Following up on the "No Rich Kid Left Behind" discussion, I wanted to get some (ideally dispassionate) discussion of the twin and adoption study based estimates of IQ heritability. One summary of them is on p. 85 of this article:

http://www.gifted.uconn.edu/siegle/research/Correlation/Intelligence.pdf

This is not my field, but let me start with a couple thoughts:

1. With the twin adoption studies, the result seems to be that monozygotic twins who are raised together have IQ that is almost as correlated as MZ twins who are raised apart (presumably because both were adopted by different parents). So same DNA and same home yield almost as high a correlation as same DNA and different home.

It seems like one, and perhaps the main, critique of the external validity of estimates of the relative role of heredity and environment to a broader population (e.g., the entire U.S.) is that the variation in environments among families who adopt children may be smaller than the variation in environments across all families. Which seems plausible, given that adopting families are screened pretty heavily.

2. There are also studies comparing correlations among mono and di-zygotic twins. So same DNA and same home compared with half of the same DNA and same home. The MZ twins have higher correlations than the DZ twins, and if one assumes a functional form for the relationship between DNA similarity and IQ similarity, one can back out the role of heredity.

It seems like the grounds for criticizing this approach would focus on either the functional form assumption or on the some "MZ twins get more correlated environments than DZ twins" (dressed alike, mixed up at school, etc.) type of story.

Is that a fair summary of the evidence and the grounds for criticism? What am I missing?

That's about what I read into it, too.

I really liked Val's meta-study link. Haven't had a chance to look at that one completely, but the other thing that I'd suggest is that MOST study participants are probably coming from a place where home environments are likely to be more-or-less idealized, or at least "good."

Horrible homes seldom offer to become research study participants. KWIM?

What I've read is that adopted children have IQs that correlate highly to their non-biological adopted siblings as children (there is an effect of environment on IQ) but by adulthood the adopted subjects have IQs with higher correlations to their biological parents and one hypothesis is that the environment the subject chooses to be a part of in adulthood is somehow more greatly modulated by genes than the childhood environment of adopted children, which in turn influences IQ.

I have not read the study but I think it is worth investigating whether they used DNA testing to determine zygosity or whether it was determined from chorionicity (as was typically used prior to DNA testing being readily accessible for research purposes) . Importantly, the understanding now that 25% of dichorionic twins are indeed identical has cast doubt over a lot of previous twin research. Added to which there is now the knowledge of a third type of twin called semi identical or polar body twins.

FWIW I have twins who I assume are fraternal but have not been DNA tested with identical iq scores.

AP,

Just to sign the potential bias for those who are following along at home, if 25% of dichorionic twins are really monozygotic, then studies that use chorionicity as a prozy for zygoticity would underestimate the correlation difference between MZ and DZ twins (by a factor of around 4/3, assuming sharing a placenta doesn't confound things).

And thus they would *underestimate* the role of heredity. Which is not usually the direction from which these studies are critiqued. Am I right?

Thanks for keeping it dispassionate.

UV

Amniocity refers to membranes, chorionicity referes to placentation. All monochorionic (shared placenta) twins are monozygotic but not all dichorionic twins are dizygotic. So a shared placenta doesn't confound things but a fused placenta does. My twins had a placenta that was unable to be differentiated as fused or shared on gross inspection. Semi identical twins adds other confounding thought but the incidence is going to be very low.

Well, sadly for the research, not so much for the twins, I think it is less likely that in the future there will be a significant population of twins separated at birth to study. No ethical agency would do it. Of course, there are still unethical agencies--I know of twins my dd's age who were adopted to different families as infants, U.S. parents only found out after adoptions were completed, and one family relocated across the country so that twins would grow up together.

Just as a note, realize that a correlation of even say .8 could work out to a mean difference of 20 IQ points. Depends on the shape of the data and lots of other factors. Also, in these studies, as in most, the two percent tails barely contribute even noise to the results.

Excellent point.

It's also the reason why adoption studies in general are less useful than they used to be-- screening is SO thorough now that marginal/low SES homes are rare for adopted children.

Zen, I don't follow. The SD of IQ in the full population is ~15 pts. If we identify a group of pairs of twins whose IQs are correlated with coefficient 0.8, then I think the within-twin SD of IQ is 15*sqrt(1-.8^2) = 15*.6 = 9. For the normal distribution, mean(abs(x)) ~= 0.8 if SD(x) = 1. So mean absolute differences should be about 7 pts.

On the broader point, the "adoptive families are more similar than non-adoptive families" critique of the twin adoption studies seems to me more serious than the "MZ twins get more correlated environments than DZ twins" critique of the MZ-DZ twin studies. The latter I'm sure is true, but seems unlikely to be a big deal, since DZ twins are already getting very correlated environments in most families. Misidentifying MZ/DZ status will likely lead to underestimates of heritability. So is it fair to interpret the MZ-DZ estimates as a lower bound on heritability and the adoption study estimates as an upper bound? What would this be missing?

Yes, but it goes even deeper than this. Agencies are screening the applicants, but applicants are also screening the agencies, so to speak. Applicants learn all about the process and what to expect, and can opt out at any time.

And we're talking about a self-selecting group even before the screenings begin.