Legalized abortion may lead to reduced crime through any of three channels. The first channel is simply a smaller cohort size. If abortion reduces the number of births, when that cohort reaches the late teens and twenties, there will be fewer young males, and thus less crime. Abortion may also lower per capita offending rates in affected cohorts. The children may on average be less criminal due to either a ‘selective-abortion effect or an ‘improved-environment’ effect. The selective-abortion channel will operate if the women who have abortions are the most at risk to give birth to children who would engage in criminal activity. . . . The improved-environment effect may be present if women use abortion to optimize the timing of childbearing.
—John J. Donohue, III, and Steven D. Levitt
Donohue and Levitt’s contention that legal abortions in the 1970s explain half the fall in the crime rate during the 1990s is couched in what appears to be a scientific study. Yet a closer look at the data on both abortion and crime contradicts their conclusions. Equally important, their argument ignores the new social dynamics that emerged in the wake of the legalization of abortion. The “selective abortion” effect that stands at the center of their thesis is nullified by the fact that about two-thirds of the rise in teenage pregnancies following the legalization of abortion were the result of behavioral changes. In other words, legal abortion was a cause of the bad socioeconomic environment, not its cure. As documented in my feature essay, by sharply cutting the net marriage rate and increasing out-of-wedlock births, legal abortion caused a sharp fall in household living standards, especially among “teenagers, unmarried women, and African-Americans,” groups that Donohue and Levitt claim will have children most likely to engage in criminal activity.
Among the dynamics brought on by abortion and overlooked by Donohue and Levitt is what might be called the “worm-in-the-apple effect.” After doing away with an “unwanted” child through abortion, the parents belatedly discover-that they, too, are unwanted: the woman is unwanted as a wife or mother; the man is unwanted as a husband or father. When the child is unwanted by the father but wanted by the mother, it usually means the mother is also unwanted by the father. The typical result: an out-of-wedlock birth and a home that is “broken” from its conception. In either case, the less men are occupied with producing and economically supporting life, the more they are likely to injure life. Donohue and Levitt may coyly argue that “recent abortions will not have any direct impact on crime today since infants commit little crime,”1 but many fathers of aborted children do.
The age distribution of persons arrested (four-fifths of whom are men) matches almost exactly the age distribution of women seeking abortions, especially after allowing for the fact that young men are typically two years older than their female sexual partners.2 Among persons arrested in 1995, 18.3 percent were under the age of eighteen years; 26.1 percent between the ages of eighteen and twenty-four years; 47.4 percent between the ages of twenty-four and forty-four years; and 8.2 percent older than age forty-four.3 Among women undergoing abortions in 1995, 20.2 percent were nineteen years old or younger; 32.4 percent between the ages of twenty and twenty-four years; 47.4 percent between the ages of twenty-five and forty-four years; and a fraction of a percent older than forty-four.4 This means that convoluted speculations about what crimes African-American children might commit a couple of decades from now, if they are permitted to be born, are meaningless. The relation between abortion today and crime today is direct and brutal. Although yet to be statistically quantified, most crimes are committed by men the age of the fathers of aborted children; a man who has been party to either killing or abandoning his own child is more likely to harm other human beings.
To claim their results, Donohue and Levitt must torture a ridiculously short snippet of available data. Comprehensive annual data on crime begin in 1957; on homicide, in 1900. But the authors try to explain only the data from 1985 to 1997 by using abortion data fifteen to twenty years before, arbitrarily ruling out any contemporaneous effect of abortion on crime. They even concede:
It is impossible using the data alone to distinguish the impact or 1970’s abortions on current crime rates from the impact of 1990’s abortions on current crime rates. Put another way, we obtain similar results regardless of whether we include 1970’s abortion rates or 1990s abortion rates, but when both are included multicollinearity leads to enormous standard errors. Consequently, it must be recognized that our interpretation of the results relies on the assumption that there will be a 15–20 year lag before abortion materially affects crime.5
This assumption represents a fatal flaw in their analysis. A strongly positive contemporaneous relation between abortion and crime exists when the crime and abortion rates are measured against other social indicators. The total fertility rate strongly correlates inversely with the crime rate. Since at least 83 percent of past legal abortions reduced the total fertility rate, a strong positive relationship exists between the current abortion rate and the current crime rate. In addition, the share of the population on welfare (and general assistance, which was important before 1960) strongly correlates with the crime rate.
Donohue and Levitt ignore these factors, comparing the changes in abortion between 1973 and 1976 with the change in various crime rates between 1985 and 1997. Why did they compare a three-year change in abortion (starting when abortions had already reached about half their peak number) with a twelve-year change in crime? Perhaps they discovered that using comparable time periods–thus starting with zero abortions–would have yielded mathematically nonsensical results, revealing that the study is “improperly specified.” Even so, their “fit” of the data is underwhelming. The data consist essentially of a single pair of data points–the 1973-76 change in abortion and the 1985-97 change in the crime rate–for each of the fifty states and the District of Columbia. While a 100 percent statistical correlation must always be found between any single pair of data points, the Donohue-Levitt study of fifty-one simultaneous pairs could find only a 31 percent correlation between lagged abortion and the violent crime rate, a 39 percent correlation with the property crime rate, and a miserable 14 percent correlation with the homicide rate. Even by adding seven extra variables, the study could still explain only 37 percent of the variation in the violent crime rate, 65 percent of the variation in the property crime rate, and 45 percent of the variation in the homicide rate.
Analysts looking for more satisfying explanations for changes in the crime rate ought to look at variables Donohue and Levitt ignore. One single explanatory variable, the total fertility rate, can account for 84 percent of the annual variation in the violent crime rate, 94 percent of the total crime rate between 1957 and 1997. Two explanatory variables, the total fertility rate and the share of the U.S. population on welfare, can explain 90 percent of the variation in the violent crime rate, 92 percent of the homicide rate, and 96 percent of the total crime rate since 1957. The regression results indicate that a 1 percent rise in the total fertility rate is associated with a 1.4 percent drop in the total crime rate and that a 1 percent fall in the population on welfare is associated with a 0.85 percent fall in the total crime rate.6 Therefore, the rise of the total fertility rate coupled with the sharp decline of the welfare population accounts far better empirically for the recent drop in the crime rate. Conversely, the sharp decline in the total fertility rate coupled with the sharp rise in the welfare population in the 1960s and 1970s explains the sharp rise in the crime rate during the same period.
1. John J. Donohue, III, and Steven D. Levitt, “Legalized Abortion and Crime,” unpublished paper, p. 17.
2. Statistical Abstract of the United States: 1998, p. 112.
3. Ibid., p. 220.
4. Ibid., p. 92.
5. Donohue and Levitt, “Legalized Abortion and Crime,” p. 22.
6. The standard errors were 0.33 percent for the total fertility rate and 0.25 percent for the welfare population, confirming that the statistical results are robust.