May 8, 2005

Levitt caught embroidering truth:

From a U. of Chicago Graduate School of Business News report on a talk given by Freakonomics author Steven D. Levitt:

His research tells him that 35 percent of the “incredible drop in crime” in the early 1990s was due to abortion legalization.

While conceding that some people aren’t completely convinced of his findings, “I think it’s pretty compelling,” Levitt told students in the Milton Friedman Group on April 25 at the Hyde Park Center.

Levitt said he based his theory on two pieces: unwanted, unloved children are at highest risk for crime; and fewer unwanted children—a million a year—came into existence because of abortion’s legalization.

Levitt needs to bamboozle the public about the size of the reduction in unwanted births because it's central to his theory's appeal. Yet, it's also imprudent of him to blatantly mislead the public like this because we can quote him on what really happened: "Conceptions rose by nearly 30 percent, but births actually fell by 6 percent …" Most legal abortions, as Levitt has admitted, are pure waste -- a fetus who never would have been conceived without legalization gets aborted.

The peak number of abortions per year was 1.6 million. The number of births was running at about 3.7 million before legalization of abortion. This would suggest a reduction in the number of births due to abortion of around a quarter of a million. Other methodologies might raise that to a half million, but Levitt knows very well that it wasn't a million. Levitt himself told an interviewer regarding the impact of legalization:

“One in four of the pregnancies which took place were just because people were lazy,” he says. “That’s a lot. That’s a lot of abortions.”

The article on his U of C talk continues:

He tested his theory on the data and it fit. In addition, states that legalized abortion three years earlier than Roe v. Wade saw their crime rates dramatically dip three years earlier than other states.

Notice how Levitt tries to skate by the fact that the early-liberalizing metropolises are also precisely where the serious violent crime rate first went up. As Levitt acknowledged to me in 1999, “[T]he high abortion places like New York and California tended to have a bigger crack problem, and tended to have crack arrive earlier.

Levitt seems to think longer lag periods are more trustworthy. If NYC and LA legalized abortion in 1970 and the juvenile murder rate went up in those places 17 years later and then fell there 22 or so years later, Levitt wants us to believe that we should trust him that legalization's effect was not felt 17 years after 1970, but was felt 22 years after 1970. In Levitt's Private Universe, the longer the lag in years between a potential cause and its hypothesized effect, the more trustworthy the connection is!

But what about the spike in crime among young black males in the late 1980s and early 1990s, which critics offer as a puzzle piece that doesn’t neatly fit Levitt’s solution?

“The answer, to me it seems obvious,” Levitt said. “It’s crack cocaine.”

You can see how Levitt tries to have his cake and eat it too. The enormous rise in murder and serious violent crime rates among precisely the group most culled by the legalization of abortion -- young urban blacks -- is assumed away as having nothing to do with abortion. It's purely the result of an exogenous event, the rise of the crack wars. Okay, but then, in contrast, the fall in serious violent crime rates is assumed to be the result of abortion, not the decline of the crack wars.

Levitt, when given a choice between two sets of data, consistently uses the fuzzier, more uncertain evidence to justify his theory and pointedly ignores the more straightforward, more precise comparison. For example, he likes to look at data for two groups of criminals, over and under age 25 and make surmises about the effects of abortion legalization that can't be prudently drawn out of such coarse categories. In contrast, he hates to look at data for sharply defined groups, such as 14 to 17 year olds, where you can actually focus on individuals born before or after legalization.

Similarly, he prefers to look at the change in the crime rate for two points twelve years apart, one before and one after legalization, but hates to look at changes, say, six years apart. Levitt's philosophy seems to be that the more intervening years and intervening events, the clearer the reading you can get on the effects of legalization!

For example, the article goes on:

Crime statistics show that crack cocaine use hit young black males especially hard in their teenage years but didn’t translate into an increase in the amount of crime committed over the course of their whole lives, he said, thus accounting for the spike.

But, obviously, this is an apples to oranges comparison. By their late 20s, the most dangerous members of the cohort of black urban males who were born soon after legalization were, more than any other cohort in American history, confined to prisons, wheelchairs, or coffins, where their ability to commit more crimes was limited, at best. That in their late 20s, this highly-culled cohort was still committing serious crimes at an average rate after all this depletion of their most violent members says something about what they were like before all this post-natal culling of their most criminally-inclined members. Instead, Levitt wants us to focus on the pre-natal culling a quarter of a century before.


Colby Cosh points out that Levitt's abortion-cut-crime theory triggered even comedian Jon Stewart's BS detector on The Daily Show:

Alex Tabarrok--an American economist whose acquaintance overlaps slightly with mine--recently watched a fascinating exchange between overexposed freakonomist Steven Levitt and overexposed fakejournalist Jon Stewart on The Daily Show.

...Levitt said that in estimating the effect of abortion on crime he controlled for other variables like police and prisons. Jon Stewart pressed Steve for an explanation of how someone could "control" for other variables--amazingly, Stewart seemed genuinely interested in an answer but, wisely, Steve demurred...

Stewart's question reveals both a gap in his education and a laudable ability to spot legerdemain (which is a pretty good cement for such educational gaps). It might be helpful, in fact, if the public at large knew that the phenomenon of "experimental control" covers many orders of rigor...

This matter of experimental control is a pretty deep pool. I would be the last person to advise us to dispense with inferential "controls" in the squishier sciences. But Jon Stewart is probably quite wise to dig in when someone tries to skate with him across that pool, arm-in-arm.

You can read the rest of Colby's explanation of why economists so often develop too much confidence in their non-robust statistical models here.

My published articles are archived at -- Steve Sailer

No comments: