There’s a really interesting article in Slate by Steven D. Levitt and Stephen J. Dubner (the authors of Freakonomics) about female births and heptatitis B. The disproportionate number of male births in some Asian countries has been attributed to causes such as selective abortion and infanticide. But, as explained in the paper “Hepatitis B and the Case of the Missing Women”, by Harvard Economics graduate student Emily Oster, Hepatitis B infection rates actually explain a lot of the discrepancy. Pregnant women who have Hepatitis B are more likely to bear sons than daughters, and Hepatitis B is more common in those parts of the world where the proportion of male births is so high. Pretty cool.
Again, though, the reason I’m writing about the article doesn’t have much to do with its subject matter. What struck me more than anything were the article’s opening sentences:
“What is economics, anyway? It’s not so much a subject matter as a sort of tool kit–one that, when set loose on a thicket of information, can determine the effect of any given factor.”
Whoa! That’s a really bold statement! I won’t fuss about whether what’s in the tool kit is actually economics or statistics–methods like regression can be shared by many different fields, and there’s enough SAS for everyone. What’s more troubling is the idea that most any question can be answered by throwing enough regression-type analyses at a large data set. Regression isn’t magic, and causal inference is hard; while most statisticians and economists know that (and I’m not accusing Steven Levitt of not knowing that), we could maybe do a better job of convincing the rest of the world.
In an interesting twist, the New York Times review of Freakonomics ends by praising the work of Amartya Sen, whose work Levitt and Dubner use to introduce the lopsided sex ratio in Asia.
I remember reading somewhere that Levitt is a breed apart from "greek letter" economists. Perhaps this says more about his unconventional approach than anything else, but could be misconstrued.
Econ and Stats can be described similarly in other ways. No other mathematical dicipline (except maybe string theory) suffers from such differences in opinion umong experts.
I think you've mistaken what Levitt was trying to say. The economist's toolkit Levitt is referring to is much larger than regression, or indeed 'statistical techniques generally. Indeed, its primarily a way of formally generating and setting out theories to test – maximisation subject to constraint, the role of rationality, etc.
My own view is that this toolkit is a bit more limited than Levitt thinks, but he's far from the naive data miner you imply.
Deb Nolan and I mentioned this example in our book but we didn't know about the hepatitis story. We wrote, "Demographers think the discrepancy in girl births is due to infanticide…"
I think you are correct, and if one were to actually look dispassionately at Levitt's work separated from the surrounding hype, he has done an awful lot of this sort of thing over time. Even the famous abortion/crime paper "controls" for the crack wars in a very superficial way.