19 Comments
User's avatar
Steve Sailer's avatar

My vague hunch would be that identical twins reared apart tend to often both be raised about the 25th percentile of income. Say, identical twins are orphaned because the parents are killed in a car wreck or the parents both become drug addicts or whatever. So the extended family, maybe both sets of in-laws, gets together and says, "OK, I know this will be a major sacrifice, but we've got to dig deep and figure out who can afford to raise the two kids."

But nobody has the money to take both, so one couple takes one twin and another couple takes the other.

Expand full comment
Steve Sailer's avatar

So, there don't seem to be too many Prince and Pauper-style examples of identical twins being raised apart at different ends of the social scale, because if one relative is really rich, she'd probably take both twins.

Expand full comment
The Critical Citizen's avatar

What’s the correlation for the mother to child and father to child? Is one larger than the other?

Expand full comment
mattysoup's avatar

Sorry if dumb question but, would the same person/same genetic theoretical individual, one raised poor and one raised wealthy, score identically on IQ tests?

Expand full comment
Sebastian Jensen's avatar

Depends on the extent. Poor, as in Africa tier? No. Poor, by America’s standards, probably.

Expand full comment
mattysoup's avatar

Ah so insofar as poverty may lead to things like malnourishment during childhood it negatively impacts IQ?

Expand full comment
Sebastian Jensen's avatar

It’s a controversial topic, but I support the view that malnutrition does lead to lower IQ as an adult, though empirical tests of the hypothesis have been disappointing.

Expand full comment
Henry Rodger Beck's avatar

Illiteracy is probably a big one, too. The harder it is for one to read, the harder it is to maintain a large vocabulary, or review information. The smaller the vocabulary, the harder it is to sharpen one's thinking, understand abstract concepts, communicate ideas, or solve problems.

But I don't see further educational literacy efforts making much difference. There's still only so much you can do when someone's in the 80s and below, so I really do think gene therapy will be the only truly effective means to solve the Third World's low-average-I.Q. problem.

Expand full comment
Johannes MK's avatar

The only problem with this looking at identical twins reared apart is there is still a womb effect. How much of our intelligence is actually shaped by the conditions of the mother's womb having an effect on the brain?

Expand full comment
TonyZa's avatar

Embryo splitting is done regularly in cows and creates identical twins. In theory one could split a human embryo, implant the resulting identical twin embryo in different moms and compare.

A lot of effort for a study unlike to pass the IRB and given the results above I doubt the womb effect is a big thing unless moms smoke crack.

Expand full comment
Drunk Wisconsin's avatar

Do we know to what extent the heritability varies? Is it additive? For example, if the dad is high IQ and the mom is low IQ, are their offspring likely to average out in the middle or inherit either high or low IQ? If both parents are high IQ, do their kids have an increased chance of having an even higher IQ than either parent?

Expand full comment
TonyZa's avatar

"If both parents are high IQ, do their kids have an increased chance of having an even higher IQ than either parent?"

No. On the contrary the kids are more likely to regress closer to the mean of their ethnic group. But with luck or enough kids it is likely that some will be higher IQ than their parents.

Expand full comment
Hyperdupont's avatar

For the less knowledgeable among your readers: I suppose that test reliability is estimated b the correlation when you test the same person twice. Correct?

Expand full comment
Sebastian Jensen's avatar

Internal consistency. Usually with cronbach’s alpha or (better), the omega reliability statistic.

Expand full comment
Headless Marbles's avatar

I found some more context on Wikipedia, because this also interested me, being a psychometrics-outsider, as it were. I believe the idea that Hyperdupont refers to would be like using "parallel tests" to estimate reliability ("parallel" because you wouldn't actually want to give the same person the *exact same* test twice with the same questions on it, because they'll remember them; it would have to be a "parallel test" with different questions but the same score expectation and variance).

The Wikipedia article on Classical Test Theory elaborates:

> Using parallel tests to estimate reliability is cumbersome because parallel tests are very hard to come by. In practice the method is rarely used. Instead, researchers use a measure of internal consistency known as Cronbach's α.

> Cronbach's α can be shown to provide a lower bound for reliability under rather mild assumptions. [...] Thus, this method is empirically feasible and, as a result, it is very popular among researchers.

https://en.wikipedia.org/wiki/Classical_test_theory#Evaluating_tests_and_scores:_Reliability

Expand full comment
Headless Marbles's avatar

And when you refer to reliability-adjusted correlation, is that like an errors-in-variables adjustment where you use the reliabilities to estimate the measurement error in the predictors?

It’s a little unsatisfying to just assume the .91 reliabilities for so many of the tests, but at least a decent chunk of your N has empirical reliability estimates.

Expand full comment
Sebastian Jensen's avatar

>And when you refer to reliability-adjusted correlation, is that like an errors-in-variables adjustment where you use the reliabilities to estimate the measurement error in the predictors?

No.

Expand full comment
Headless Marbles's avatar

My apologies if I've provoked your curtness somehow :(

I was just asking if this:

> The mean correlation is 0.7, mean reliability-adjusted correlation is 0.77...

... referred to disattenuation, i.e., correcting for the dilution that results from considering the individual-level IQ measurements to be noisy variables themselves, enabled by the reliability estimates, which tell us something about that noisiness.

Having now spent some more time reading about classical test theory, that seems to be exactly what it must mean? So I'm not sure what your "No" is in reference to, exactly.

My background is closer to econometrics, you could say, and I find that the various applied fields (psychometrics, econometrics, etc.) tend to organize and relabel the statistical concepts/literature in their own somewhat idiosyncratic ways, and I find it useful for understanding to call out the connections. As in, "ah this is another instance of the errors-in-variables problem, and here's how psychometricians talk about it."

Expand full comment