Birth Control and Crime

John Donahue and Steven Levitt, in 2001, created one of the biggest academic controversies to hit economics when they provided evidence that legalized abortion during the early and mid-1970s was responsible for the decline in crime rates observed in the United States in the early 1990s. While the original findings have not held up to deeper scrutiny, and the results cannot always be replicated using alternative methods of identification, several papers have found evidence that suggests those abortions performed eliminated future criminals. A new working paper by Juan Pantano, graduate student at UCLA on the job market this year, extends the original Donahue and Levitt paper by focusing on a different set of exogenous shocks - legalized access to the birth control pill. The paper is here.

I heard Juan present this paper at the Population Association of America a year ago, and was impressed. I've not read the paper close enough to speak about its strengths, either, so I'll just explain what he does in this paper. Goldin and Katz noted in an earlier set of papers that the Vietnam War created a natural experiment relevant to contraception access, which they exploited to identify the effect of birth control access on fertility outcomes as well as female employment outcomes. Ordinarily, females were unable to get access to the Pill (legally) until they were 21. The Vietnam War put pressure on several states to lower the "age of elibility" for voting, completely unrelated to the age of eligibility for getting the Pill. The argument was that if you're old enough to fight in the war (age 18), you're old enough to vote for or against it (then 21), so several states lowered their age of eligibility from 21 to 18. These legislative changes coincidentally allowed 18-20 year old women to get access to the Pill, slightly ahead of other cohorts and other states. Goldin and Katz use this quasi-experiment to find the effect of birth control access on cohort outcomes. This kind of empirical approach is essentially an instrumental variables strategy, and during the 1990s, was quite the rave. And still is, to some degree, but not like then, where original and creative experiments like these were almost enough to get you published in the best journal.

Pantano's reasoning goes like this - if abortion lowered crime, then wouldn't any contraceptive technology do the same, namely access to the Pill? He then uses Goldin and Katz's coding of the states with early repeal of their state age of eligibility laws in a standard OLS framework, and finds indeed that access to the pill is negatively associated with crime rates later in time.

Because of the controversy of the original Donahue and Levitt (2001) study - only some of which has to do with moral protests about the normative implications of their study (which frankly are outside the scope of the study and irrelevant to their findings, which is purely scientific) - you have to take Pantano's findings with a grain of salt initially. One should read the forthcoming Quarterly Journal of Economics (2008) paper by Foote and Goetz, for instance, that documents the programming errors in Donahue and Levitt's original piece, as well as their specification of the models which may in and of itself create negative correlations where there are none (because they used numbers of effective abortions rather than rates is, I think, the main problem noted), as well as the forthcoming, Review of Economics & Statistics (2008) by Ted Joyce that replicates their study using simpler tests and finds no effect of abortions on crime, and you should read them closely and check whether Juan successfully persuades you he has dealt with those many problems. I am encouraged to see he does include state-year fixed effects, for instance, which was the source of the programming error. But then I'm actually astonished the results hold up. Someone should replicate this study themselves and figure out how robust this is. If it is robust, then I am personally closer to believing the abortion hypothesis. But having written a similar paper myself looking at the effect of abortion legalization on other outcomes correlated with crime but not crime itself, I've actually become a skeptic. At least for the things I was examining, the declines observed were coincidental to the abortion effect, and not causal, because when you compare outcomes of the emerging cohort to the older cohorts in those states who had not been in utero during Roe v. Wade or early repeal of abortion laws in the early repeal states, you find the declines are generic within the state across all age groups. Which suggests that what people may be picking up is actually an unobserved, generic state effect, and not a cohort effect - some kind of event or events in those states in those years which for reasons known or unknown is correlated with the decision of states to legalize abortion early.

But this paper does make a contribution. Even though skeptics may accuse Juan of simply being a part of this cottage industry of papers devoted to the abortion-crime hypothesis that Levitt started, I think given the controversial nature of that original study and the controversial nature of abortion in general, we need to know whether that theory holds any water. And science progresses through this kind of iteration. I see that Juan has many interviews on the job market this year at excellent schools, so it sounds like his paper is getting attention. Good for him. He's a very cordial person, and exceptional as an economist as far as I can see.

Update That is not Juan's job market paper. This is his job market paper: "On Scarlet Letters and Clean Slates".