Abortion and Health

Back To The Future: Roe Versus Wade As An Instrument

Advances in research design and insistence on greater rigor in the application of instrumental variables greatly has improved applied economics since the late 1990s. The literature on abortion and health was similarly affected. Researchers realized that changes in policies regarding Medicaid financing of abortion, PI laws, and mandatory delay statutes did not alter the timing or number of children sufficiently to power analyses of maternal health and child well-being. Thus, researchers returned to abortion legalization in the US and abroad in which there was greater evidence of changes in fertility associated with the more dramatic fall in the price of fertility control. Two papers led the way. In the first, researchers used the legalization of abortion in the US as an instrument for teen childbearing in models of schooling and labor market outcomes. With data from the 1980 Public Use Microdata Samples (PUMS) from the US Census , the authors showed that the longer a teen was exposed to legalized abortion, the lower the likelihood of becoming a teen mother or married before the age of 20 years. The impact of legalization on childbearing was substantially greater among Blacks than among Whites. The racial pattern persisted in the reduced form models of high school graduation, college attendance, and labor force participation. The authors then used exposure to legalized abortion as an instrument for Black teen out-of-wedlock childbearing in models of school, work, and poverty. They did not pursue a similar analysis for Whites because there was no reduced-form evidence to support it. The results were large. Teen motherhood reduced college entrance by 20 percentage points when estimated by ordinary least squares (OLS) but by 56 percentage points when estimated by TSLS. Differences between OLS and TSLS for labor force participation were even greater. The authors concluded that on balance the data suggested that abortion legalization increased schooling and employment among Black women. Nevertheless, the authors noted that despite the change teen fertility, it was difficult to detect the consequences of teen childbearing even with large samples from the US Census. They go on to encourage researchers to find other sources of exogenous variation in fertility in order to identify the effects of teen childbearing on downstream outcomes.

In the same year, Gruber et al. (1999) published an important paper entitled, ‘Abortion Legalization and Child Living Circumstances: Who is the Marginal Child?’ They too used the 1980 PUMS to analyze changes in the health and well-being of cohorts born before and after the legalization of abortion. Legalized abortion, they argued, changed the distribution of women who gave birth which, in turn, altered the average circumstances under which subsequent cohorts of children were raised. Improved circumstances after Roe would be evidence of positive selection. They also noted that increases in the average circumstance of a cohort implied that the conditions of the marginal child, the one who would have been born had the women not ended the pregnancy, would have to have been worse for average well-being to rise.

The authors estimated both reduced-form and structural models of child well-being using the two phases of abortion legalization in the early 1970s. The reduced form showed that the average change in each outcome was associated with increased access to legalized abortion. In these regressions, the authors found that the rate of low birth weight birth associated with pre-Roe legalization fell from 7.7% to 7.6%, whereas infant mortality dropped from 1.9 per 1000 live births to 1.86 per 1000. The reduced-form results also suggested that children after legalization were less likely to live with a single parent, to live in poverty, or to receive welfare. Effect sizes were approximately 3% of the mean for each outcome. Changes in well-being associated with the marginal child were much larger. The TSLS estimates suggest that the probability of dying in the first year was 40% greater for the marginal child, although the rate of low birth weight was 14% greater. The results by race were less consistent. Although the impact of abortion legalization on the birth rates of non-Whites was twice as large as on Whites, none of the reduced-form estimates of changes in non-White living circumstances or infant health were associated with abortion legalization. The same was true for the marginal child as estimated by TSLS.

In a sequel to the marginal child, the researchers analyzed the impact of abortion legalization on adult outcomes with data from the 2000 census. As before, cohorts pertained to individuals born between 1965 and 1979 and who were 21 to 35 years of age as of the 2000 census. As in Gruber et al. (1999), they regressed measures of well-being on the two-phase of legalized abortion in 1970s. The outcomes include the percent in poverty, in single-parent household, on welfare, incarcerated, employed, a high school dropout and a non-college graduate. In only 2 of the 7 outcomes was there an association with early legalization and in only 3 of the outcomes was there any association with all phases of legalization. In the TSLS models in which each outcome was regressed on the birth rate instrumented by the cost of abortion, less than half the outcomes were associated with worse conditions for the marginal child.

The ‘marginal child’ papers provided a novel and more general empirical framework for estimating the impact of abortion legalization on the child that was not born. Instead, of only associating abortion legalization with average changes in affected cohorts, these authors provided a clever method of estimating the counterfactual outcome. There are, however, important limitations to the empirical work and results. First, in both papers, the authors could not separate age from period effects because they only had data on each outcome at a single point in time. The inclusion of state-specific quadratics in age may have accounted for some of the variation in period effects, but period effects can be very powerful determinants of crime, employment, single parenthood, etc. Second, a lack of selection effects among non-Whites is difficult to explain, especially in light of other work that demonstrated robust effects of abortion legalization on education and employment among Black women. Not only did the legalization of abortion affect non-White fertility more than Whites, but also the non-Whites are more likely to be incarcerated, on welfare, single parents, and high school dropouts. If abortion is improving the circumstances of White children, indicative of positive selection, why would an even greater relative and absolute decrease in fertility among non-Whites not affect their circumstances even more? Either there is negative selection among non-Whites or unmeasured period effects are confounding estimates. Third, it is difficult to interpret the first-stage estimates in this study. There are many interactions in which the omitted category is obscure and the exclusion restrictions are hard to justify. Despite these issues, the marginal child papers were an important advance in the literature.

Abortion And Crime

Clearly, the most sensational association with abortion came from Donohue and Levitt’s (2001) paper linking the legalization of abortion to the precipitous drop in crime. The mechanism was not novel. Citing Grossman and Jacobowitz (1981) and Gruber et al. (1999), Donohue and Levitt (2001) argued that the child who was not born would have grown up in worse living circumstances, received less parental support, and as a result would have been more prone to criminal behavior as a teen and adult. The paper received remarkable attention in the popular press and its basic finding reached an even broader audience with the publication of Levitt and Dubner (2005) book, Freakonomics. The empirics were simple. The authors regressed total crime rates on lags of the abortion rate adjusted for state and year-fixed effects. They also regressed age-specific arrest rates for those of 15–24 years of age on the lagged abortion rate. In both specifications they found that abortion rates could explain upward of 50% of the decrease in crime in 1990s.

The results were quickly challenged. It was straightforward to show that their story did not line up with simple plots of age-specific homicide rates (Joyce, 2009). For instance, homicide rates soared between 1985 and 1992 among young, African-American males in large urban areas and then dropped almost as precipitously thereafter. There were relatively modest changes in murder rates among other groups who were also exposed to legalized abortion in utero. Most criminologists attributed the increases in homicides to the crack cocaine epidemic which spurred a rise in gang violence. However, no credible data on crack-cocaine use by state, year, and age existed which created a potentially significant omitted variable problem. This was aptly demonstrated by two economists who first replicated Donohue and Levitt’s regressions but then added state–year interactions. The association with the abortion rate fell by 50–60%. Another economist used a triple difference strategy to eliminate the confounding effect of crack cocaine by comparing the crime rates of 19-year olds born before abortion was legalized to that of 17-year olds born just after. Both groups experienced the same period effects (i.e., the crack-cocaine epidemic) but only the younger cohort was exposed to legalized abortion in utero (Joyce, 2009). Joyce found no association between legalized abortion and crime. A full airing of the debate is beyond the scope of this article. Regardless of the ultimate judgment of the Donohue and Levitt thesis, their work stimulated further research. Economists examined the association between legalized abortion and drug use, whereas others correlated legalized abortion with teen pregnancy, a female proxy for delinquent behavior. Economists also convincingly linked legalized abortion to sexually transmitted diseases. The strength of these papers rested on use of abortion legalization as the identifying source of variation. Legalization, much more than subsequent policies regulating abortion, had a clear, measurable impact on fertility. And yet the challenge in all these papers is identification of a cohort effect amidst often powerful period effects. In the case of abortion and crime, it was the crack epidemic of the late 1980s and the early 1990s that confounded estimates. With teen pregnancy, it was welfare reform and the expanding economy in the 1990s. Thus, studies that analyzed changes in outcomes around the time of legalization are more convincing because the confounding from period effects is arguably more easily controlled. Even with more proximate outcomes, the health effects of abortion are exceedingly difficult to identify. Recall that Gruber et al. (1999) found exceedingly modest declines in low birth weight and infant mortality among cohorts exposed versus unexposed to legalized abortion. In fact, more recent research suggests that the 1–2% declines in their paper are probably too small to be detected with the proper adjustment of the standard errors.

One paper illustrates just how difficult it can be to associate even dramatic changes in the cost of fertility control with wellbeing (Pop-Eleches, 2006). In December of 1966, Romania outlawed abortion and all methods of fertility control in response to the declining birth rate in the country. The result was an immediate doubling of the birth rate from 14.3 births per 1000 population to 27.4 a year later. The author used this unprecedented fertility shock to estimate its impact on the educational and labor market outcomes of the birth cohorts born just before and after the ban. The overall result was an increase in well-being, a result directly at odds with the US experience. The seemingly contradictory finding resulted from the positive increase in childbearing among families of higher socioeconomic status. Once the author adjusted for the composition change, exposure to the ban was associated with decreased schooling. The author interpreted the latter effect as the negative impact of unwantedness. The author found no association with labor market outcomes. The author also reported a 27% increase in infant mortality and a 30% increase in low birth weight. The changes in infant health were relatively short lived and thus may have been caused in part by lack of prenatal and obstetric services.

The increase in fertility in Romania was 20 times the decrease observed with abortion legalization in US and yet, even with such a huge jump in the birth rate, changes in well-being were somewhat modest or relatively short lived. This underscores the point made previously: detecting cohort effects on downstream outcomes is extremely challenging. Without large, exogenous shocks, distinguishing cohort from age and period effects may exceed researchers’ ability to detect them with extant data.

Addiction and Health