April 14, 2006

The Economist on race-specific medicine

It reviews new evidence that (surprise!) people of different races sometimes tend to have different genes. Then it issues this common caveat:

"This sort of work raises another question, with its own set of sensitivities: how, scientifically, do you define someone's race? In studies such as Dr Cohen's this question is fudged, because people defined themselves. That is a reasonable starting point. But America's population is descended mainly from immigrants and those ancestors have been meeting and mating for centuries. Few native-born Americans trace all of their ancestors back to the same part of the world, so how closely are genes affecting disease linked to genes affecting racial self-identification?"

But that's getting the issue backwards for the purposes of understanding relationships between race and medical conditions. In reality, if researchers were using more accurate measurements of patients' racial backgrounds, they would get even higher correlations between ancestry and various medical statuses. Why? Because if they were using DNA admixture tests to group, say, the Halle Berrys and Henry Louis Gateses in a mulatto category instead of in their self-identified black category, they would be reducing the random noise in their studies, and thus likely increasing the correlations.

You see the same mistake with many critiques of Lynn and Vanhanen's finding of a high (r=0.73) correlation between national average per capita GDP and national average IQ. Critics point to all the noise in the IQ data (e.g., Are the samples really nationally representative? What is the effect of using different IQ tests given by different researchers in different years on different ages?), and triumphantly claim that the noise invalidates the high correlation with GDP. No, actually, the opposite is more likely true. If you had more reliable IQ measures, you'd most likely get an even higher correlation with per capita GDP. You are already getting an awfully high one using just a grab bag of IQ studies.

That would only not be true if there was some biasing factor that made the noise systematically push the scores in one direction based on GDP. That's a possibility, but I haven't been able to imagine one.

Levitt and Donohue similarly argue that the looooong time lag between abortions in a state and then crimes committed a generation later, which causes a huge amount of noise in the data by people moving from state to state, means that the real correlation between abortion and reduction in crimes must be higher than their data indicates. The problem, though, is that there is an obviously biasing interplay among crime, abortion, and interstate migration -- one of the big reasons people move is to find a different moral environment in terms of crime and sexual morality.

I don't know what the net effect is. My guess is that people who feared their children were in danger of becoming criminals tended to flee the high abortion / high crime big cities like New York and Los Angeles, while sophisticates who didn't have to worry about such lower class problems as their kids turning into thugs flocked to the culturally progressive cities from the retrograde hinterlands. The impact of these flows of at risk young people from the high abortion states to the low abortion states, combined with the reverse flow of well-educated, law-abiding people from the low abortion states to the sophisticated high abortion cities thus artificially made it look like high abortion rates reduced crime a generation later. But that's just a plausible guess. Levitt and Donohue are certainly making a heroic assumption that all the effects net out to zero. But, more likely, they haven't even thought systematically about this problem.

By the way, ABC's "20/20" will devote an entire hour to Levitt and Freakonomics on Friday night, 4/14 at 10pm Eastern time. (Levitt and Dubner are under contract to ABC so ABC is, surprise, hyping their own hired talent. On the other hand, 20/20's John Stossel is a cut above most everybody else on TV in brains and honesty, so it might not be the usual complete corporate conflict of interest travesty.) You can watch the webcast preview here. Economist Ted Joyce gets about 7 seconds on the preview to say the data don't support Levitt's abortion-cut-crime theory. When Stossel's people emailed me, I recommended they interview blogger La Shawn Barber. I don't know if they did. Here's La Shawn's run-in with Levitt.

Meanwhile on his Freakonomics blog, Levitt responds to my recent posting about his shortcomings with this charmingly entitled post: "I will see you in hell, apparently." Poor Levitt ... it's not enough for him that he's turned himself into a brand name celebrity and zillions of people idolize him. The mere existence of one vocal skeptic is driving him to do irrational things like angrily respond to my criticism, when the smart thing for him to do would be to not publicize my itemizations of his theory's failings. He's got all mass media momentum on his side, but perhaps the failure of his abortion-crime theory to make much progress in persuading his fellow economists, as amply displayed at last month's American Enterprise Institute colloquium on it, is irking him.

My published articles are archived at iSteve.com -- Steve Sailer

No comments: