September 30, 2005

Levitt on Bennett

On his Freakonomics blog, economist Steven D. Levitt, the main promoter of the old theory that legalizing abortion cut crime, writes:

2) Race is not an important part of the abortion-crime argument that John Donohue and I have made in academic papers and that Dubner and I discuss in Freakonomics.

C'mon, Steve, try being frank about your abortion-crime theory for once. Your widely circulated draft paper in 1999 argued that one reason abortion should have cut crime is because blacks, per capita, have more abortions and commit more crimes. (See this NYT story from 1999 for the details). You dropped that reference later to stay out of trouble.

It is true that, on average, crime involvement in the U.S. is higher among blacks than whites. Importantly, however, once you control for income, the likelihood of growing up in a female-headed household, having a teenage mother, and how urban the environment is, the importance of race disappears for all crimes except homicide. (The homicide gap is partly explained by crack markets).

Oh, boy ... where to begin?

- "Except homicide"? -- That reminds me of the old joke: "Except for that, Mrs. Lincoln, how did you enjoy the play?"

- The homicide gap existed long before crack was invented in the 1980s. Back in 1976, the first year of separate black and white data in the Bureau of Justice Statistics website, the black homicide rate per capita was 9.5 times the white rate.

- Further, I'm highly dubious that the racial aspect of homicide is all that different. In the important new report "The Color of Crime, 2005," the incarceration rate is broken out for a variety of crimes by ethnicity. (The authors split out Hispanics from whites, unlike many of the other crime statistics, which makes them more accurate and the black to white ratio even higher because Hispanics are imprisoned, overall, 2.9 times more per capita than non-Hispanic whites.)

Judging from the graph in Fig. 9 of "The Color of Crime, 2005," (I don't have the exact data), the black to non-Hispanic white ratio for incarceration for murder is 8.3 to 1. But for all crimes overall, blacks are imprisoned 7.2 times more often than whites, so the difference isn't that great. And blacks are incarcerated 33 times more than Asian-Americans!

For robbery, the black-white ratio looks like about 14.8 to 1, or nearly 2 to 1 over the homicide rate.

Aggravated assault looks like about 7.3 to 1. Other violent crimes are lower (rape is about 3.0 to 1), but the overall violent crime incarceration ratio is about 7.1 to 1, not too different from the homicide ratio.

Strikingly, the non-violent incarceration ratio is just as bad, also in the 7 to 1 rate. This is driven in part by drug offenses, which are in the 12.5 to 1 area. But, blacks are incarcerated for non-drug property crimes about 5.1 times the white rate.

Blacks even get themselves thrown in jail for white collar crimes disproportionately: 4.0 times more often for fraud, 5.1 times more often for "Bribery / Conflict of Interest," 3.2 times for racketeering, and even 2.9 times more often for embezzlement. I suspect you'd have to go all the way to high end white collar crimes like anti-trust violations and insider trading to find ones where whites have higher per capita rates.

In other words, for most crimes a white person and a black person who grow up next door to each other with similar incomes and the same family structure would be predicted to have the same crime involvement. Empirically, what matters is the fact that abortions are disproportionately used on unwanted pregnancies, and disproportionately by teenage women and single women.

First of all, for the purpose of discussing whether or not the Levitt Effect of abortion driving down the crime rate works in part by aborting more black fetuses per capita than white fetuses, these kinds of attempts at "underlying explanations" are largely irrelevant. (The real objection to the Levitt Effect is that, judging from the historical record, it didn't work at all.)

Second, I find this highly dubious. Levitt doesn't cite any research supporting this. And even if he did, I've found that when I go an read his reports, his track record for veracity in his claims that prior research supports his views is dubious.

Third, this is the kind of thing "explaining away" that gives correlation analysis such a bad odor with the public. As Colby Cosh pointed out, on the "Daily Show," John Stewart rightly grilled Levitt on exactly how you "control for" other variables, and Levitt couldn't come up with a trustworthy answer.

You can make all sorts of things disappear by "controlling" for variables that are closer to symptoms than causes. For instance, you can make the average height gap between the Dutch and the Japanese disappear by "controlling for" inseam length of the pants hanging in their closets.

What Levittt has done is pick three variables that currently correlate closely with being black and used them as a proxy for blackness. This is the opposite of Occam's Razor, which says you ought to be biased in favor of the fewest number of explanatory variables.

Fourth, Levitt's three variables sound extremely dubious historically. Think about that 9.5 to 1 difference in homicide rates between whites and blacks back in 1976. Most of those killers in 1976 were born in the 1940s and 1950s, when over 80% of black children were being born to married women. And during the Baby Boom, lots and lots of white babies were being born to teenage mothers.

Yet, the homicide rate went shooting up in 1965, just when the illegitimacy rate went shooting up too. We didn't have to wait a generation to get the effects of rising illegitimacy on crime, we saw them instantly. That's because a major effect of society deciding to allow sex without marriage is on the young men who now don't need to get a job so they can get married so they can have sex. They can hang around, do a few crimes, and still have a girlfriend.

In summary, Levitt is one slippery operator.

My published articles are archived at -- Steve Sailer

No comments: