Bryan Caplan  

Not Junk? A Question for Arnold

PRINT
Good Sense on Fiscal Policy... CoA Watch...

Arnold writes:

The junk-science aspect is to look at correlations and say that they prove what intuition and common sense might wish to be true. I just do not believe that it is possible to use correlation to untangle the relationships among health, socioeconomic status, and intelligence.

He adds:

But my view is that the statistical tricks are just that--tricks. That is why I don't care much for Steve Levitt's work.

So if ordinary statistical work is junk, and clever Levitt statistics is junk, what empirical work is worthwhile, if any? It's one thing to say that no statistical work gives FINAL ANSWERS. I agree with that. But it's another thing to say that statistical work is WORTHLESS, and it almost sounds like that is Arnold's claim.

Arnold, please tell us: What's not junk?


Comments and Sharing





COMMENTS (2 to date)
Eric writes:

and Arnold please explain why you think Levitt's approach is not valid.

Steve Sailer writes:

The general problem with Steven D. Levitt's work is that he publishes ambitious theories about topics that he doesn't know all that much about before he performs simple reality checks that would raise questions about his rather grandiose conclusions.

His most famous theory -- that legalizing abortion in 1970-1973 -- lowered the crime rate among those born after legalization -- was based on looking at crime rates in 1985 and 1997. Back in 1999, I pointed out to him that he had forgotten about the vast crack war that came between 1985 and 1997, and that when you look at the data in more detail by age cohorts, you find that the homicide rate nearly tripled among the first cohort of youths aged 14-17 born after legalization compared to the last cohort born before legalization.

http://www.slate.com/id/33569/entry/33571/

Ever since, Levitt has been scrambling to patch up his theory against ever growing attacks from more careful social scientists. Last fall, Foote and Goetz pointed out that Levitt's 2001 paper was based on two technical errors he had made. Now, here is the latest abstract:

Further Tests of Abortion and Crime

Ted Joyce
Professor of Economics
Baruch College, City University of NY
& National Bureau of Economic Research
May 2006

The association between legalized abortion and crime remains a contentious finding with major implications for social policy. In this paper, I replicate analyses of Donohue and Levitt (2001, 2004, 2006) in which they regress age-specific arrests and homicides on cohort-specific abortion rates. I find that the coefficient on the abortion rate in a regression of age-specific homicide or arrest rates has either the wrong sign or is small in magnitude and statistically insignificant when adjusted for serial correlation. Efforts to instrument for measurement error are flawed and attempts to identify cohort from selection effects are mis-specified. Nor are their findings robust to alternative identification strategies. A convincing test of abortion and crime should be based on an exogenous change in abortion that had a demonstrable effect on fertility. Thus, I analyze changes in abortion rates before and after Roe to identify changes in unwanted fertility. I use within-state comparison groups to net out hard to measure period effects. I also follow Donohue and Levitt (2004) and average the effects of abortion on crime over 15 to 20 years of the life of a cohort to lessen the impact of the crack epidemic. I find little support for a credible association between legalized abortion and crime.

Comments for this entry have been closed
Return to top