As I noted in the comments to my earlier post on the AEI event, and by request, I thought I would briefly sketch out some of the theoretical and empirical issues surrounding the Donohue & Levitt (D&L) result (Quarterly Journal of Economics, 2001) that increased abortion access resulted in a lower crime rate. Because of the emotionally charged nature of this debate, I feel it appropriate to start with a few important caveats:
1. I have no horse in this race. I have (now) read the papers. I’ve read Freakonomics. I’ve watched the videotape. I do not know, and quite frankly, have never met, any of the scholars directly involved in the debate surrounding the D&L result. I do not have a position as to what the true relationship between abortion access and crime really is. The only impression I have regarding the debate is that the participants care about getting it right. D&L have posted their data and made it available for the world to see, use, and dissect their results. They also were quick to admit to a coding error in earlier work, correct it, and present new results (which appear to be even stronger). This is largely a case of empirical research as it should be: an open debate about theory, methods, and a search for the true relationship, if any, between abortion and crime. I view the personal attacks on the researchers on both sides a very unfortunate waste of time.
2. I do not pretend to have mastery over the host of theoretical and empirical issues involved here. I merely claim to have a modest sense of what the debate is about. Part of the reason that I watched the event and read the papers is to learn more. I am not purporting to attempt an exhausive literature review, or level criticisms at these papers. I am merely attempting to highlight some of the issues and offer a few thoughts in response to some reader requests and my general inclination that folks are interested in the issue.
3. These are all initial and preliminary thoughts (I did just read the papers!). Future research, a closer read on my part, or just about anything else, might alter any views expressed below. For those who have also read the papers or have their own thoughts, I am sure our readers would appreciate them. Feel free to comment. I am interested in hearing from folks who saw the debate first hand, watched the video, are familiar with the papers, or who are interested at more of a “I’ve read Freakonomics and want to know more” type level.
So, here goes.
D&L almost have to be right about the plausibility of the theory. I don’t think there is much dispute about the following. The evidence underlying the theory is that poor black men commit a large portion of violent and property crime. Legalized abortion had a substantial effect on black birth patterns: small reduction in rates accompanied by a fairly large delay in the timing of births primary leading to black women having kids later in life).
There is also a strand of research showing that abortion access leads to changes in sexual activity. For instance, Klick & Stratmann show that the adoption of parental notification laws, which raise the cost of abortion access to teens, results in a reduction in risky sexual behavior. It is likely, however, that these sorts of effects are second-order concerns that will be swamped by the primary effect. Nonetheless, the empirical result that teens respond to incentives with respect to risky sexual behavior appears to be an important one for much of the normative policy debate surrounding abortion.
All of that said, the empirical questions involve the strength of the D&L econometric study in identifying the immediate effect. The basic intuition of the empirical strategy is that if abortion access lowers the number of children born, it must also lower the number of individuals in the age cohort that commits the most crimes. They further hypothesize that since the reduction in children will come disproportionately from women the least willing and able to care for their children, children born from these mothers are also more likely to commit crimes in their late adolescent years than the rest of the population.
The bottom line is that D&L purport to identify a 20% abortion-induced decrease in the crime rate for those cohorts whose mothers had access to legal abortions. This effect would be large enough to represent what would amount to about 50% of the crime decrease witnessed in the 1990s. It is also important to note, if only because it is so infrequently noted in popular accounts of the D&L study, that the authors explicitly state that:
“Our drawing a link between falling crime and legalized abortion should not be misinterpreted as either an endorsement of abortion or a call for intervention by the state in the fertility decisions of women.”
Many have levied criticisms of these (and later revised) estimates. An exhaustive summary of the criticisms and replies by D&L would be an article, not a blog post. In fact, Jon Klick has a nice literature review in this paper that is non-economists should have no trouble with, and is more detailed than my post allows. John Lott and John Whitley, for example, have argued that a countervailing effect would result from the legalization of abortion in the form of an increase of single family homes, pointing to data suggesting that children raised in single parent homes have systematically lower educational outcomes. This increase in less educated children will result in an increase in the crime rate that should offset the decrease predicted by the D&L model. On a more technical note, one of (there are others) Lott & Whitney’s criticisms of the D&L study is that one would expect, according to the D&L hypothesis, that the crime rate would decrease for younger individuals first, with the decrease spreading to older individuals over time. Using the FBI’s Supplemental Homicide Report to disaggregate state level homicides by age of perpetrator, they do not find the expected pattern.
Perhaps the largest debate, and perhaps the most intuitive, surrounds the possible confounding effects of the crack epidemic on the D&L analysis. Ted Joyce, who also spoke at the AEI event, and others have suggested that the presence of crack markets at different levels of development across states is a confouding variable impacting the crime rate that is very difficult to control for because it tended to be concentrated in NY and LA in the early 1980s (and these two cities are in 2 of the 5 states used by D&L as early legalizers). Joyce also offers other criticisms to which D&L reply here. See also this critique by Steve Sailer, and a debate in Slate between Levitt and Sailer which focuses on some of the issues surrounding the crack confound. Finally, a recent paper by Foote & Goetz criticizing D&L has also received a good deal of attention. F&G criticize one of four pieces of evidence offered by D&L regarding the abortion/ crime link. F&G’s criticisms involve the now infamous coding error and measurement error. The F&G paper is actually very accessible to non-technically trained readers, and I recommend it to those looking for a paper that summarizes some of these points in more detail. D&L have responded both formally and informally to the F&G criticisms. Levitt’s blog post summarizes these issues and D&L’s responses, so better to get it straight from the horse’s mouth.
So … if you have made it this far through the post, you might have noticed that I still have not answered Gordon’s question in the comments to my earlier post: “Josh, did either side convince you?” A loaded question if there ever was one. But a good one. So what is my answer?
I am fairly familiar with the techniques of exploiting variation in legal change to identify changes in behavior. As a general matter, it can be pretty hard to identify an immediate effect. D&L are purporting to identify an effect at a 20 year lag. Not only that, their results imply that this effect is substantially larger than anything else — though Levitt’s subsequent work examines other forces reducing the crime rate in addition to abortion access. Further, if the effect is real, it would show up exactly after the most important urban crime phenomena since prohibition (crack), which is a pretty sizeable confound. Add to that the fact that you’re really only identifying the effect from two shocks (CA and NY legalizing in 1970 and then Roe v. Wade), and one should at least approach the results with a healthy skepticism. I guess that is where I ultimately stand: somewhat skeptical of the result but nowhere near convinced that it is implausible as a matter of theory or empirics. The scholars involved in this debate are working to improve the precision of the estimates by dealing with these issues head on. One of the nice things about econometrics is that these debates such as these take place out in the open, assumptions and methods exposed for all criticize, with data available to all. That particular form debate is likely to move us closer to the truth and increase our knowledge over time.