Monday, September 08, 2003

IQ, Truth and Logic

It seems that at least one member of the right is in a bit of a panic over IQ. That intelligence is mostly hereditary and that science has established this as fact, has been routinely asserted by various social conservatives for over a century. To them, ideology must win, not evidence. They have developed many standard strategies to attempt to discredit experimental findings that challenge their position and their responses to the latest such study, reported by Rick Weiss in the Washington Post has provided another opportunity for them to put them into practice. The study was conducted by Eric Turkheimer, a psychologist at the University of Virginia with several colleagues.

The Misanthropyst got in quickly, with a heartfelt plea on behalf of People for the Ethical Treatment of Data:

Here's an example of having your cake and eating it, too - the unfortunate data must have been screaming in pain at the abuse it was put through to achieve these politically correct conclusions:

Back-to-school pop quiz: Why do poor children, and especially black poor children, score lower on average than their middle-class and white counterparts on IQ tests and other measures of cognitive performance?...

...Now a groundbreaking study of the interaction among genes, environment and IQ finds that the influence of genes on intelligence is dependent on class. Genes do explain the vast majority of IQ differences among children in wealthier families, the new work shows. But environmental factors -- not genetic deficits -- explain IQ differences among poor minorities...

Following on, Ozblogistan's very own John Ray is also skeptical about the study and has done his own analysis of the study. He begins with a well worn standard: misconstrue the study's conclusion, to make it appear as absurd as possible:

... It claims that heredity is the main factor in determining white IQ but is NOT the main factor in determing black IQ! So are blacks a different species from whites, then? That would seem to be the conclusion if we took the study seriously.

Ray presumably bases this caricature on this paragraph from Weiss' report of the

Now a groundbreaking study of the interaction among genes, environment and IQ finds that the influence of genes on intelligence is dependent on class. Genes do explain the vast majority of IQ differences among children in wealthier families, the new work shows. But environmental factors - not genetic deficits - explain IQ differences among poor minorities.

Ray goes on to challenge the study's methodology, despite the fact that:

Full publication of the study has not been done as yet but from what we know so far it seems that what they found was in fact much simpler than that.

This leads on to a passage I find delightfully ironic: a discussion of statistical theory that leads to the conclusion that the finding was a statistical artifact, similar to regression to the mean. In this case, the artifact in question is the "restriction of range effect" which Ray describes thus:

... if you take ANY group and select out a subset that is relatively homogeneous with regard to some variable, differences in that variable will tend to have less importance in explaining other differences. Since socioeconomic status and race are substantially correlated with heritable IQ, that is precisely what these researchers have done: Selected a group that is relatively homogeneous in genetic inheritance for IQ and then said: "Hey! Differences in genetic inheritance are not so important here!" [my emphasis]

Just as I did after reading Ray's use of regression to the mean to argue that, over time, you could expect blacks to breed children dumber than their parents, I did a quick Google for the restriction of range effect. There's a brief explanation here. In general, it agrees with Ray's description of the effect, however it is worth noting that, like regression to the mean, the restriction of range effect occurs as an artefact when calculating correlation co-efficients for correlation of two variables. If you calculate a correlation coefficient over a whole data set, then select a sub-range of the set, you will find a much lower correlation. Ray's major argument against the validity of the study is that this is what the researchers have done. How he can reach this conclusion without having seen the full report of the study eludes me.

Ray's insistence that socioeconomic status and race are substantially correlated with heritable IQ seem to me to ignore the standard warning, given by most lecturers in statistics when introducing the subject of correlation for the first time: correlation is not causation. I'll repeat that: correlation is not causation. And, for the third and final time: correlation is not causation.

There's a standard example used to illustrate this point; you'll find it all over the internet with a bit of imaginative Googling. It's the strong correlation between monthly sales of ice-cream and the rate of deaths by drowning. The two correlate very strongly but it's pretty obvious that there's no causal relation between the two. Something else - something very obvious - is going on here.

The case of the correlations between IQ and socioeconomic status and IQ and race are similar; they give no grounds for concluding that IQ determines social position or vice versa. All you can say, in the absence of other evidence, is that the two correlate and ain't that interesting. The race-IQ-status correlation is often interpreted as indicating that race determines IQ which determines social status, but it would be equally valid, on the basis of the correlation alone to assert that IQ determines both race and social status or that status determines IQ and race or ... There are a few too many permutations to enumerate fully.

Ray's position (stated at the beginning of his post) is that IQ is mainly hereditary, which leads to the logical error of circular argument. He introduces into his argument, as a bald assertion, the very conclusion he is trying to establish; that is that IQ determines socioeconomic status and not vice versa. Stripped of the statistical theory window dressing, Ray's argument is:

I: If IQ is hereditary then the new study is wrong.
II: IQ is hereditary.
III: The new study is wrong.

This is an eminently logical classic syllogism (modus ponens) but the conclusion does not follow if either of the two premises is false. Ray's minor premise is in fact the very assumption that the new study challenges, so he is on very thin ice in asserting its truth as a refutation of the study's findings. An equally logical argument is the following modus tollens:

I: If IQ is hereditary the the new study is wrong.
II: The new study is not wrong.
III: IQ is not hereditary.

The only conclusion that logic can reach in this situation is that either IQ is hereditary or the new study is wrong. Which leaves us right where we started. Weiss' report, although admirably detailed and clear doesn't give us enough information to determine this question. The only way to demonstrate that the study reported by Weiss is wrong, or flawed, is to look at the study design, methods, data and data analysis and assess whether they support the conclusions reached.

I may be a little rusty on this - it's over twenty years since I got my scientist's licence and I've never used it professionally, but unless the scientific method has changed radically over the past two decades, I'm pretty confident that if you want to show that a scientist has conducted a flawed experiment, that's how you do it.

None of this has prevented Ray's post from receiving approbation elsewhere: here, here and here, where "That Bitch" offers some unintentionally ironic advice:

No matter what your political or social leanings are, I challenge you to take a look at your favorite set of statistics. Go digging for the devil in the details, and find out whether or not your "statistics" are nothing more than someone elses warped view being presented as "The Truth." I think that 99% of the time, you'll be unpleasantly surprised.

Dan at Tubagooba has also offered a little seriously intended advice, suggesting that John Ray might take a look at Stephen Jay Gould's The Mismeasurement of Man. Offering reading suggestions for John Ray to ignore is one of those blogospheric rites of passage; Dan is now up there with Rob Schaap to name just one blogger who has completed this initiation.

Update: As if I needed reminding of the futility of this kind of post, John Ray has taken the opportunity to revise Update: As a further reminder of the futility of posts such as these, John ray has taken the opportunity to summarise his critique of the Turkheimer et al survey here:

I showed that IF IQ is generally hereditary and related to race and class, [then] you would still get the survey results reported -- so therefore the survey results reported do not upset the claim that IQ is generally hereditary and related to race and class!

Give or take an unstated premise, this is an impeccable piece of logical reasoning, whose conclusion follows with the same rigor as the conclusion (III) of the following syllogism.

I If Gummo Trotsky doesn't understand simple "if ... then" logic then he shouldn't argue with John Ray.
II Gummo Trotsky shouldn't argue with John Ray.
III Gummo Trotsky doesn't understand simple "if ... then" logic.

No comments: