Tuesday, August 22, 2006

Book Review: Freakonomics

Steven D. Levitt and Stephen J. Dubner, Freakonomics (B+)

I am somewhat torn about this book. On one hand, I'm always in love with the idea of debunking stupidly held conventional wisdom. I've opined before that "correlation does imply causation" is a valuable lesson for anyone, but for those who shape opinion or craft public policy, it should chanted hourly as a mantra, warding off the evil spirits of rushed judgement and horrible, expensive legislation. The authors stress this distinction early and often and I like that. On the other hand, when someone starts strutting their certitude around like a smug little rooster, my big temptation is to knock them down. Levitt and Dubner ultimately appeal more to the former sentiment than the latter, but their admonition to always question the data applies to their data too.

Well, it applies to Levitt's. Dubner seems to be the prose stylist of the pair, and does a good job of reinterpreting Levitt's technical work with accessible and humorous style for the brighter-than-average everyman. (Conversational science reporting is a style category that is every bit as hackneyed and necessary as noir, technical writing, pulp, whatever. It succeeds when it's effective, but many of these nonfiction books are told in interchangeable voices.) I could have done without opening each chapter with a fawning editorial about Stephen Levitt.

An expert demanding distrust of experts, Levitt sets a high bar for himself. Though he presents them as off-the-wall, his ideas tend to be more interesting than controversial. He goes on a bit that people will cheat with the proper incentives, and that you can't fake your values as parents. Neither is surprising. He also describes how economic achievement correlates with, but isn't caused by, race. I thought that was the conventional wisdom, but I was pleased to read a discussion of data which bear that out.

Probably his most contentious essay is the one in which he suggests that legalizing of abortion in the 70s was a major contributor to the drop in violent crime in the 90s. He draws an interesting correlation between these data sets, advances an interesting theory, and provides a reasonable counterexample (abortion instantly made illegal in communist Romania) that supports his idea of causality. But he also rolls over some of monster holes. He dismisses incarceration rate as accounting for only a third of the drop, and that crime rates have a 1:1 linearity with unemployment rates, also ruling out an improving economy as a factor for major crime reduction. It's brazen to weight things this precisely when you're doing epidemiology, and worse, Levitt provides no citation while assuring us that "studies show…" He likewise glosses over the causes of the increased crime rates in the 1960s in the first place (I suspect that more inclusive definitions of crime, especially drug crime, was a big factor), and he fails to address whether the lower socioeconomic sectors are more likely to have abortions (I expect it more appeals to people who are comfortable going to doctors).

It would have been a ballsier enterprise for Levitt to correlate crime rates with national opinions about the country say, from these ubiquitous right course/wrong course surveys. Certainly that was tanking in 1968, and, despite what Newt said, topping out in 1999. He does hint that white collar crime rates appear to change with current events.

But all that said, I accept that legalized abortion may well be a contributing factor to reducing crime, and I expect that Levitt has over-emphasized it for the purposes of being clever. That's one problem with iconoclasts. The other problem with iconoclasts is that they can be every bit as full of shit as the spewers of conventional wisdom. Just because someone's a smartass, doesn't necessarily mean they're smart, and Levitt's advice to look at the numbers is well-advised to anyone who tries to make opinions.

Keifus

No comments: