Steven Levitt and John Donohue write:
More than two decades have passed since we published an academic paper linking the legalisation of abortion to the enormous decline in American crime since the 1990s. The underlying theory is straightforward. Children who are unwanted at birth are at risk of a range of adverse life outcomes and commit much more crime later in life. Legalised abortion greatly reduced the number of unwanted births. Consequently, legalised abortion will reduce crime, albeit with substantial lags. . . .
The data available at the time strongly supported our hypothesis. We showed, for instance, that crime began falling sooner in the five states that legalised abortion in advance of Roe v Wade, . . . crime in states with high and low abortion rates followed nearly identical trends for many years, then suddenly and persistently diverged only after the birth cohorts exposed to legal abortion reached the age at which they would commit crime. . . . looking at arrest data, which reveal the age of the offender, the declines in crime were concentrated among those born after abortion became legal.
These data patterns do not, of course, make an open-and-shut case. No randomised experiment has been conducted on this topic. We also didn’t help our own case by mislabelling the set of control variables included in one of the specifications in one of our tables—a regrettable mistake (later corrected) which led some people to dismiss all of the paper’s findings. . . . We concede that reasonable people could disagree about how convincing the findings were in our initial paper.
They now come in with new data:
We could already at the time of our first academic paper in 2001 make strong predictions about what our theory would predict should happen to future crime. Indeed, at the end of that paper, we made the following prediction: “When a steady state is reached roughly 20 years from now, the impact of abortion will be roughly twice as great as the impact felt so far. Our results suggest that all else equal, legalised abortion will account for persistent declines of 1% a year in crime over the next two decades.” . . .
That’s what we did in a paper published in 2020. Our methodological approach was straightforward: mimic the specifications reported in our original paper, but limit the time period to the years that were out of sample, ie, those after our original data ended.
The results provided stunning corroboration of our predictions. For each of the seven different analyses we had presented in the initial 2001 paper, the results for next two decades of data were at least as strong as the results in our initial dataset, and in most specifications even stronger. This included what the main critics of our 2001 paper called the “crucial” test, showing that the abortion rate at the time of any birth cohort negatively correlated with the age-specific arrest rate for that cohort years later as it moved through ages 15–24, while perfectly controlling for whatever other factors were influencing crime in a given state and a given year.
The magnitude of the implied impacts we are talking about is huge. If you look over the entire sample, violent-crime rates fell by 62.2 percentage points in high-abortion states whereas they rose by 3.1% in low-abortion states.
I think they meant to say 62% and 3% here, but I get the point.
Levitt and Donohue continue:
Though there is not complete acceptance of our hypothesis among academics, all agree that if our paper is not correct, then there is no viable explanation for the enormous drop in crime in America that started in the early 1990s. Indeed, there is not even an arguable theory to supplant the abortion-crime link. . . .
I haven’t followed the details on this one, but I was struck by that last quote, so I sent a message to Donohue asking about it. It’s not clear why we should expect a single explanation for the drop in crime. There are lots of social trends that occur without a single clear explanation, right?
Donohue replied:
This may be a semantic disagreement, but what I meant to say was that “no existing explanation for the large, otherwise-unexplained portion of the crime drop has any empirical support at anything close to the abortion/crime hypothesis.” If someone said “I think social trends that influenced the high-abortion states more powerfully is what explained the crime drop,” I would say that could certainly be true, but that is not really a theory but just a conjecture that then would need explaining. But I could imagine that high abortion states were more progressive, and started doing things like pre-school enrichment or better social programs or better policing that reduced crime—but our data shows that whatever happened in these states did emerge in a particular pattern that isn’t simply “high abortion states—perhaps being more progressive—started reducing crime more effectively.”
That’s because our data allowed us to link actual abortion rates in a particular state in say 1980 with crime by 15 year olds in 1995 and compare what was happening to slightly older cohorts in the same state who had been born with lower abortion rates. I think our Table 7 column 2 (showing impact of abortion on violent crime per capita) is perhaps the most unassailable (it is what our critics in 2000 called the “crucial test” of the hypothesis). It essentially looks at crime by 15-24 year olds by single year of age to see their violent crime rates—while controlling for whatever else influences crime in that state in each year of our data. We show that if 15 year olds were born in a cohort that experienced a higher abortion rate than 16 year olds, then the 15 year olds in say Illinois in 1995 would have a bigger drop in crime than the 16 year olds in Illinois (holding constant the fixed effect of crime in Illinois in 1995). The same would then be true the next year (1996) for these same two cohorts who would then be 16 and 17 year olds. Since the 16 year olds had been born in a cohort experiencing higher abortion rates they would still show less violent crime then 17 year olds—and this controls for whatever other factors were influencing crime in the particular state in the particular year.
So yes, other factors could reproduce this effect, but it is not a simple story of some states started doing better at fighting crime and that correlates with higher levels of abortion, since that would not explain the differential effect by year of age (linked back to the abortion rate at the time of the birth cohort).
It still seems to me that there are enough differences between high-abortion and low-abortion states that lots of things could be going on here. In their paper, Donohue and Levitt look at incarceration rates, policing rates, and lead exposure and in his note to me, Donohue mentioned pre-school enrichment or social programs, but I’m not thinking so much about policy as about, yes, what Donohue called “social trends.” We recently discussed similar issues regarding the less politically-contentious topic of the decades-long decline in cigarette smoking. Policy changes can make a difference, but policy changes are also occurring in the context of changes in public opinion and changes in social conditions.
I guess what I’m saying is, yeah, the abortion-crime correlation is worth pointing out and discussing, and I understand Donohue and Levitt’s frustration with critics who don’t seem willing to take yes for an answer; still, what we have here are observational comparisons of trends in different sorts of states.
Dale Lehman send me the following note by criminologists Graham Farrell and Nick Tilley disagreeing with Donohue and Levitt. Farrell and Tilley write:
We support the rights of women to control their fertility. However, we were surprised to see Steven Levitt and John Donohue reiterating their claim that abortion caused a drop in crime. Roe v Wade, the case in 1973 that legalised abortion in America, related only to that country. But the drop in crime was international, occurring in all high-income countries. Abortion was not a factor causing the steep falls in crime observed in Canada, Britain and many other places. Declines in big-volume crimes in America, including burglary and larceny theft, began in 1980 not 1990. This was just seven years after Roe v Wade, so too soon for abortion to have had an impact on adolescent crime.
We have been researching the role of security for many years. Evidence showing the importance of security in generating the reduction in crime has grown.
The Levitt and Donohue analysis relied on five states that introduced abortion earlier than 1973. Any effect is probably spurious and related to income. These states had median incomes far above the national average. Higher-income residents replace their cars and renovate or buy new homes more quickly and frequently, buying more and newer security measures. There is scientific consensus on the effectiveness of electronic vehicle-immobilisers in reducing car theft across multiple jurisdictions, and of door deadlocks in reducing break-ins.
America, followed by other rich countries, experienced what has been termed an avalanche of security in all walks of life. With easy property crimes no longer available, fewer adolescents became involved in crime. This led to reductions in violent crime, which are far fewer in number than property crimes.
There does seem to be a lot more security than there used to be, so I guess this addresses Donohue and Levitt’s claim that “there is no viable explanation for the enormous drop in crime.”
I have two linguistic comments to make:
1. The choice of the adjective, “viable,” which appears twice in this posting, whether as a catchphrase or out of laziness, is unfortunate.
2. Even worse is the 30 plus times the word “abortion” appears when not immediately followed by the word “rights.” It is the linguistic triumph of the anti-abortion rights movement to have framed the issue this way. To put it bluntly, no one is “pro abortion”; the majority of the country is for abortion rights.
Choice of words often do make a difference. For example, Fisher’s famous concept, “statistical significance” has been much discussed and misunderstood particularly because it promises so much more merely because of the choice of the adjective.
Good points!
Do you mean because something could be statistically significant but and argument would still have to be made that the results are literally significant, as in greatly important or definitely correlated
> Choice of words often do make a difference.
What word should they have chosen then?
Sure, people are pro abortion. Have you listened to any of the DNC speeches this week? They even have a bus outside giving free abortions.
If abortions are really reducing the crime rate, then that is another reason to be pro abortion.
Do you also insist that people say “pro gun rights” instead of “pro gun”, “pro union rights” instead of “pro union”, and “pro war declarations” instead of “pro war”? No, the simplified terms are preferable.
If you want clarity, you should say “gun rights.” I’ve owned one or more guns for over 75 years, and lost track of how many deer I’ve shot, so I’m not anti-gun. But, I think the Second Amendment was about militias, so I’m sure as hell anti-gun rights if you are talking about the right to own an assault rife.
The whole point of the second amendment is that people should have the right to own the means to protect themselves from the absolute power of the government. So what would be the point of the second amendment of “gun rights” were restricted to hunting rifles and flintlocks when the government has far more sophisticated and deadly weapons?
I’m personally *not* a gun owner and never have been and don’t want to be. But the second amendment has a purpose and it’s not so people can freely shoot chickadees and deer.
Roger –
So to clarify. Would your argument be that if I want a woman to have the right to decide whether to get an abortion, I have to want her to get that abortion?
If not, then how would that not reflect your view?
I am not trying to impose my views on anyone. The activists on one side says “pro life”, while the other side says “pro choice”. Being pro or anti abortion seems like the more neutral terminology today.
Democrat politicians used to say they wanted abortion to be safe, legal, and rare. Now they seem to have shifted towards saying that abortion is a good thing, for women who want one.
Most people favor abortion with restriction, and so do not fall neatly into pro-abortion or anti-abortion. I understand that. I just don’t think that adding the word “rights” helps much.
Roger –
I just don’t think that adding the word “rights” helps much
I’m trying to interrogate your logic. I didn’t think you were trying to impose your views.
When I first read you comment I thought maybe you have a point. I tried to think of other parallel contexts where we distinguish “rights” from being “pro” something and couldn’t come up with any.
But then I thought about it more and I think there’s a logic problem. Because I think women should have the “right” to decide to get an abortion doesn’t mean I’m “pro” them getting an abortion. That’s just logically flawed, IMO. I’m in favor of them having the right, but I’m not “pro” abortion.
So that’s why I asked you that question. I think adding the word “right” is more accurately descriptive of my viewpoint. I would still like you to answer my question.
Roger,
Here in California, my wife had a right to get an abortion; instead, as I wanted, she had my son. Does that clarify things for you?
John, California is a very pro-abortion state, and a wife does not need her husband’s permission to get an abortion. She does not even need to tell him that his future child is being terminated. Not sure what that clarifies.
Also, you seem to favor pistols and hunting rifles, but not the AR-15 that is the most popular gun in the USA. So not really pro-gun or anti-gun.
Joshua, if you think women should have an unrestricted right to an abortion, anytime during pregnancy, then I would say you are pro-abortion. Just like I would say you are pro-gambling if you favor gambling casinos everywhere. It is understood that you do not necessarily think women should have to get abortions, or gamble at a casino.
Now maybe you think that abortions are bad, and should only be used as a last resort. Okay, that’s fine, but it is not what the typical pro-abortion politician says today. They say that abortion frees and empowers women. And D-L say that it lowers the crime rate.
> Do you also insist that people say “pro gun rights” instead of “pro gun”, “pro union rights” instead of “pro union”, and “pro war declarations” instead of “pro war”? No, the simplified terms are preferable.
Note that it’s not that he objects to saying “pro abortion” instead of “pro abortion rights”.
Apparently his issue is with the use of the word “abortion” not immediately followed by the word “rights.”
Maybe a more appropriate analogy would be complaining about the use of the word “gun” to refer to a device designed to propel a projectile using pressure or explosive force outside of the syntagm “gun rights” – because that would indicate the linguistic triumph of the anti-gun rights movement in framing the issue in some way or something.
The topic is not abortion rights, it is actual abortions. A right that isn’t exercised would be irrelevant to Levitt’s hypothesis, whereas illegal abortions would count just as much.
I was reading along waiting for the cross country comparison. But is there not another explanation in the ageing populations of these countries? Doesn’t the propensity to do crime peak at a relatively young age (early 20’s)? So just from the ageing populations you would expect a decline in crime per capita. This is all just from vague memories about crime stats.
Interesting:
…but that is not really a theory but just a conjecture that then would need explaining.
What is the level of control for confounding variables, or mediators or moderators, or interactions, that is required to cross the line from “conjecture” to “theory?”
I dunno, I think that “social trends” is a huge fly in the theory ointment. If they want to argue for abortion as a mediator or moderator or interaction variable, then sure, the distinction he’s drawing between their “theory” and others’ “conjecture” would make sense to me. If the argument is that more abortions “cause” a reduction in crime, “conjecture” fits, imo.
Andrew wrote:
“It’s not clear why we should expect a single explanation for the drop in crime. There are lots of social trends that occur without a single clear explanation, right?”
If there is a significant change from a reasonably stable previous global crime rate that begins at a certain time, we should indeed expect a single root cause.
Crimes such as larceny predate civilization. The battle between criminals and law enforcement has been waged since the dawn of civilization. Excluding instances when the civil order broke down, a phenomenon subject to constraints – a bump in crime leads to a bump in enforcement – will tend to become relatively stable, with minor fluctuations around a mean. A significant and unprecedented excursion from that stable mean would then require a rare occurrence. Hypothesizing simultaneous rare occurrences to explain a single excursion from stable behavior is unlikely to be right.
Has anyone done a similar study on the introduction of security cameras and facial recognition software?
“We concede that reasonable people could disagree about how convincing the findings were in our initial paper.”
No, according to Wikipedia, their initial paper had a bad computer error, and no one should accept it.
https://en.wikipedia.org/wiki/Legalized_abortion_and_crime_effect
“all agree that if our paper is not correct, then there is no viable explanation for the enormous drop in crime”
I very much doubt this. As noted above, there are many plausible factors, such as the rise of cell phones, surveillance cameras, and 911 services. Glancing briefly at the D-L paper, it does not appear that they consider such factors.
I thought lead was a widely accepted major contributor to the crime rise & drop worldwide.
It’s my understanding that there’s solid evidence of income inequality being an associated factor as well?
How would that explain crime increasing during the Great Society and declining after welfare reform?
Inequality is strongly associated across time and space with crime. REALLY strongly. It’s more or less crime ~ exp(k * gini)
It is however, to be expected that the experience of inequality is not an instantaneous thing. If in one day a guy down the street earns $1000 a day and you earn $100 a day this is unlikely to cause crime. If for a decade the guy down the street earns $1000/day and accumulates it and flaunts it and hires many servants and tears down buildings and builds big mansions and buys off politicians and gets laws made to make his income increase and puts pressure on the govt to clear out the old neighbors… then you expect the surrounding neighborhood to be upset and be willing to do things they otherwise wouldn’t.
So, you might best describe crime now as related to some recent average over time of the experience of inequality. An exponential weighted average with say 1/10 of a lifetime as the exponential decay time might be reasonable, which means let’s say 20 years of history would have some nontrivial contribution.
Also, it certainly isn’t the only thing driving crime. During that period the Vietnam war was driving a lot of animosity between generations and soforth.
As already mentioned, the Lead / Crime hypothesis has a well understood mechanism and is a reasonable theory with some empirical support https://www.sciencedirect.com/science/article/pii/S0166046222000667
Lead pollution increased throughout the Great Society period. Lead was banned in catalytic converter containing cars in 1975, and decreased significantly between 1975 and 1985 and banned completely for passenger cars in 1996.
Lead, Income inequality, and drug policy, all operating together potentially explain a LOT of the variation in violence since say 1946.
“Inequality is strongly associated across time and space with crime. REALLY strongly. It’s more or less crime ~ exp(k * gini)”
Income equality dropped dramatically in the US from 1990 to 2010 (see graph in post)?
Yes and from 1975-1995 lead exposure dropped dramatically. The lead hypothesis is that lead exposure from say 0-15 years of age affects crime committed by 15-25 year olds. So we expect that effect to be seen from 1990-2010 ish and we do, and we see it in other parts of the world all timed to coincide with the local lead exposure timing.
The income inequality argument is that the experience of inequality leads to crime, but also you’d expect the effect of changes to cause delays because in an “step response” of a sudden change in inequality it’s not really visible and notable until the effects accumulate through time… So as inequality rose from the 90s through 2010 ish, so eventually the lead decrease aged out and crime increased from accumulated inequality experience, and that’s consistent with the large increases in crime from say 2014-2020… So the hypothesis that lead, inequality, and drug policy all are important for crime makes sense as long as you understand these things are dynamic processes that play out through time in ways that have timescales about on the scale of 15-20 years (because for example most crime is committed by 15-25 year old males and their penchant for crime is related to the experiences they have between birth and age 15-20
I just wanted to clarify that for you the words:
“Inequality is strongly associated across time and space with crime. REALLY strongly”
are consistent with:
“crime and inequality can move decisively in opposite directions for 2 or 3 decades”
They can yes, because as Gelman said ” There are lots of social trends that occur without a single clear explanation, right?” whose corollary is there are lots of social trends with potentially multiple simultaneous explanations, and also in general dynamical systems can move in opposite directions to the causal forces. The sun attracts comets, and yet it is possible for comets to move away from the sun for decades before they finally turn around and come back towards the sun.
