Crime rates in the U.S. have fallen by about half since the early 1990s. A new working paper from the National Bureau of Economic Research finds that legalized abortion following the Supreme Court’s landmark Roe v. Wade decision in 1973 accounts for 45% of the decline in crime rates over the past three decades.
The paper’s authors, Stanford University economist John Donohue and University of Chicago economist Steve Levitt, take new data and run nearly the same model they used in their influential — and controversial — 2001 analysis published in the Quarterly Journal of Economics, where they first suggested an association between abortion and crime.
In the 2001 paper, they found that legalized abortion appeared to account for up to half of the drop in rates of violent crime and property crime to that point. They also predicted crime would fall an additional 20% over the next two decades. Levitt featured the research in the 2005 bestseller Freakonomics. The new paper also looks at violent crime and property crime.
“The thing that’s most interesting about the [new] paper is we simply repeated the regression process we went through 20 years ago with more data and the results got even stronger,” Donohue says, referring to the statistical method researchers often use to study relationships among variables. “That was a pretty interesting and powerful affirmation of the original hypothesis which was initially proposed.”
Several states recently passed legislation restricting abortion access, and research on the social effects of abortion since Roe might inform legal arguments as restrictive laws are appealed and litigated in the courts. The Roe decision effectively legalized abortion in the U.S., based on a 7-to-2 ruling that decided prohibiting abortion was unconstitutional because it infringed on a woman’s right to privacy.
The headline finding from the new paper works on the idea that, as the authors put it, “unwanted children are at an elevated risk for less favorable life outcomes on multiple dimensions, including criminal involvement, and the legalization of abortion appears to have dramatically reduced the number of unwanted births.”
In the new paper, they use an almost identical analysis model as they did in 2001 but add abortion and crime data covering 1997 to 2014. The original analysis covered 1985 to 1997.
“By imposing the restriction of using the same model we used for a paper published in 2001, no one could claim that we were fiddling with the model to generate a particular result,” Donohue says.
Changes in crime within states
The 45% crime-reduction-due-to-abortion figure comes from several types of analyses. The authors examine crime in states that legalized abortion before Roe; crime in states with high and low abortion rates after Roe; differences in crime patterns in states among people born before and after Roe; and differences in arrest rates within states among people born before and after Roe.
In the analysis on differences in crime patterns within states, they find a 10% to 20% reduction in crime associated with abortion, after controlling for several variables: prisoners and police per 1,000 residents, state personal income per capita, welfare assistance, unemployment rates, poverty rates and beer shipments.
“There has been this literature that talks about how abuse of alcohol is related particularly to violent crime,” Donohue says. “So [beer shipments are] a crude measure of alcohol abuse but the best we could correlate fairly well.”
As with their 2001 paper, not all economists are convinced by Donohue and Levitt’s new findings.
“Modern econometrics is focused on looking at really sharp changes in the variable driving your analysis,” says Theodore Joyce, an economics professor at Baruch College whose research on abortion has been published in several academic journals, including the New England Journal of Medicine and The Journal of Political Economy. “In other words, you look for really sharp breaks [changes] because you can isolate what changed and if anything moves when this moves. When you have smooth changes going on, smooth changes in employment, crime, demographics — abortion is part of that smooth change as you move into ‘80s, ‘90s and 2000s — you can’t identify abortion from the other things going on at that time.”
“To run these long panel regressions in which you are picking up the sweep of humanity and say you controlled for poverty — it moves very slowly and it’s mixed in with all these other trends, and I don’t think they can sort it out,” he says.
Donohue and Levitt replied to several of Joyce’s criticisms in a 2004 paper in the Journal of Human Resources.
Changes in crime across states
Another portion of the new NBER analysis compares crime patterns across states that had low and high rates of abortion post-Roe. The authors report that there is no relationship between abortion and crime before 1985, because there were fewer criminals affected by abortion being legal. In their analysis, a 20-year-old arrested for a property crime in 1985 is part of a cohort not affected by legal abortion because he or she was born in 1965 — before Roe. But a person born in 1975 would be part of a cohort that was affected.
From 1985 to 1997, the researchers show a pattern where crime fell more in states with high rates of arrestees affected by legalized abortion, compared with states with low rates. Again, other economists are not convinced that state-to-state comparisons are appropriate.
“There are a lot of reasons New York crime might decline relative to Utah crime,” says Christopher Foote, a senior economist at the Federal Reserve Bank of Boston. “You can try to control what’s going on with unemployment and beer sales but at the end of the day you’re never really sure whether a state like New York is better prepared to prosper in the new economy than perhaps Oklahoma is, so maybe there’s different types of people moving into New York rather than Oklahoma. There’s all sorts of other factors you can’t account for when you’re doing this state-to-state comparison.”
Foote and another economist, Christopher Goetz, identified several technical flaws in Donohue and Levitt’s original analysis in a 2008 comment published in the Quarterly Journal of Economics. They noted then that cross-state comparisons came with too much data noise, meaning those comparisons include variables that can’t be controlled for. More precise findings could be drawn from within-state comparisons, they argued.
“The best way to determine if abortion has a causal effect on crime is to compare two people who are in a similar environment today, but who had differing probabilities of being wanted at birth,” they wrote.
Donohue and Levitt responded to Foote in 2008 in the Quarterly Journal of Economics, acknowledging and correcting mistakes in their 2001 analysis while reiterating their finding that “abortion legalization reduces crime.” They acknowledge in the new working paper that they added more within-state analysis because of Foote’s 2008 comment.
Questions about the validity of comparing data across states are, “certainly appropriate to think about,” Donohue says. “One thing that’s so interesting is we are able to look not only across states at current levels but also within states.”
An ongoing scholarly discussion
Donohue and Levitt have been forthcoming with their data and methods over the years. Authors who have written papers critical of their past work often thank them for sharing their data or for providing comments.
One NBER paper from 2008 allows for the possibility that legal abortion had an impact on crime, but questions the magnitude that Donohue and Levitt found. A 2008 paper in Criminal Justice Policy Review cautions against using events that happen on an individual level — such as a person having never been born into a situation that might have led them to want or need to commit crime — to draw broad conclusions — such as that abortion legalization accounts for a huge drop in national crime rates.
Other research has found links between abortion and crime reduction, but for different reasons than Donohue and Levitt articulate. A 2007 paper in The B.E. Journal of Economic Analysis and Policy looks at data from Canada and suggests that lower crime rates are not because of fewer “unwanted” births in a given year, but rather because fewer women were becoming teen moms.
Likewise, a 2015 study in Crime and Delinquency finds that, “if there is a statistically significant relationship between crime and abortion, it is due to varying concentrations of teenage abortions across states, not unwanted pregnancy.”
Whether the new working paper stirs up another academic controversy may depend on where the paper is published. Donohue and Levitt finished their NBER paper less than two weeks ago and are still refining it for submission to peer-reviewed journals. Their 2001 paper made a splash in part because the Quarterly Journal of Economics is among the most influential economics journals in the world.
No matter where the new analysis lands, Donohue and Levitt remain confident: “It is rare for an economic theory to make predictions for 20 years into the future that are both bold and precise,” they conclude in their working paper.
Abortion rates are down — will crime go up?
Levitt, in a 2006 blog post, explained a subtle point that’s worth restating. A critic had brought up that abortion rates among white women were declining and asked if Levitt would expect a rise in crime among white teenagers.
(Total abortion rates have declined substantially since the early 1980s.)
As Levitt explains in his blog post, the idea is that it is not access to abortion per se that leads to a decline in crime, but rather a decline in unwanted births. Here’s how he put it:
“It appears that the 1990s were a time when factors such as AIDS were leading people to, for instance, use condoms or abstain from sex altogether. Not conceiving an unwanted baby is equally effective in reducing unwantedness as having an abortion.”