade school dropouts? No, because unlike adoption studies where environments are relatively random, for most of us, our environment reflects our genes. That is, if you have low IQ parents who raise you to flunk out of school, you probably have low IQ genes, while someone who had high IQ parents and was raised to graduate college, probably has high IQ genes. So controlling for schooling will reduce the absolute difference in test scores, but it will not much change the rank order because the latter is set by genes, and IQ is a measure of the latter.

However if everyone were adopted, then controlling the environment (including schooling) would indeed affect the rank order of people on cognitive tests.

Recently I posted about a popular formula for estimating heritability (H^2):

H^2 = 2[(correlation between MZ twins raised together) – (correlation between DZ twins raised together)]

This formula assumes that since MZ twins share twice as many unique genes as DZ twins, then if a trait is heritable, MZ twins should correlate much more strongly than DZ twins do.

Implicit in this formula is the assumption that since MZ twins reared together and DZ twins reared together, both shared the same home, and were both born to the same mothers, then the only reason why MZ twins would be more similar in a trait is because they are more genetically similar. Thus the greater the similarity of MZ twins compared to DZ twins, the more heritable the trait is thought to be.

However a major flaw with this model is that MZ twins are not only more similar than DZ twins genetically, but at least before birth, they are more similar environmentally, because MZ twins typically grow in the same placenta while DZ twins almost never do.

One way to correct for this environmental confound is to compare the correlation between DZ twins with the correlation between MZ twins who grew in different placentas (dichorionic MZ twins).

I recently blogged about an IQ study that did exactly that and what they found was that on tests like vocabulary, even when using dichorionic MZ twins, you still get the sky high heritability estimates that traditional twin studies found. However on a culture reduced test of spatial reasoning (Block Design) heritability suddenly dropped to less than 10%.

However that study had a very small sample size, and commenter “Lion of the Judah-sphere” was immediately skeptical.

Well I recently found a similar study but with an absolutely enormous sample size, and the results are very different. In this study, when you take the difference between the correlations between dichorionic MZ twins and DZ twins (see table 2 at the end of chapter 3 of this document), and apply the formula cited at the start of this post, you get a heritability of 0.48 for vocabulary and an astonishing 0.8 for block design. For overall verbal IQ, you get a heritability of 0.72, and for overall non-verbal IQ, you get a heritability of 0.6, and for full-scale IQ, you get a heritability of 0.76.

And this study was conducted on children, so a study done on adults would be expected to yield even higher numbers.

Of course the formula I cited at the start of this post assumes the Phenotype = Genotype + Environment model so disputed by anti-hereditists, however the most formidable attack on twin studies is that they don’t control for prenatal factors. This criticism has now been rebutted, and it’s a great day to be a hereditist.

Commenter Swank posted a link that lead me to an excellent large-scale study where Swedish young men (ages 18-20) were adopted and compared to their same age genetic relatives who grew up in different homes. What the study found was that being raised by educated parents increases IQ, even in young adulthood, even when you control for genetic background.

The study ranked the education of the parents on a scale of 1 to 5. I’m not sure how this scale worked because I don’t have access to the full study, but you can imagine how such a scale might look in the United States:

Level 5 = Advanced university degree
Level 4 = University degree
Level 3 = High school diploma
Level 2 = 8th grade education
Level 1 = Less than an 8th grade education

The study found that each additional level of parental education added about 1.83 IQ points (1.71 to 1.94) to the young man’s IQ (controlling for the young man’s genetics).

What this study seems to suggest is that if you have a pair of identical twins, and one of them is raised by a seventh grade dropout (level 1 education?) and the other is raised by a PhD (level 5 education?), then by the time they are 19, the one raised by the PhD should score 7 IQ points higher.

Does this mean we should start subtracting IQ points from people raised in educated homes and adding IQ points to people raised in uneducated homes? Probably not because most people are not adopted, and so the extra IQ points caused by their educated parents correlate with the high IQ genes they inherited from their educated parents. So while differences in parental education widens the absolute gap between people raised in different homes, they don’t much change the rank order for non-adopted people, and IQ is a measure of rank order, not absolute differences.

It’s worth noting that the vast majority of Americans are not raised by PhDs or seventh grade dropouts, so that’s quite an extreme case. Randomly selected Americans would not differ anywhere near that much in parental education levels, so such differences would not significantly contribute to IQ variation, except perhaps on tests like the SAT, where socioeconomic effects are suspected to be more pronounced (though this is controversial).

Although being raised by more educated parents props up IQ, that’s not the same as saying it props up intelligence. I suspect being raised by educated parents exposes kids to more cultural experiences, which increases their performance on measures of knowledge and vocabulary. Educated parents probably also encourage their kids to stay in school longer, and train them to value intellectual tasks like sitting still and concentrating, and such attitudes can add a few extra points on IQ tests, but they don’t reflect genuine gains in intelligence. Further, these effects are probably large in childhood, but become small in late adolescence as this study shows. I suspect if there were a followup study at age 40, the effect would have all but vanished.

Only when environment increases the biological environment, particularly in the prenatal stage, do the IQ gains tend to reflect a genuine rise in intelligence. This is because intelligence appears to be an overwhelmingly physiological variable, that is not amenable, to any significant degree, to psychological intervention, except perhaps in the most extreme of cases. That doesn’t mean the way you raise your kids is irrelevant, but it’s perhaps, largely irrelevant to intelligence.