December 5, 2005

Levitt strikes back!

Levitt strikes back! Steven D. Levitt has responded to the Foote and Goetz paper pointing out two major errors he made in his abortion-cut-crime theory in a particularly striking way -- by introducing, at this late date, a whole new data set!

Foote and Goetz showed that Levitt arrived at his popular conclusion that abortion cut crime due to technical incompetence, and that objective analysis of Levitt's own data shows no impact of legalizing abortion on crime. So, Levitt now introduces a new data set, which he claims provides less noisy data on abortion rates by states, than the one he and Foote-Goetz worked with. Unsurprisingly, Levitt claims this new data set proves he was right all along, even though his original data said, when analyzed correctly, that he was wrong.

Is his new data set really better? Is it worse? Did Levitt botch up his analysis again? Who knows? I'm sure it will take months for objective analysts to look it over.

And is that the last word in data sets? I strongly doubt it. For example, America's most dangerous criminals were performing "selective post-natal abortions" on each other at an unprecedented clip in the gang wars of the 1990s. AIDS was also taking a toll on criminals then. Levitt hasn't adjusted for how many criminals died during this period.

Levitt likes to look at arrests for property crimes, but one of the obvious trends during the youth of the first cohort born after legalization was that property crimes were declining in payoff due to target hardening -- more locks, more alarms, more video surveillance cameras etc. In contrast, drug dealing was a booming market. In 1990, for a criminally inclined 15 year old born in 1975, it didn't make sense for him to set out to learn the craft of the thief when there were already so many experienced older thieves out there all searching for the dwindling set of soft targets to steal. No, the hot business was dealing crack, one in which older criminals weren't well established yet. That's why the homicide rate among black 14-17-year-olds born in the late 1970s was about four times as high as among black 14-17 year-olds born in the late 1960s, before legalization.

Levitt notes that something like 30% of older teens aren't living in the states where they were born. That should raise some questions about state-level analyses, especially when they don't agree with the national level analysis. If movement between states was random, that shouldn't cause too much of a problem for his methodology. But what if crime is a driving force in causing people to move?

What if, say, people in socially liberal states (that had lots of abortion earlier in the 1970s) who worried that their children were be crime-prone tended to move to socially conservative states. And what if people who ware less worried about the crime proneness of their children due to the family's affluence were more likely to move to socially liberal states? That would foul up the state-level analyses fatally if there was a steady net flow of crime prone kids to socially conservative states and a steady net flow of crime-unlikely kids to socially liberals states. I'm not saying it happened, but it might have.

But, from a marketing standpoint, in terms of preserving the value of the Freakonomics brand name, Levitt has put a marker down that his true believers can use to ward off Doubts.

We're now way, way out in how many angels can dance on the head of a pin territory. If Levitt really is explaining close to half of the huge decline in crime that occurred in the 1990s, as he has claimed, the evidence shouldn't be so fragile that it collapses when somebody else stares at it hard and Levitt has to throw away his old data and replace it with a new set of data that nobody has seen before.

When I studied marketing models in MBA school a few eons ago, the professor constantly pointed out that the true test of the statistical analyst is creating robust models. You can always fiddle with historical data and variables until you obtain a high r and a high degree of statistical significance and declare victory. But that's not a robust model and it's not much use in making real world business or policy decisions.

Economist Roehlano Briones writes on his Go Figure blog:

"Extraordinary claims require extraordinary evidence" is a phrase popularized by Carl Sagan - in turn derived from Hume's examination of miracles-claims. Now the original abortion-crime hypothesis is far from alleging a miracle. It is however extraordinary as it implies that causal mechanisms of crime originate from circumstances prevailing at the time of birth. Moreover, the claim that the behavior of eliminating live births is skewed against this causal mechanism (that is, abortion does not neutrally eliminate future crooks and law-abiders on a 50:50 ratio).

The issue remains, as it has since Levitt and I debated in 1999, who should have the Burden of Proof on his shoulders.

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

No comments: