March 2006
Abortion Legalization and Crime Rates: Is There a Relationship?
In 2001, John Donohue of Yale University and Steven Levitt of the University of Chicago published a paper entitled “The Impact of Legalized Abortion on Crime,” in which they argued that legalized abortion in the 1970s significantly contributed to decreased crime in America during the 1990s. The article sparked a fierce controversy which has yet to abate. The controversy further captured public attention when Levitt featured the argument in his bestselling book, Freakonomics. At a March 28 AEI discussion, panelists who have studied whether there is a link between abortion and crime weighed in on the available empirical evidence.
Jonathan Klick
Florida State University and AEI
Ever since they began to publicize the results of their economic inquiries into the effects of legalized abortion on crime rates, Professors Donohue and Levitt have managed to seriously irritate both ends of the political spectrum. The Right is angry because it does not like to hear any evidence about potentially positive effects of abortion, and the Left is angry because of the racial implications of the theory. For those people actually interested in the evidence, though, the debate is complicated by the many challenges to substantiating the abortion-reduces-crime hypothesis, including the eighteen- or twenty-year lag time in seeing effects; the relatively imprecise nature of much of the data; and the possibility that legalized abortion might also cause people to relax the precautionary measures they take regarding sexual behavior, therefore leading to more unwanted pregnancies. We might wonder if and why this debate matters, but the answers it has to offer have considerable implications on various policies affecting the ease of access to abortion, such as whether we should allow Medicaid to pay for abortions.
Christopher Foote
Federal Reserve of Boston
My Federal Reserve colleague and coauthor, Christopher Goetz, became interested in this debate because of the type of data being employed rather than because of the subject matter. We are interested in the effectiveness of using data that looks at specific groups of people born in specific states at specific time--for example, seventeen-year-olds in Massachusetts in 1990. It is potentially a very incisive tool for analyzing data, but we think there are some serious problems with how Professors Donohue and Levitt have put it to use because of endogeneity problems. In our paper, we find that there is a significant correlation between the absolute number of abortions and the crime rate, meaning that high abortion states tend to be high crime states. What we consider is that maybe there is simply a convergence between high and low crime states over time, and that is what is driving Donohue and Levitt’s finding that abortion is having a significant negative impact on crime. In other words, it is just high crime states that saw a decrease in crime in the 1990s, rather than high abortion states. When we test for that, the significance of abortion falls out unless one is willing to make heroic assumptions about its importance.
Ted Joyce
Baruch College, CUNY
In my paper for this conference, I attempted to replicate the Donohue and Levitt test, which regresses age-specific arrests and homicides on cohort-specific abortion rates. When I corrected those tests for serial correlation, the significant results of abortion on crime that Donohue and Levitt found evaporate completely. The original model put forth by Donohue and Levitt would predict huge changes in crime due to abortion, ones that I believe strain credulity. Their current method is more credible, but they now have a significant problem with their data, because they lack the abortion rates for cohorts before 1973. Without these data, their current results are based on an alarmingly small set of information, and even that set does not robustly support their hypothesis. I use data from states that had early legalization of abortion in 1970 as a test and find little support for their strong hypothesis.
Leo Kahane
California State University
My coauthors and I wanted to test this hypothesis by introducing an international comparison, so we turned to the United Kingdom. The data there has certain advantages and disadvantages, but it was certainly comparable enough to allow fruitful comparison. First, we used the original Donohue and Levitt method on the UK data. When we replicated their test as well as possible, we found similar results. We then wanted to introduce modifications to their test to deal with endogeneity problems. Specifically, we wanted to use ease of access to abortion rather than the simple rate as the relevant variable, since it is ease of access that will have the greatest impact on reducing unwanted births. One measure we used to do this was the number of children in local care (such as foster homes) as a measure of social deprivation. When we made all of the corrections we thought were appropriate, we found that there was no significant effect of abortion on crime rates.
Phillip Levine
Wellesley College
The question of abortion’s potential effect on crime is complicated by the fact that we cannot observe the number of unwanted births directly, just the total number of births. It is quite possible that the total number of births could remain constant even as the number of unwanted births fell. This is because of increased pregnancies resulting from the legalization of abortion, and because an unplanned pregnancy does not necessarily translate into an unwanted birth. For these reasons, my coauthors and I think that it is necessary to look at the demand for abortions. As a result, we object to some of the fixes proposed by Joyce and Kahane, which propose to correct for the endogeneity of abortion rates, a correction we feel throws out a great deal of useful information about variation. Our results (forthcoming) support the Donohue/Levitt thesis, although they are still vulnerable to various objections, such as those brought by Foote and Goetz.
Steve Sailer
The American Conservative
As a journalist, much of the economic material discussed here today is over my head, but I understand enough to be impressed with the quality of thought that economists have given to the subject and realize that journalists as a whole have done an extremely poor job covering it. After seven years, no one has met the burden of proof regarding a link between abortion and crime, but journalists often report the “discovery” of that link uncritically and have contributed to it becoming almost conventional wisdom. Journalists should be acting as a reality check on social scientists, who are often over-enthusiastic in their assertions. Specifically, they should point out that national crime rates did not begin falling in the late 1980s and early 1990s, as the Donohue/Levitt theory predicted, but rather peaked well afterward, in 1994. They should note that black homicide rates went up more than white ones in the 1980s, again contradicting the predictions of the Donohue/Levitt model. Finally, they should note that, contrary to the rhetoric used by many about abortion positively affecting “wantedness,” illegitimate birth rates in the country continued steadily upward from the 1950s to the 1990s. As anyone willing to look at the national data can see, abortion was the cause of the millions of unwanted pregnancies it was also the cure for, and so it is no surprise that no significant impact on crime has been observed.
