[Please place all off-topic comments in the most recent open thread. They will not be published here]

Thanks to Louis Lello and his colleagues, we now can predict a person’s height just from his DNA and these height predictors correlate about 0.64 with actual within sex height.

Of course this correlation is based on a UK sample and as Mug of Pee has long argued, when the environment is that narrow, you can’t be sure the  genome is actually causing the height, or if some genomes just grow tall in particular countries for local reasons, but would not have a height advantage elsewhere (see reaction norms vs independent genetic effects)

Thus I was heartened to learn that this genomic predictor was tested in environments as diverse as South Asia, China and Africa.  One of the studies authors Steve Hsu writes:

Note, despite the reduction in power our predictor still captures more height variance than any other existing model for S. Asians, Chinese, Africans, etc.

So the predictive power falls below 0.64 as we move far from the country the predictor was created in, but “still captures more height variance than any other existing model”.

So how well does any other existing model do?  In their paper they write:

Recent studies using data from the interim release of the UKBB reported prediction correlations of about 0.5 for human height using roughly 100K individuals in the training

So this one genomic predictor correlates at least 0.5+ with within-sex height all over the world, suggesting a truly causal relationship.  Squaring the correlation tells us that within sex height heritability (in the most meaningful causal sense of the term) is at least 0.25.

But why only 0.25, when twin studies suggest heritabilities of roughly triple that?

One possibility is all the flaws in twin studies, but another possibility is that common additive genetic variants only account for a small fraction of the heritability of height and other complex polygenetic traits like IQ.  To find rare genetic variants you must look at the entire genome, but considering how expensive that is and how rare the rare variants are, few are willing to spend the money.  In addition, there may be non-additive gene on gene interactions and if these are sufficiently complex, they may never be found.

How much more heritability is left to be found?

Perhaps a study of cattle might provide a clue:

Common sequence variants captured 83%, 77%, 76% and 84% of the total genetic variance for fat, milk, and protein yields and fertility, respectively

If human height is anything like these cattle phenotypes, then maybe we’ve found 80%, suggesting genomic predictors could go from explaining 25% of the within sex height variance to 31%, implying a predictive correlation of 0.56.

And if genomic predictions can achieve that much precision for within sex height, they can likely do the same for IQ once they genotype a sufficiently large sample (one million people) taking a sufficiently valid test (the WAIS).