There’s an interesting 2013 meta-analysis by Branigan et al about the heritability of education attainment as measured by twin studies:




Source:  Branigan et al, 2013

If you average all the studies in the US and UK,  the mean heritability is 0.31.

A recent study of 1.1 million people (largely from the U.K. and U.S.) found polygenic scores predicted 0.12 of the variation in education, or roughly 39% of twin studies’ heritability.

Why so much lower than twin studies?  One reason might be that genetic samples suffer from range restriction, since relatively educated people (like our very own G-man!) seem more likely to get genotyped.

I found this quote from the supplemental materials of a 2018 study by Ritchie et al.

A personal measure of socioeconomic status is educational attainment. We compared the distributions of educational attainment in UK Biobank to the data from the 2011 Census for England and Wales (available at the following URL:; England and Wales makes up the vast majority—around 89%—of the population of the United Kingdom). In the census, for those ages 50+ years, 25.5% of males and 20.5% of females reported having a ‘level 4 qualification’, the category including college/university degrees (we might expect this figure to be slightly higher were it restricted to the 44-77 age group, but that precise age subset was not available from the census data). In the subsample of UK Biobank used here, 48.0% of males and 42.2% of females reported having a college degree. Thus, the sample was not representative in terms of educational attainment: a higher proportion of individuals in general had a degree.

Adjusting for range restriction would perhaps increase the amount of education variance explained by DNA from 0.12 to 0.2 (though that’s just a guess).

0.2 is 65% of the heritability found in twin studies.

Meanwhile twin studies find about a 0.75 heritability for IQ.

That means we might expect polygenic scores to eventually explain 65% of 0.75 of the variance in IQ, or 0.49 at least among whites living in the West.

Of course that might be a huge overestimate of heritability, if much of the genetic variance is not causal (i.e. population stratification, gene environment interaction).

On the other hand it might be a huge underestimate of heritability, if much of the genetic variance in IQ is not capture by the additive effects of common SNPs.

Both possibilities likely cancel each other out to some degree.

But if the heritability of IQ really is 0.49 (one of the fiercest critics of twin studies suggested 0.45) then square rooting the heritability gives a potent 0.7 correlation between DNA and IQ.