Abortion and Crime: Sailer Responds to Steven "Freakonomics" Levitt's Response

Having gotten softball reviews from the national press, Steven D. Levitt's hometown newspaper, the Chicago Sun-Times, finally called his bluff on his abortion-cut-crime theory. Perhaps that's why he has now deigned to answer some of his critics.

Levitt makes no attempt to defend his battered "wantedness" theory of how abortion is supposed to cut crime, but instead merely restates his interpretations of the historical data.

Let me answer Levitt's initial arguments:

First, let's start by reviewing the basic facts that support the Donohue-Levitt hypothesis that legalized abortion in the 1970s explains a substantial part of the crime decline in the 1990s:

Right away, Levitt tries to rig the deck in his favor by defining the question as whether legalized abortion caused in large measure the drop in crime in the 1990s. The answers to that question can be either "Yes" or "No." Neutral observers tend to react to disagreements by splitting the difference. So, in this case, the natural assumption of people who don't want to wade through all the arguments would be: "It had probably some effect on cutting crime." And that means Levitt is seen as being more or less right.

In reality, however, we should be looking at the more general question: "What was the effect of legalizing abortion on serious crime from the 1970s onward?" The answers to this question can be "It cut crime," "It had no effect," or "It increased crime." Given those three options, observers are likely to split the difference and say: "It had no effect." So, Levitt cleverly tries to narrow the focus to where he'll be given the benefit of the doubt.

1) Five states legalized abortion three years before Roe v. Wade. Crime started falling three years earlier in these states, with property crime (done by younger people) falling before violent crime.

Actually, it's much more complicated than this because ten other states "liberalized" abortion laws in 1970. But, for now, let's just look at those five that outright legalized it. Two of those states are Hawaii and Alaska, which are hardly representative of the rest of the country, and a third is far-off Washington state. The big two early legalizers were California and New York. Yet, Levitt admits in another place in his response:

The homicide rate of young males (especially young Black males) temporarily skyrocketed in the late 1980s, especially in urban centers like Los Angeles, New York City, and Washington, DC...

And Washington D.C. largely had de facto legal abortion from 1970, too. So, according to Levitt, those three cities are where, purely by coincidence, the teen crack wars broke out about 17-18 years after abortion was legalized. Now that correlation between legalizing abortion and increased teen violence isn't proof of anything, but it obviously raises the question of whether or not legalized abortion contributed to crack killings. But that's not an issue Levitt wants to touch.

No, what Levitt wants to talk about is not what happened about 17-21 years after abortion was legalized, but what happened about 22-24 years later. The longer the lag between the effect and the hypothesized cause, the more Dr. Levitt trusts it!

Of course, that's terrible science. If abortion has an effect, it should show up earlier in life, before more adult experience has intervened. Similarly, the effect of legalizing abortion is most trustworthy earlier in history, before too many intervening changes have gotten in the way of a clean read.

And the violence grew fastest among the demographic group with the highest abortion rate: blacks.

Levitt also claims "with property crime (done by younger people) falling before violent crime." Look, property crime apparently began falling in the mid 1970s, probably due to "target-hardening." Property crime statistics are less reliable than violent crime statistics because the victims frequently don't get a look at the perpetrator so we don't know the age, and the cops vary tremendously over time and place in how much they care about catching property criminals .

The FBI provides much better quality data on homicides (which cops care about a lot) and serious violent crimes (which "includes rape, robbery, aggravated assault, and homicide") from the annual crime victimization survey. So, that's two totally different methodologies measuring two kinds of crime, but they agree closely on the history, which is that the first generation born after legalized abortion were the most violent teens in the history of American crime statistics.

Levitt is misleading when he implies that younger people don't commit much violence. The worst year for serious violent crime by ages 12-17 was 1993, when this cohort (all born after abortion was legalized) committed 27% of all serious violent crimes. (1994 was a bad year too). Moreover, children under 18 accounted for over half of the big increase in serious violent crimes between the mid-1980s and 1993.

I'm going to lump the next three of Levitt's arguments together:

2) After abortion was legalized, the availability of abortions differed dramatically across states. In some states like North Dakota and in parts of the deep South, it was virtually impossible to get an abortion even after Roe v. Wade. If one compares states that had high abortion rates in the mid 1970s to states that had low abortion rates in the mid 1970s, you see the following patterns with crime. For the period from 1973-1988, the two sets of states (high abortion states and low abortion states) have nearly identical crime patterns. Note, that this is a period before the generations exposed to legalized abortion are old enough to do much crime. So this is exactly what the Donohue-Levitt theory predicts. But from the period 1985-1997, when the post Roe cohort is reaching peak crime ages, the high abortion states see a decline in crime of 30% relative to the low abortion states. Our original data ended in 1997. If one updated the study, the results would be similar.)

3) All of the decline in crime from 1985-1997 experienced by high abortion states relative to low abortion states is concentrated among the age groups born after Roe v. Wade. For people born before abortion legalization, there is no difference in the crime patterns for high abortion and low abortion states, just as the Donohue-Levitt theory predicts.

4) When we compare arrest rates of people born in the same state, just before and just after abortion legalization, we once again see the identical pattern of lower arrest rates for those born after legalization than before.

Data by state is extremely tricky because women travel across state lines to get abortions. New Jersey is a classic example. Arrest rates vary by state and change over time too in all sorts of ways.

Further, states like North Dakota are largely irrelevant to national crime trends.

Here, I'm going to turn to a not-yet-published paper by Ted Joyce, an economist with the National Bureau of Economic Research and Baruch College, City University of New York. In this paper, Joyce tries hard to remove the effects of crack crime from the data. Here's part of his abstract:

In this paper, I conduct a number of new analyses intended to address [Levitt and Donohue's] criticisms of my earlier work.

First, I examine closely the effects of changes in abortion rates between 1971 and 1974. Changes in abortion rates during this period were dramatic, varied widely by state, had a demonstrable effect on fertility, and were more plausibly exogenous than changes in the late 1970s and early 1980s. If abortion reduced crime, crime should have fallen sharply as these post-legalization cohorts reached their late teens and early 20s, the peak ages of criminal involvement.

It did not.

Second, I conduct separate estimates for whites and blacks because the effect of legalized abortion on crime should have been much larger for blacks than whites, since the effect of legalization of abortion on the fertility rates of blacks was much larger.

There was little race difference in the reduction in crime.

Finally, I compare changes in homicide rates before and after legalization of abortion, within states, by single year of age. The analysis of older adults is compelling because they were largely unaffected by the crack-cocaine epidemic, which was a potentially important confounding factor in earlier estimates.

These analyses provide little evidence that legalized abortion reduced crime.

And here is the abstract of the not-yet-published paper by economists John R. Lott and John Whitley replying to Levitt and Donohue. They find that abortion increased the murder rate:

Previous empirical work linking abortions and crime [i.e., Levitt and Donohue's] has assumed, with the exception of five states, that no abortions took place prior to the Roe v. Wade decision in January 1973. In fact, abortion data from the Centers for Disease Control indicate that states which allowed abortions prior to the Roe v. Wade only when the life or health of the mother was in danger actually had higher abortion rates than some states where it was legal. The use of data from the Supplemental Homicide Report also allows the direct linkage between the current age of the murderer and the abortion rate when those murders were born.).

One more abortion per 1,000 females age 15-44 (i.e., about four percent of the average) is associated with between a 0.12 to 0.9 percent increase in murders in any given year. Similar estimates are obtained using abortions per 1,000 live births. Linear estimates indicate increased annual victimization costs by at least $3.2 billion.

One of the differences between Joyce's approach and Lott-Whitley's approach is that Joyce assumes the crack war of the late 1980s and early 1990s waged in sizable measure by urban teens born soon after the rise in urban abortion rates in the early 1970s was an "exogenous" (independent) event while Lott-Whitley are assuming the legalization of abortion and the subsequent growth in the murder rate during the crack wars might be related.

Both Joyce's and Lott-Whitley's approaches seem defensible ways to explore a hugely complex social phenomenon. What's not reasonable is Levitt's cherry-picking approach, in which he assumes that legalizing abortion isn't responsible for any of the increase in murders at the beginning of the crack wars but is responsible for some of the decrease of murders at the end of the crack wars. That's called having your cake and eating it too.

Levitt continues:

5) The evidence from Canada, Australia, and Romania also support the hypothesis that abortion reduces crime.

So that's why Russia, which had such an enormous abortion rate during the Soviet years, has no problem these days with criminals!

One of the weirder passages in Freakonomics is the section where Levitt implies that if only Romanian leader Nicolae Ceaucescu hadn't outlawed abortion he'd be ruling still, instead of getting overthrown by all those unruly unaborted criminals.

May I point out that Canada, Australia, and Romania are not the U.S., and differ from America in some obvious ways relevant to crime? And are these foreign studies of a higher quality than Levitt's, or just imitations of his work?

6) Studies have shown a reduction in infanticide, teen age drug use, and teen age childbearing consistent with the theory that abortion will reduce other social ills similar to crime.

And studies have shown an increase in venereal disease, illegitimacy, and a decrease in adoption due to legalizing abortion.

And do Levitt's examples even hold water historically? He claims abortion reduces teen age drug use, but when the first generation born after legalization reached their teen years, we had a crack epidemic. And the sharp decline in teen childbearing didn't happen until the 1990s, when we also had a sharp decline in abortion rates.

So, Levitt hasn't gotten very far beyond where his slapdash original theory back in 1999 had got him. If large claims require large evidence, then he's still a long, long way from meeting the burden of proof.

P.S. Some clever analyst should figure out why so many commentators desperately want Levitt's theory to be true.

My published articles are archived at iSteve.com -- Steve Sailer