See here (if you care).
P.S. Kaus writes that, when he was on William Bennett’s radio show, “Bennett immediately zeroed in on a key political mystery: Are African-American voters on board with the Democrats’ recent amnesty-for-illegal-immigrants program?” I wonder if Kaus asked Bennett about this quote:
But I [Bennett] do know that it’s true that if you wanted to reduce crime, you could — if that were your sole purpose, you could abort every black baby in this country, and your crime rate would go down.
As Brendan Nyhan notes, Bennett wasn’t actually suggesting that black babies be aborted–in fact, Bennett said, “That would be an impossible, ridiculous, and morally reprehensible thing to do, but your crime rate would go down.” Bennett definitely sounds like the go-to guy for a savvy discussion of the black vote!
P.P.S. Just to clarify for those who might think that Bennett was simply calling-it-like-it-is, albeit in a politically incorrect style . . . On his Freakonomics blog, Steven Levitt supported Bennett’s reasoning, as follows:
If we lived in a world in which the government chose who gets to reproduce, then Bennett would be correct in saying that “you could abort every black baby in this country, and your crime rate would go down.” Of course, it would also be true that if we aborted every white, Asian, male, Republican, and Democratic baby in that world, crime would also fall. . . .
As John DiNardo points out, Levitt seems to be confusing the number of crimes with the crime rate. The latter has a denominator. Beyond this, the intervention being hypothesized would certainly have many effects, and it is highly doubtful that the result would be to leave the crime rate among non-blacks unchanged. This is pretty basic causal reasoning, although given what I’ve heard about ed schools (sorry, Jennifer!), I guess it’s not completely unsurprising that a former Secretary of Education could get confused on the matter.
Perhaps (to return to a familiar Kaus theme), it’s a problem with the teacher’s unions? With a more dynamic education system, less bound by bureaucratic constraints, we’d surely be appointing cabinet-level education who had a better understanding of causal inference.
Andrew, there seems to be some sort of serious editing problem with this post, the same sets of paragraphs appears multiple times… I think you want to have a look at it and straighten it out.
Damn cut-and-paste. This must be what happened to Doris Kearns Goodwin . . .
Anyway, it's fixed now. Thanks.
I'm not a fan of Bennett, but this isn't really a fair analysis of his comment. His core point, that policies are not desirable simply because they reduce crime, is an obvious one. Whether or not his particular example holds water, it works well enough as a ceteris paribus argument to make his point. I don't think this reflects at all on his understanding of causal inference.
Andrew J.:
Maybe Bennett's statement is fine by the standards of radio talk show hosts.
But to the extent that Bennett is considered an intellectual, author of books and former Secretary of Education, then, no, no way.
Bennett flat-out says, "I do know that it's true that if you wanted to reduce crime . . . you could abort every black baby in this country, and your crime rate would go down." How does he know this? Or, should I say, how does he "know" this? The simple answer is, he doesn't. He's claiming to know the effect of a hypothetical intervention. The basics of causal inference would tell him (and, for that matter, Steven Levitt) that this is wrong. As I wrote above, the intervention being hypothesized would certainly have many effects, and it is highly doubtful that the result would be to leave the crime rate among non-blacks unchanged.
If Bennett said, "Suppose we could lower the crime rate by aborting only blacks, still I don't think that would be a good idea," then, fine. But that's not what he said. He said he knew what would happen. Again, this might be fine by the standards of Larry King, but I'd hope to see better from a former secretary of education.
It's just a good thing Bennett doesn't gamble. This kind of lack of understanding of statistical principles could translate into big losses!
Andrew,
Huh? It seems to be that Bennett using the Rubin Causal Model in a perfectly defensible fashion. (He takes Steven Levitt's work on the effects of abortion as given.)
Causal outcome: crime rate in the US.
Treatment: Aborting all black fetuses.
Causal effect: Crime rate lower with race-based abortion.
He (like Levitt!) assumes SUTVA. If you believe Levitt's analysis (which I, at least, don't), then Bennett's hypothetical demonstrates a fine "understanding of causal inference."
Now, you could be making a subtle point that SUTVA is unlikely to hold in this case, but is it any less likely to hold with Levitt's work? Or with any sort of public policy example?
If your point is that Bennett shows as much of an "understanding of causal inference" as Levitt, then that might be a reasonable point. But where is the evidence that he understands it less?
Andrew J. – you also do have to ask yourself why a communication and political professional like Bennett decided to use that particular example – and the answer is, of course, that it plays nicely into "black people are criminals" narrative that many of his viewers undoubtedly believe.
And this is distasteful on yet another level – there is a (mostly 19th century) history of eugenics thinking (and partial practice) directed against blacks in the US. This isn't super present in the white consciousness, but it's quite present for blacks, especially black intellectuals. And so "jokes" or "examples" about eugenics against blacks aren't received much more kindly by many blacks than "jokes" or "examples" using gas chambers are by many jews.
I'd guess that's where Toni Morrison was coming from, too.
David:
I know of no reasonable version of the Rubin causal model in which Bennett, or anybody else, would "know" the outcome of a hypothetical treatment that has never been tried. In your terms, yes, I think Sutva would be violated. But really it goes beyond that: a proposal such as "aborting every black baby" is so much of an extrapolation to be beyond any reasonable claim of certainty. Rubin knew about this, which is why he's spent so much of his career thinking about matching and other methods of avoiding undue model-based extrapolation.
I agree with you that in this case Bennett is showing as much understanding of causal inference as Levitt. I'm assuming this was a momentary slip on Levitt's part. I mean, in this case Levitt didn't even catch the distinction between a count and a rate, and I know he understands this..
All joking aside, my impression is that Bennett treats his radio show and Levitt treats his Freakonomics franchise as a way to blow off steam and get attention, not as comparable to their more serious intellectual endeavors.
I'd bet that both Bennett and Levitt would understand causal inference just fine if it were presented to them in the form of a textbook problem, but given the above evidence, I don't think they feel causal inference in their bones the way a statistician might. No big deal, especially if you consider radio and blogs to be throwaway media.
I expect that if you probed Bennett on his black-babies-and-crime statement, he'd say that he doesn't really care if his causal statement is true because nobody's going to be doing the treatment anyway. As a statistician, though, I'm very sensitive to extreme and inappropriate statements of certainty. The sort of casual thinking that will lead Bennett to "know" the effects of aborting black babies also seems related to him "knowing" the effect of various education policies, drug policies, etc.
So your complaint is that during a conversation that aired live he used the phrase "I know" rather than "I'm reasonably certain"?
I don't see how he could be "reasonably certain." The whole thing is ridiculous.
In his 2001 academic paper with John J. Donohue, Steven "Freakonomics" Levitt wrote:
"Fertility declines for black women are three times greater than for whites (12 percent compared to 4 percent). Given that homicide rates of black youths are roughly nine times higher than those of white youths, racial differences in the fertility effects of abortion are likely to translate into greater homicide reductions. Under the assumption that those black and white births eliminated by legalized abortion would have experienced the average criminal propensities of their respective races, then the predicted reduction in homicide is 8.9 percent. In other words, taking into account differential abortion rates by race raises the predicted impact of abortion legalization on homicide from 5.4 percent to 8.9 percent."
In other words, race accounts for 39% of the putative Levitt Effect on supposedly reducing homicides.
Personally, as I pointed out to Levitt in Slate in 1999, the historical record suggests that legalizing abortion didn't reduce the homicide rate a generation later, but he went on arguing that (just forgetting to mention the race part) and became a celebrity with his 2005 bestseller.
Dear Andrew:
If you wanted to study this question empirically, you could look at urban areas that have changed demographically over time and see the impact on the relative homicide rates.
For example, Compton was famously black 20 years ago in the NWA "Straight Outta Compton" era, but LA Times today says the student body of Compton H.S. is now 76% Latino. What has happened to the homicide rate in Compton over the last 20 years relative to areas that didn't undergo massive demographic change?
Similarly, there was massive demographic change on the huge West Side of Chicago between 1965 and 1975. What happened to the homicide rate on the West Side relative to areas without that kind of demographic change?
I don't specifically know the answer to those questions, but I wouldn't rush to denounce Bill Bennett unless I had checked a few examples like those first.
demographic change != an entire ethnic/racial group without kids by gov't decree
Not even by a long shot. There is no way to make reasonable causal statements about things that have no parallel in the past unless you're very sure about your underlying causal model and how it extends – which, I think everyone can agree, you shouldn't be in this case.
Steve: What Sebastian said. Demographic changes and ethnic differences are interesting and worth studying, but it's a big leap from there to making claims about the potential effects of genocidal policies. Or even to be so sure, as Levitt seemed to be, that reducing the number of people of any particular group would lower the number of crimes. I just think it's silly for intellectual authorities–a former secretary of education and a leading academic economist–to talk like this. If Bennett had talked about Compton etc., that would be fine–but that's not what he was doing. (To take the Levitt discussion to a more technical level, there's a big difference between changes at the margin (changing the number of abortions) and aggregate changes (aborting every baby), and there's a bit difference between abortions chosen by the mother and those forced upon her.) The whole thing is just a mess.
"The whole thing is just a mess."
Then why reopen the mess by denouncing Bennett in 2010 for something he said off the top of his head in 2006? Especially when your line of argument isn't terribly well thought out despite having 3.5 years to think it over?
Bennett wasn't making an empirical prediction, he was making a moral reductio ad absurdum argument using economists' ceteris paribus logic, which isn't that shabby by the standards of talk radio.
Bennett wasn't advocating genocide, he was responding off the top of his head with a reductio ad absurdum to a radio caller who agreed with him that abortion should be illegal. But Bennett didn't approve of his caller's argument for why it should be illegal: because abortion was bad for Social Security solvency in the long run. Bennett replied that he didn't like that kind of cold-blooded pragmatic argument about abortion, and offered the obviously repugnant reductio ad absurdum his caller's logic of aborting every black baby as a way to reduce crime.
The reason this became such a big brouhaha that you remember it in 2010 is because it's the kind of thing that everybody sort of figures must be true, so everybody gets upset when somebody comes out and actually says it out loud, just like so many people in 2007 got outraged at America's most prominent man of science, James D. Watson, for coming out and saying out loud what most people fear is true about race and IQ.
As he explained later, Bennett was referring to all the intellectual adulation that Steven Levitt had received since his paper with John J. Donohue on how legalizing abortion had supposedly cut the crime rate had leaked out in August 1999, especially because the black abortion and crime rates are so high.
As Levitt and Donohue argued in their 2001 paper, blacks have about a 3x higher abortion rate than whites and about a 6+x higher murder rate, so increasing abortion through legalization would, all else being equal, reduce the murder rate a generation later.
Levitt's view on abortion, crime, and race is straight-forward and is widely shared privately. I've been told many times in private that outlawing abortion would increase the black birth rate and thus the national crime rate.
Personally, I was the first to challenge Levitt empirically, in Slate in a dialogue on August 23-24, 1999:
http://www.slate.com/id/33569/entry/33571/
What Levitt predicted just hadn't happened historically. The first cohort born after legalization had the worst teen homicide rate in history, about 3x higher than the last cohort born before legalization. Among black teens, those born after legalization had about a 5x higher murder rate than the previous black cohort.
In the Levitt-Sailer empirical debate, the burden of proof has been almost universally put on me since Levitt's logic is so compelling within the framework of economics' "all else being equal" logic. That's how economists reason: ceteris paribus.
In late 2005, Foote and Goetz showed that Levitt had simply screwed up his code, but the NYT has never revealed that their star blogger messed up. If the flaws in Levitt's empirical argument had been more popular, perhaps Bennett wouldn't have used that reductio ad absurdum.
Still, I have to admit that it wasn't idiotic for so many in the intellectual establishment to give Levitt rather me the benefit of the doubt. After all, his theory makes sense in theory.
To understand why it didn't work, you'd have to understand a lot more about the underclass than most people in the intellectual establishment do.
What I think actually happened, and there is quite a bit of empirical evidence to support this, is that the legalization of abortion drove up the unwed pregnancy rate dramatically. The women who got legal abortions tended to be the more responsible middle class / working class types while the underclass types didn't get abortions and had their kids. This is unexpected to the kind of upper middle class people who read popular economics bestsellers, but what do they know about life at the bottom of society, where the future criminals come from?
Steve:
To answer quickly, my purpose was not to denounce Bennett. I just thought it was funny that Kaus was yapping about the African American vote with a guy who was notorious for musing about the potential consequences of a genocidal abortion strategy–not a strategy that Bennett endorsed, but something that he was nonetheless sure of its effects. To me it all just seems like a big joke: on one hand, I suspect Kaus's Senate run is part of a book contract and, in any case, I hope and expect it will lead Kaus away from gossip-blogging and toward a more interesting career as a policy thinker. On the other hand, I found Kaus's speculations about the black vote to be implausible, and it seemed funny to me that he'd be bringing all this up with Bennett, of all people.
P.S. I haven't had "3.5 years" to think this over! After posting on Kaus and following some links, I came across the Bennett quote, which I vaguely remembered from a few years ago. For reasons described above by myself and commenter Sebastian, I think Bennett's (and Levitt's) attitude of certainty is inappropriate, but I only bothered to comment on this because I'd already been writing about Kaus. Public figures say silly things all the time, and I wouldn't have brought up a 3-year-old quote all on its own.