Thread by @SpencrGreenberg

There’s a really interesting debate raging in the field of genetics about how heritable different human traits are. It could end up overturning 100 years of research:
🧵

First off, what is heritability? It’s the fraction of the variance of a trait (in a particular population) that’s caused by genetic differences.

Everyone agrees that height is at least fairly heritable since, in most populations, much of the variation is attributable to genes.

A classical and very commonly used way to estimate the heritability of a trait is to compare how similar that trait is among identical twins relative to how similar it is among fraternal (non-identical but same-sex) twins.

The logic is that…

Since identical twins have ~100% of their genes in common, whereas fraternal twins have only ~50% (considering just those genes that are not typically shared between all humans), if identical twins are more similar to each other for a trait, it’s probably due to genetics.

This can be formalized with Falconer’s formula, which says that if the correlation of a trait (say, height) is ri for identical twins and rf for fraternal twins, then, under certain assumptions:

heritability = 2 * (ri - rf)

This means that if ri = rf, then heritability is 0.

If we use the method of comparing identical and fraternal twins, we get heritability results like those in this image (which I compiled from papers).

However, there is another, very different way to estimate heritability, which the revolution in DNA sequencing technology has made possible. Rather than looking at twins, people’s genetics can be measured, and traits can then be predicted directly from DNA.

The more accurate the DNA-based predictions are, the more heritable a trait is.

In the early days of these new methods, the heritability estimates were extremely low compared to the twin studies, which was puzzling.

As DNA technology has improved, the statistical approaches for making these predictions have advanced, and the data sets to train these algorithms on have grown in size, the DNA based methods have predicted higher heritabilities than before.

But here’s the problem: even with all these advancements, the DNA-based methods still usually predict much lower heritabilities than the twin studies - often less than half of what the twin studies say! This is called the “Missing Heritability” problem.

Why would the different approaches lead to such different results? Well, both approaches rely on assumptions.

For instance, the twin-based method assumes that:

i) the shared “environment” for identical twins is not more similar than the environment for fraternal twins

ii) that there is not a substantial amount of “assortative mating” - where parents end up more similar than expected by chance because they seek out traits that they have, such as people with college degrees seeking out others with college degrees

iii) that genes do not substantially change the probability of being exposed to different environments

On the other hand, the DNA methods make their own assumptions. These assumptions depend on the exact method applied, but a common one is that there aren’t “rare genetic variants” that contribute substantially to heritability.

So, what’s the result? Twin study based methods give us higher heritability estimates than DNA based methods. Is this because twin studies overestimate heritability or because DNA based methods underestimate it? Or does the truth fall somewhere in between?

This debate is still contentious. Hopefully, high-quality science will do its thing, and a new consensus will emerge. But if DNA based methods turn out to be more accurate, nearly 100 years of heritability estimates may need to be thrown out!

If you found this interesting, I'd appreciate a follow at @SpencrGreenberg !

You may also enjoy my newsletter (One Helpful Idea) - where I send out one idea weekly (a 30 sec read) about psychology, philosophy, or society:

bit.ly/onehelpfulidea