In general dynamics are ignored in social sciences, and I think this is perilous. Economists imagine equilibrium occurs at the timescale of anywhere from milliseconds (in the stock market) to a couple hours or days maybe, but in fact it’s pretty clear there are lots of Economic processes that take timescales of 20 to 100 years to occur. The creation of the Fed very clearly has created continuous dynamics since 100 years ago. Similarly criminology people and sociologists and such tend to regress outcomes in one year against predictors in that year, or perhaps in the previous year. But rarely if ever fit models involving real feedback dynamics.
So, yes my own personal feeling is that it’s likely that the lead hypothesis holds some water, there are known mechanisms for its action on the brain, and there’s consistent evidence across multiple regions of space. In general i’m not a huge fan of mother jones but they occasionally do useful stuff, and this seems like a reasonable roundup of some of the evidence, and includes multiple graphs from around the world https://www.motherjones.com/kevin-drum/2018/02/an-updated-lead-crime-roundup-for-2018/ If the hypothesis holds water, then it predicts a decline in crime from 1995 to 2010 caused by a decline in exposure to lead in the ~ 1975-1995 timeframe, since the hypothesis predicts crime 20 ish years in the future from exposure.
If we believe the lead hypothesis then since lead has remained out of gasoline, its effect from say 2010 to now should be relatively minimal in the US, which means that other less dominant effects can be involved in this region of time, including inequality.
At the same time, inequality as measured by the income Gini in the US hasn’t changed that strongly. It changed more strongly between 1980 and 1995 than it did between 1995 and 2024.
https://ourworldindata.org/grapher/economic-inequality-gini-index?tab=chart&country=~USA
So, in fact, crime decline from 1995-2010 doesn’t move “opposite” to increasing inequality, since in this timeframe income inequality stayed relatively constant having already increased 1980-1995.
In any case, dynamic models (where for example rates of change of crime now are a function of say weighted averages of this history of lead exposure and income inequality in the previous say 10 to 20 years) would be a smart lens to look at crime through. Perhaps when I’m done with a current project on migration I’ll fit such a model.
Kevin Drum backs the lead theory and has reported on it extensively. Very convincing, unlike abortion or policies.
https://jabberwocking.com/legalized-abortion-has-nothing-to-do-crime-rates/#comments
Reminds me of this:
https://journals.sagepub.com/doi/10.1179/002436309803889043
A slightly different take is improved health is not required to “make life expectancy numbers go up”. Encouraging the abortion of unwanted pregnancies can have a huge effect on its own.
Three things from someone who hasn’t followed this debate at all:
1. Surely the relevant factor for L&D shouldn’t be abortion rights but abortions themselves. If you measure the second by changes in the first, you could have a lot of measurement error, yes? Do they make an attempt to use estimates of abortions per, I don’t know, pregnancies/women of fertility age/something else as a variable?
2. If the mechanism is about avoiding unwanted births, then surely the availability and use of birth control should be part of the story. As I understand it, rates of unwanted and teen pregnancies, after rising many decades ago, have been on a secular decline. Is this confounding?
3. If I understand correctly, the evidence for the L&D hypothesis is all ecological. Do they examine disaggregations (crimes, abortions) to bring the ecological levels down and test the likelihood that their mechanism could work at the individual or prospective individual level?
All good points. Yet what I find most striking in the L&D statement is its tone – not just doubling down, but feeling vindicated because reality has matched what they would have predicted (just imagine investment advisors using that logic) and virtually dismissing all other possible explanations. I’m not saying L&D are wrong, or that the F&T counterarguments are correct, but they should at least address the international context where their explanation doesn’t seem to fit as well. If good science is achieved through replication and a converging of understanding, it seems like the abortion-crime link is a long way from looking like that.
Franklin Zimring discusses this in his book *The Great American Crime Decline*:
“To the extent “cohort effects” depend on a reduction of the size of the population born after 1973, any cohort dividend from Roe v. Wade would be minimal. Using 1972 as a baseline, the average birth volume in the four years after that was about 3% lower, but then the volume turned up, so that the average number of new births was 6% higher in 1979 than in 1972.”
In addition, the percentage of greater at-risk births — single mother and single mother ages 15-19, all are steady or increasing.
The abortion case fits with the drop from the peak in the early 90’s but doesn’t fit with the rise from the mid-70’s. If unwanted pregnancies were a driver I would expect a plateau or random variation prior to the RoevWade decision, no? Should have been lots of disgruntled youth built up in the years prior. So it seems like there are other factors maybe more important at play here.
Some claim the rise of video games explains the decline in crime. One narrative appears here:
https://www.psychologytoday.com/ca/blog/the-main-ingredient/201908/video-games-cause-crime-or-prevention
The article is lightweight, and I don’t endorse it in whole or in part, but it does name possible explanations for the fall in crime rates other than abortions. The web is munificently offers discussions about this and other matters.
Karl Smith wrote on explaining the rise & decline of crime:
However, he was talking about lead as the explanation.
Oh Yeesh. Leavitt makes Nate Silver look less of a scam artist by comparison.
How the theory accounts for the crime drop being an international phenomenon?
As Donohue writes in his reply to Andrew, a key point is that cohorts born in a high abortion year have lower crime rates as teens and the pattern persists as the cohort ages. I don’t see how improved security or social programs explain that finding. I guess there could be other factors that vary annually that affect both abortion rates and future crime rates, but none come immediately to mind.
I’m going to guess that there isn’t any place on the planet where abortion policies aren’t associated with attitudes towards the rights of women.
Say attitudes towards the rights of women is what drives policies about abortion. If so, and crime varies in a association with abortion rates, and you’re absolutely certain that nothing else explains variation in crime rates, how would you determine if it’s the attitudes towards the rights of women or abortion policies that is “causal?”
I agree, but what we´re talking about is interannual variation (unless I’m misunderstanding Donohue’s message). Attitudes about women´s rights would be associated with some average abortion rate, but it wouldn’t fluctuate year-to-year at the same rate as abortion rates. The argument being made is that an increase in abortions from t to t+1 is associated with a decrease in crime from t+15 to t+16. But I don’t believe there would be coincident shift in policies or attitudes or security or lead or whatever between t and t+1.
turn –
Yes, that is the argument presented here:
That’s because our data allowed us to link actual abortion rates in a particular state in say 1980 with crime by 15 year olds in 1995 and compare what was happening to slightly older cohorts in the same state who had been born with lower abortion rates. I think our Table 7 column 2 (showing impact of abortion on violent crime per capita) is perhaps the most unassailable (it is what our critics in 2000 called the “crucial test” of the hypothesis). It essentially looks at crime by 15-24 year olds by single year of age to see their violent crime rates—while controlling for whatever else influences crime in that state in each year of our data. We show that if 15 year olds were born in a cohort that experienced a higher abortion rate than 16 year olds, then the 15 year olds in say Illinois in 1995 would have a bigger drop in crime than the 16 year olds in Illinois (holding constant the fixed effect of crime in Illinois in 1995). The same would then be true the next year (1996) for these same two cohorts who would then be 16 and 17 year olds. Since the 16 year olds had been born in a cohort experiencing higher abortion rates they would still show less violent crime then 17 year olds—and this controls for whatever other factors were influencing crime in the particular state in the particular year.
So yes, other factors could reproduce this effect, but it is not a simple story of some states started doing better at fighting crime and that correlates with higher levels of abortion, since that would not explain the differential effect by year of age (linked back to the abortion rate at the time of the birth cohort).
I guess in my zeal to question their conclusions I didn’t allow that to register enough. Still, I think that it could be important to look under the hood a bit more re changes in abortion rates. For example, first hit on a Google.
Jones and Jerman (p. 1904) present evidence of a dramatic decrease in the US abortion rate between 2008 and 2014 on the basis of reports from abortion clinics. In just six years, there was an unprecedented decrease of 25%, concentrated among young women, women with higher household incomes, and women of color. Among women aged 15 to 19 years, abortion rates dropped by almost half.
So not to dismiss your point, I still wonder if there’s something missed by just looking at abortion rates as an influence that’s independent of other potentially interacting variables.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5678419/
“Abortion was not a factor causing the steep falls in crime observed in Canada, Britain and many other places.”
I’m not entirely sure this is such a damning observation that these critics think it is. Roe was 1973.
The UK legalized abortion in 1967. France did it in 1975. Canada had two stages, and while abortion was effectively only decriminalized in 1969, the full legalization in 1988 didn’t really affect abortion rates (which were effectively identical in 1972 and 2020). Germany did it in 1974 (struck down by the court, then decriminalized in 1976).
In other words, the temporal era of when these countries did it is effectively indistinguishable from the US. Unless the critics are going to do the data analyses for these countries, just saying “other countries had similar drops in crime!” is not persuasive (imo).
“Unless the critics are going to do the data analyses for these countries, just saying “other countries had similar drops in crime!” is not persuasive (imo).”
Excellent point. Great info thanks
Along with that, when the critics say crime declined “world wide” coincident with the decline of crime in the US, what does that mean? Western countries = “world wide”? Or did countries outside the western mileu also experience a decline in crime?
There are other factors. As David Hackett Fischer contends in his book, The Great Wave: Price Revolutions and the Rhythm of History, inflation cycles played a role.
https://www.google.co.th/books/edition/The_Great_Wave/-efB6GTgNdAC?hl=en&gbpv=0
From 1990 to roughly 2005 the prison population went from .6M to 1.6M. Given that crime is highly concentrated in a small number of hard core criminals, maybe removing 1 million of them from the populations they’re victimizing had a causal effect. You don’t always need a 5-D chess explanation.
I agree with Brad that when those other countries legalized abortion is important if we’re going to use them to dismiss Levitt and John Donohue’s hypothesis. I also agree that security likely has a non-trivial impact on crime over time, but I don’t think that would convincingly explain the cohort effect in the data.
Cunningham showed a similar cohort pattern with the incidence of gonorrhea in black girls 15-19 in the wake of abortion legalization.
https://mixtape.scunning.com/09-difference_in_differences#abortion-legalization-and-long-term-gonorrhea-incidence
He then brings his own conclusion into question when he looks at 25-29 year olds as an untreated cohort as a within-state control, even though the shape of the pattern is about right, and the error bars don’t seem to rule out the pattern he’s looking for. He worries over a couple of small magnitude wrong signs on point estimates and not having statistical significance, but that model just seems noisy to me.
The point about the cohort effect is very important. We need a theory that can explain both the state-wide drop in crime and the cohort effect. L-D advocate that abortion is the only theory that currently can explain both phenomenon. Of course, we are far from being able to accept the theory because of the multitude of alternative hypotheses that have yet to be proposed, let alone tested. However, some of the alternative hypotheses mentioned here do not, on the surface, seem to be able to explain both phenomenon. Everyone is focusing on the drop in crime and ignoring the cohort effect.
Personally, I think that the numbers generated by summing up official crime statistics are basically meaningless, especially in a time series context. (1) Crimes that are in no way related are lumped together (e.g. bank robbery and wire fraud). (2) Definitions of crime change over time, leading to discontinuities in the data that have nothing to do with what is happening in real life. (3) Crucially, crime reporting behaviour has changed over time. This is linked to a reduction in the social stigma attached to crime victims. (4) Official statistics are collected by the police, who have their own agenda. I am not saying it is a bad agenda per se, but there is a non-trivial risk that the data are prepared in a way that suits the interests of the police.
One might argue ‘Should this not lead to attrition of our parameter estimates? What does not kill my statistical significance makes it stronger!’ Because the measurement error is systematic, not i.i.d., it would bias the results rather than lead to attrition. I am not saying that Levitt and Donohue’s research project is dead from the start, but they have some tough questions to answer about their left-hand side variable.
Homicide data tends to be reliable. Hard to explain away a body. Car theft next most, since most people report when their cars are stolen.
@Wonks Anonymous:
‘Hard to explain away a body.’ You would think so! The problem is not so much what happened in the case of homicide, but how it is counted and by whom, and whether these counts find their way into the national crime database. I have no objection to your point about car theft, although in this case I would rather try to get the data from the car insurance companies. Unlike law enforcement, they have a financial incentive to get their data right.
This chapter from the Handbook on the Economics of Crime gives an overview over some of the problems of UCR data: https://mason.gmu.edu/~atabarro/Measure.pdf (page 59/ 7 in the document)
There are many independent ways to cross reference homicide data. You can examine:
1. Homicide data directly from municipal police departments
2. Missing persons reports. Obviously, many of these turn out to not be homicides
3. Homicide data from medical examiners. The problem is medically, “homicide” is just related to cause of death, and doesn’t match up with the legal sense of the word
4. Scrape news websites. Not everything gets reported though
When you do this, you end up with numbers that match up not exactly, but do move up and down together in the way that you’d expect. Like all data, it’d be wrong to say it exactly represents what it says on the tin, in this case the number of people who were murdered by another person, but it’s way better than meaningless and the time trends are definitely not meaningless.
@somebody:
I agree (although I doubt many researchers will contact the literally tens of thousands of law enforcement agencies in the US for local homicide rates; see the link above). The point about meaninglessness was about adding up crimes that are unrelated, e.g. the sum of all ‘violent crimes’. For example, it makes a big difference if A starts pushing B after B has insulted A, or if B is stabbed with a pair of scissors by A while laughing manically. Technically, both are violent crimes. I am not sure about the exact definition of violent crime in the UCR, but I do know that threatening someone with a weapon, even if it is just a stick or a Swiss Army knife, would be counted as a violent crime in the German Police Crime Statistics (PKS). Threats of violence should not normally be added to actual acts of violence. Again, my point is not to badmouth the UCR or any other crime statistic, but to point out that on should not accept the numbers as meaningful and correct without asking questions.
A few quick points come to my mind from a quick scan of some of the comments. We did address in our paper the similar impact of abortion legalization across Europe. Here is the quote: “The strong evidence of the impact of legalized abortion on crime in the United States would of course be strengthened by similar evidence from a different continent where the timing of abortion legalization and frequency of abortions varies greatly from ours. In fact, François et al. (2014) provide such evidence with a panel data analysis with country and year fixed effects from 1990–2007 for 16 Countries in Western Europe. The paper “confirm[s] the negative impact of abortion on crime for both homicides and thefts….” [Their model} implies that over the ensuing decade, abortion legalization reduced homicides by 12– 40% and reduced theft by 23–43%. These estimates are roughly comparable to and therefore provide significant support for our own estimates on data from the United States.”
On the issue of whether police data is problematic (which it certainly is), we show the results for murder are even stronger when one uses CDC murder rates (instead of FBI murder rates), and the CDC gets this data not from voluntary reports from police but from mandatory reporting from death certificates. Better data seems to strengthen our results (on this and other matters), which is typically a good indication for one’s theory.
Finally, increased incarceration certainly played a role in reducing crime but the big crime decreases in the 1990s and later came in states with high abortion rates and relatively smaller increases in incarceration
Thank you for engaging with us here online and for addressing the problems with crime data in your comment. Vital statistics have their own pitfalls, but I guess one can only work with the data that are available.
I have taken a look at your NBER working paper (the link above does not work for me) and your publication in the American Law and Economics Review. I wondered two things. First, is the data you used for this analysis available for inspection, especially the data from publicly available sources? Second, you exclude 25 states from your analysis: AL, AR, CO, DC, DE, FL, HI, IA, IL, KS, KY, ME, MI, MN, MT, NH, NV, NY, OK, PA, RI, SC, SD, VT, and WI. That leaves you with only half of the US states. This seriously undermines your argument that your findings apply to the whole of the United States, would you not agree?
Sorry for this drive-by comment, but I’d emphasize that NY + FL + PA and a few of the other Midwestern states would exclude about 1/3 of the population and several of the 10 largest cities.
It’s great to see authors of an important hypothesis putting their theory out into a more public forum and addressing a few questions. I’ll take another read of their paper in due course.
I was surprised by the idea expressed by the authors that there are no other theories that really explain the drop in crime. Some commentators have raised some of the other theories.
About 10 years ago I read a whole book by Gary LaFree (spelling of his name is from memory) on American Homicide. I think it was printed around 2000. He put forward one idea that you can explain the changes in homicide by trust in government. Don’t think that will hold up now. His graphs from 1930 looked pretty convincing at the time.
Anyway, at the time I went looking for papers by academics and there were lots of theories for why crime went down, but none of them (except possibly the lead hypothesis) explained why crime had gone up. After all, the homicide rate (going from memory again) was about 4/100,000 in 1960, rose to 11 (?) in the early 1990s and by 2000 had dropped to maybe 6 or 7 and by 2015 was 4/100,000.
Each paper with one theory pointed out the flaws in the other hypotheses, so there didn’t seem to be any kind of consensus.
Of course, most of the general public, and academics who don’t have to write papers on changes in the violent crime rate over time, know the answer. It’s poverty that causes crime. Why that is believed seems worthy of lots of study by psychologists. So much to understand about the human condition, and about our beliefs about the human condition.
I’m sure it’s complicated. I’m sure there are multiple factors. My strongest takeaway from doing my own research via peer-reviewed papers was how difficult this subject is and possibly how little we all know.