Are Twin Studies "Pretty Much Useless"?
By Bryan Caplan
The latest Slate features a new critique of twin studies by Brian Palmer, entitled “Double Inanity: Twin Studies Are Pretty Much Useless.”
Palmer starts with the standard guilt-by-association critique: Galton, the founder of twin studies, “coined the term ‘eugenics’ and was the inspiration for the push to manipulate human evolution through selective breeding.” Hopefully this won’t lead us to recoil in horror from Galton’s key statistical innovations – including standard deviation and regression. But Palmer’s critique of twin studies also includes two substantive arguments.
First, identical twins aren’t literally genetically identical: “That identical twins do not, in fact, have identical DNA has been known for some time.” Palmer doesn’t tell us how genetically related monozygotic twins really are on average, because the answer would still be “vastly more similar than any other humans on earth.” But in any case, the fact that identical twins don’t have identical genes actually implies that standard hereditability numbers underestimate the effect of genes. Here’s why:
Standard twin methods basically say, “Let’s see how much extra observed similarity an extra 50 percentage points of genetic similarity causes.” If you narrow the genetic distance between monozygotic (MZ) and dizygotic (DZ) twins to say 45 percentage points, you increase the estimated effect of genes by 50/45=11%. If you hold the observed similarity between two groups constant, and decrease their genetic similarity, genes look more powerful, not less.
Second, Palmer says, MZ and DZ twins don’t, contrary to standard assumptions, have “equal environments.” True enough, but does it matter? The best evidence says no.
a. Two studies (see * below) explicitly address this point by looking at twins with socially misidentified twin types. How do identical twins falsely believed by family and friends to be fraternal compare to fraternal twins falsely believed by family and friends to be identical? Both studies confirm that greater similarity of treatment does not make MZ twins are more similar.
b. Separated twin studies and standard twin studies give very similar results. Critics are quick to point out that many twins in separated twin studies were not totally separated. But many were totally separated. And if similarity of treatment matters, shouldn’t the “dosage” of this treatment matter too?
c. Adoption studies and standard twin studies also give very similar results.
There is a kernel of truth in Palmer’s critique. Twin studies aren’t perfect. I discuss all the main problems in Selfish Reasons to Have More Kids. The biggest problem is just that what you find depends on where you look. Twin studies don’t show that growing up on the streets of Haiti is harmless. Why not? Because the twins under observation almost invariably come from vaguely normal First World families. If you’re a vaguely normal First World parent or potential parent, twin studies are a gold mine of useful information. Outside of this range, twin studies remain merely suggestive. Even “suggestive,” though, is far from useless.
* Kenneth Kendler et al., “A Test of the Equal-Environment Assumption in Twin Studies of Psychiatric Illness,” Behavior Genetics 24 (1) (January 1993); Sandra Scarr and Louise Carter-Saltzman, “Twin Method: Defense of a Critical Assumption,” Behavior Genetics 9 (6) (June 1979). Kendler et al. and Peter Borkenau et al., “Similarity of Childhood Experiences and Personality Resemblance in Monozygotic and Dizygotic Twins: A Test of the Equal Environments Assumption,” Personality and Individual Differences 33 (2) (July 2002) also show that twins with more similar appearances and childhood environments are not more similar.
HT: Ananda Gupta
Update: Alex Tabarrok makes many of the same points, then adds:
[F]or an article that goes on about “modern genetics” the author seems
completely unaware that it is now possible to do a whole-genome
analysis. That is, instead of assuming that siblings share
50% of their genes on average it is possible to estimate, sibling-pair
by sibling-pair, how many genes siblings share and then correlate that
with various characteristics. Obviously, it takes a lot more data to do
a study like this but it has been done. Visscher et al.,
for example, use data from 3,375 sibling pairs to estimate the
heritability of height. Interestingly, they find a heritability of 0.8,
very close to that found in traditional studies.