Isn't it? Kudos to the authors for a super clear paper. The graphs make everything crystal clear. Succinct and clear graphs are oddly satisfying for me.
I'll gladly look at anything submitted by gwern, or hosted at his site, but can someone help out with a summary and some interpretations? I'm both not familiar enough with the field to be confident I am taking away the right conclusions form what is said, and without enough time to give to this topic to rectify that.
Sure. So here are a few of the highlights of the paper:
- They looked at nearly every single twin study over the last 50 years
- The combined population of the all the studies is 14 million twin pairs
- The results strongly suggest that every human trait is heritable as they didn't find one trait that had 0% heritability.
- The most surprising result for me is that across all the traits they compiled, the heritability is 49%. Which is much much lower than I would have expected.
There's a few more interesting points about monozygotic vs dizygotic, but I'll spare you the technicalities since it has a narrower scope of interest.
> - The most surprising result for me is that across all the traits they compiled, the heritability is 49%. Which is much much lower than I would have expected.
But much much higher than almost all people would believe.
And the fact that all studied traits in all categories were found to have a significant degree of heritability strongly suggests that where groups obviously differ in traits that are easily visible, a similar degree of group difference in non-visible traits is to be expected.
This does not follow at all, because there is no evidence for any master mechanism that would correlate the group average of one trait with another. So simply because a group can be clustered by one trait there is no reason to expect they will differ from the population average in others.
For example, if you were to sort the population the length of their pinky finger, and then look at two groups of people, one the longest-pinky quintile and one the shortest, there is nothing in this study that would suggest those two groups would have different resting heart rates. This is true even though we can assume on the basis of this work that both pinky length and resting heart rate are about 50% heritable.
The only way they would be correlated would be if there was a single selection mechanism for both, and given that visible traits and non-visible traits tend to have completely different selection mechinisms you'd have to show positively and directly by experiment that such a common mechanism existed in any given case.
Ab initio there is simply no basis for the claim any random non-visible traits like resting heart rate will have different group averages for groups that are segregated from the general population by any random visible characteristics like pinky length.
No, the grouping I'm talking about is not by some arbitrary trait but by genetic relatedness. Two groups with recent common ancestry within the group and only distant common ancestry across groups.
Genetic similarity is the "master mechanism that would correlate the group average of one trait with another."
Ah ... you think so gwern? I would assume that most people would think that twins are incredibly similar. Or maybe I'm in an echo chamber and need to get out more.
> > The most surprising result for me is that across all the traits they compiled, the heritability is 49%. Which is much much lower than I would have expected.
> But much much higher than almost all people would believe.
People's casual usage of the term heritability doesn't match the way it's used in scientific contexts. For instance, people wouldn't normally say that the trait of "being a slave" is explained by genetics, but the heritability (in the technical sense) of being a slave in the mid 1800's USA was rather high.
> but the heritability (in the technical sense) of being a slave in the mid 1800's USA was rather high.
No, it wouldn't be. Monozygotic black twins would have the exact same chance as dizygotic black twins of being slaves, and slavery would cluster in families, with whole families tending to be free or slave, leading to very high shared/nonshared environment estimates.
Actually, "heritability" in the scientific context can be counterintuitive sometimes. I'd bet that the heritability of slavery in the 19th century was actually pretty low - I wouldn't imagine that monozygotic twin pairs are any more/less likely to be be slaves than same-sex dizygotic counterparts. Indeed, the majority of variation of the "is a slave" phenotype would be due to environment as slaves get set free.
Think of a trait like hair colour in East Asians. Having black hair is completely determined by genes, but heritability is pretty much zero since (with maybe a few extreme exceptions) variation in hair colour in that population is entirely environmental (use of hair dyes, etc.). In Europeans however, heritability is probably much higher due to greater variation in hair colour.
So every trait can be heritable, but often they only are about half the time? To me, 49% seems kind of high, considering multiple parent, dominant and recessive traits, etc. Then again, I may not be interpreting heritability to mean the same thing as you meant in this context.
Heritability has a very precise definition in this context:
> Heritability is the proportion of observed differences on a trait among individuals of a population that are due to genetic differences. https://en.wikipedia.org/wiki/Heritability
It is also commonly misunderstood. In particular, heritability is always relative to an environment, and not independent from it. Height might be 50% heritable in one area but 75% heritable in another, for example (even with the same genetic population in both areas).
Yes you can’t compare heritability between populations only within populations.
The other big misunderstanding is that environment means the contribution of family or society. It is really just everything that is not genetic and is mostly noise. When you try to measure the effects of family background on most personality traits the contribution is close to zero.
> The results strongly suggest that every human trait is heritable as they didn't find one trait that had 0% heritability.
I can't find that conclusion in the article. I guess that there is some selection bias here. Even if some trait has a very small inheritability, the p values are very big and I guess it will be dropped from the original article because it's not interesting.
For example: "Preferred fingernail color" is probably not inheritable.
I had to read some of secondary sources to understand the implications of this paper.
Ultimately, studies show that on average human traits are typically due 1/2 to genetics and 1/2 to environment. However, some traits are more skewed one way or the other. For example, bipolar disorder among twins was about 70% due to genetics and 30% due to environmental factors. All traits researched were somehow impacted by genetics.
Also for about a third of traits, the nature part wasn't dependent on a specific gene, so gene isolation to help predict traits is not always possible.
What other major takeaways from this study do you find interesting?
>The most interesting aspect of these results is that for many traits there is no detectable non-additivity. That is, gene-gene interactions seem to be insignificant, and a simple linear genetic architecture is consistent with the results.
We already knew this sort of from animal studies and evolutionary theory. But it's interesting to see it confirmed in humans.
The implications of this, is that it will be very easy to predict phenotypes from genetics. We don't need to model complicated interactions between genes like many people thought, and can simply take the genes that correlate the most with a desired trait.
This means you could do a large enough study and find every gene that correlates with intelligence, and create the optimal combination. Creating humans orders of magnitudes more intelligent than any human who has ever lived.
Thanks! That is interesting. I would love to see that happen -- even if there are ethical issues to consider. Maybe we can start by making the most genetically intelligent rabbits or something.
> Ultimately, studies show that on average human traits are typically due 1/2 to genetics and 1/2 to environment.
Well, no. Without a qualifier, it only makes sense to say that traits are 100% due to genetics. Replace a human in whatever circumstance with something that shares no genetics with a human (say, a rock), and you'll find that the power of the environment to make it behave in normal human ways is underwhelming.
What this is saying is that of the variation that humans display, about half is generally explained (in a statistical sense) by (a particular definition of) genetic variation among those humans. Humans vary from zero to zero wings, and the heritability of wings (in humans) is therefore undefined. It would be hard to conclude from this that our lack of wings isn't due to genetics.
