Psychologists love twin studies, but they're mostly worthless. The unstated assumption is always "A twin study holds genetics constant while randomizing the environment".
Of course the families of two identical twins are far from independent variables. They're both made of the kinds of people who would adopt a child in the same city at the same point in time. You haven't managed to separate the variables at all!
Why? The twins who were adopted were of the exact same race, gender, age, and appearance at the moment of adoption. People aren't just given a random child when they adopt, they choose. So you still haven't untangled selection effects entirely.
This is correct. After the article was submitted, I went to the University of Minnesota main library (Wilson Library) and Bio-Medical Library to look up monographs on IQ and on human genetics. Adoption agencies, as one widely used genetics textbook notes, have taken care to place separated twins in matched, similar environments, so a study of identical twins reared apart still tends to understand the influence of environment on the development of most human traits.
The only one of those factors that can't be controlled for very easily is appearance at the moment of adoption.
def random_twin_like_pair(children):
x = random(children)
y = random([c for c in children if (c.race, c.age, c.gender) == (x.race, x.age, x.gender)])
return (x,y)
At which point you're down to n of 1-2 per group, because separated-at-birth twin studies commonly contain no more then 30-40 pairs of twins, and your study is fatally flawed.
I'm not saying twin studies couldn't be done well, I'm saying in practice most real twin studies are not large enough or well designed enough to actually control for those outcomes. There just aren't enough identical twins separated at birth to allow it to be done well.
(That's not to mention further issues like womb environment, which further muddy the issue and are very important)
You don't need to restrict to demographic cross sections. You can take a set of 30-40 twins (of all demographics) and compare them to a set of N>40 randomly selected twin-like pairs (of all demographics). As long as your twins and twin-like pairs have the same demographic mix, you have an adequate sample.
This sample will average out demographic variation, but the twin-induced correlations are controlled for.
Womb environment can be controlled for by comparing identical to fraternal twins.
It may be the case that most twin studies are performed badly. I'm just pointing out that most of the issues you raise are not a real problem for anyone who understands basic statistics (admittedly, this may exclude most psychologists).
Yes, I generally agree that it would be theoretically possible to do a separated at birth identical twin study. In practice, the requirements are so stringent that they've never (to my knowledge) been met.
The setup you describe could indeed separate genetic/genetic+environment feedback component from the purely environmental component. Albeit with caveats:
- Identical twins given up for adoption and then separated at birth are probably not drawn at random from the population.
- You can't draw any conclusions at all about which genetic variance is causing the issue, unless you restrict the demographic cross sections. So it can't possibly give you evidence that one race has a higher average IQ, for example.
- The conclusions you draw may be much more about environmental feedback than genetics - maybe with a different environment, the differences will go away. For example, maybe you've found genes that code for susceptibility to iodine deficiency, and your society hasn't discovered iodized salt yet. In that case it's hard to say that you've determined the degree to which intelligence is "genetic" except in the most general sense.
In practice, locating identical twins separated at birth is hard. Finding a representative sample of them across demographic mix is harder. Doing it again for fraternal twins to control for womb environment makes it yet more work. I'm not sure how you control for the fact that the identical twins who are given up and then separated at birth are probably a different population from random adoptees, but I don't doubt with sufficient cleverness it could be done. The problem is that real twin studies almost always fail to control all these factors, and failing to control even one factor dooms your entire experiment.
As long as your twins and twin-like pairs have the same demographic mix, you have an adequate sample.
Proof of this statement? There are mathematical definitions of what size of sample is "adequate" for given effect sizes. (What are the effect sizes of some of the differences in IQ we are talking about here, and how do they compare to error of estimation in the most reliable brands of IQ tests?)
It may be the case that most twin studies are performed badly.
That would definitely be an important matter to look at.
And yet, the odds of a random adoptee growing up in the same city, choosing the same profession, favorite color, and rubber-band wearing habit are astronomically lower than those of an identical twin doing so.
Many twin studies, particularly those done in the past few decades have been rigorous. And the correlation in IQ, career and age of marriage between such twins, are far higher than those of siblings, or adopted siblings growing up in the same house.
There may be a political reason for rejecting the role of heredity or the large body of research supporting it, but there isn't a scientific one.
That's an anecdote, not a study. The criticism I leveled is at twin studies in general, not twin studies .
Heritability is very specific measure, and it does not mean "genetic". Zip codes are highly heritable; number of arms is not heritable at all. I refer you to a great article on the subject of what "heritable" really means in this context:
A single example of twins who are similar proves nothing. I've personally met identical twins who couldn't be more unlike each other - which is no more evidence than the single case (case study, anecdote, they mean the same thing) you cite.
The author works as a professor of statistics, and his work is specifically in the field of uncovering the latent variables which underly human achievement. His arguments seem cogent and clear to me. Just because he comes to different conclusions from you does not mean he "has an axe to grind", which is a purely ad hominem accusation. Where do you think he makes a mistake?
He has admitted that part of his skepticism is due to the fact that he doesn't like the concept of intelligence in a general sense. He is far more critical of claims of heredity than he is of environment. Even half the level of scrutiny he applies to twin studies would invalidate nearly the entirety of sociological research on account of "not controlling for genetics". That is why I say he has an axe to grind.
As for his credentials, he's an associate prof with little to no background in biology, psychology or the study of the brain.
Steven Pinker, on the other hand is a full professor at Harvard (formerly MIT), and is the world's premier on the subject he's writing on.
The only reason number of arms is not heritable is because the variance is close to zero.
Once you include a sample with non-trivial variance (a sample including humans, snakes and spiders), you discover that number of arms is almost 100% heritable.
You're missing the point. Saying something is "heritable" is essentially meaningless. It tells you how that trait would change in the group measured in response to evolutionary selective pressure, and that's it. It doesn't tell you whether the trait is genetic, it doesn't tell you whether the trait is important, and it doesn't tell you whether the trait even exists in a meaningful way.
You can measure quite decent heritability from entirely random traits:
Twin studies are not worthless, but they are limited. If you are trying to figure out why people from the same city and class have differences in trait X, then a twin study will help find that out. But if you are trying to find out why people from different classes and/or locations have a difference in trait X, then a twin study will not really randomize the environment, because both twins will likely be adopted by someone of the same class.
Moreover, studies of broad heritability (a pre-Mendelian concept) merely establish a minimum influence of environment on the trait of interest. More extreme variance in environment, or more specifically directed variance in environment, could still result in greater variance in the trait.
Of course the families of two identical twins are far from independent variables. They're both made of the kinds of people who would adopt a child in the same city at the same point in time. You haven't managed to separate the variables at all!