John R. Lott Jr.
AEI
The Donohue/Levitt theory is quite plausible and even intuitively appealing. Akerloff, Yellen, and Katz have put forward an equally plausible hypothesis, however, suggesting that legalized abortion along with other legalized contraceptives changed the dynamics of the marriage market such that more first births occurred out of wedlock. To know which of these two effects predominates is purely an empirical matter, and to presumptively assume that one or the other is true is not justifiable. When we go to the data, the predictions made by the Donohue/Levitt theory do not hold up. If they are right, then the crime rate in the youngest groups should drop first, but the opposite happens when we look at national murder rates. When we look at groups born pre- or post-legalization of abortion, the theory predicts that the post-legalization groups should have lower murder rates, but the opposite happens. Another flaw in Professor Donohue’s paper is that it uses the age of the criminal when they are arrested, when what should be used is their age when they committed the crime, which is included in supplemental homicide reports. Interestingly, the focus on murder that was so apparent in the first Donohue/Levitt paper (2001) is no longer present, because their case is weakest with murder data. The bottom line is that the evidence we have suggests that the effect of abortion legalization on crime is to increase it, if anything.
John J. Donohue III
Yale Law School
The various criticisms of my work here today vary in their merit, but the discussion of them should not lead us away from the basic story about the negative relationship between abortion and crime, which is overwhelmingly compelling and substantiated by other types of studies, such as one in Prague that followed children whose mothers were denied abortions and found that they were significantly more likely than their peers in otherwise similar situations to develop mental health problems. At the macro level, there are also four prominent facts we should not overlook. First, the legalization of abortion caused a very large increase in the number of abortions that took place. Although the number was not zero before legalization, it is not the case that legalizing abortion only had a trivial impact on the frequency of the procedure. Second, there was a dramatic downturn in crime beginning in the early 1990s. Third, this downturn was greater in high abortion states than in low abortion states, but only in age groups born after legalization. Finally, there is no other significant competing explanation that could account for the massive drop in crime--clearly, something very important happened.
Many of our critics continue to produce harangues against our original 2001 paper, but we have moved on. The data we use in our more recent papers is improved in several ways, and if this debate is to move forward in any way, it is important that everyone adopt the best data. There are many challenging issues still outstanding in this debate--I will have to give more serious thought to the question raised by Foote and Goetz about a possible endogeneity problem resulting from convergence. But it is quite counterproductive to simply bring out national data and show how there is no huge drop-off in crime in the early 1990s. Those who engage in such obfuscation make precisely the same mistake as all of those people who claimed that rising rates of incarceration had no effect on crime rates simply because both were going up simultaneously. Our model predicts a modest decrease in crime each year as a result of abortion, not a precipitous drop, and so the gradual but eventually very substantial drop in crime that we have seen over time is exactly what our model predicts. Furthermore, our results at the state/age level are more robust than they were before. When critics simply point to a lack of a downturn in a graph in a particular year and claim to have debunked our work, they add nothing to the debate.
Florenz Plassman
SUNY Binghamton
There are three major types of issues being debated in the context of abortion and crime: data issues, econometric issues, and modeling issues. Regarding the data, there is not much argument. We wish it were better, but are ultimately forced to use what is there. Most of the discussion today has been about the proper use of econometric techniques to fix endogeneity problems. I would like to suggest that this debate is not a particularly useful one. Each critic applies his preferred fixes to the data and then declares that he has solved the problem and reached the truth, but no one seems to be asking whether the model is sound in the first place. For various reasons, the model that everyone is taking for granted, which tries to find a connection between the abortion rate and the crime rate many years later, is far from ideal, and in fact may not be able to tell us anything even in the best of circumstances. We should be looking for a better-specified model.
David Paton
Nottingham University Business School
Ideally, we should be going through an iterative process in which we get closer to a consensus with each iteration. As it is, we have three different models at work in Donohue and Levitt 2001, Lott and Whitley, and Donohue and Levitt 2006. There is no attempt to reconcile these models, and so we do not move forward. It is also vital that everyone stops trying to correct for age-year effects, because in fact those are the whole story. They are the most powerful tool available in analyzing this data, and to regress them out of the equation is to guarantee that you will not find anything significant. It is also important to try to get at unwanted children more directly. When we think about it, not all unwanted infants end up as unwanted children; many are adopted, for example. When you think about it in this light, you find that the main effect of legalizing abortion was to essentially wipe out the market for infant adoptions, which was crime-neutral, since people adopted as infants tend not to diverge in any significant ways from the rest of the population, and may even come from better homes on average. It is still possible that abortion will have an effect at the margins, but we should not expect too much from the data.
AEI research assistant Philip Wallach prepared this summary.
