Abstract
Abortion is a safe and common medical procedure. Roughly one in four women in the United States will have an abortion before the end of her reproductive years. Because of how common this experience is and how rapidly abortion policy is shifting, understanding the relationship between abortion and women’s socioeconomic futures is well worth exploring. Extant research has demonstrated that the transition to parenthood is a critical inflection point in women’s socioeconomic trajectories, often leading to poorer outcomes. In this article, we connect previous sociological work elucidating mechanisms of socioeconomic stratification and gender by considering the relationship between abortion use and access and future socioeconomic outcomes such as education, income, and financial stability—as measured by several measures, including evictions, debt, ability to pay bills, and a separate index of economic instability. We use national longitudinal survey data to assess socioeconomic outcomes associated with abortion using two statistical approaches. We find that women who lived in a location with fewer abortion restrictions in adolescence, and women who had an abortion, compared to a live birth, in adolescence, are more likely to have graduated from college, have higher incomes, and have greater financial stability at two time-points over an almost 25-year period. Our results provide evidence that policy environments allowing access to abortion, and teenagers having the option to use abortion to avoid early parenthood, are important axes along which women’s economic lives are shaped. Our research implies that the widespread abortion bans and restrictions in the United States are likely to lead to lower educational attainment and adult economic stability among women living under such restrictions, as compared to women in locations with better access to abortion.
Abortion is common in the United States: about 25 percent of women 1 will have an abortion by age 45 (Jones and Jerman 2022; Luker 1985; Watson 2018). Given the links between pregnancy, childrearing, and women’s socioeconomic status (Benard and Correll 2010; Correll, Benard, and Paik 2007), there are likely significant economic and educational outcomes associated with having an abortion or access to abortion. Indeed, some researchers have argued that much of the gender gap in wages and occupational prestige is due to the transition to motherhood (Calarco 2024; England 2005). For example, previous research has shown that in experimental settings, respondents rate mothers as less competent job applicants and recommend lower starting salaries than they do for non-mothers (Correll et al. 2007). Longitudinal survey data repeatedly shows that having children results in decreases in labor force participation and wages and in fewer promotions for women (Budig and Hodges 2010; Kahn, Garcia-Manglano, and Bianchi 2014; Wilde, Batchelder, and Ellwood 2010), a distinction that persists over the life course (Goldin, Kerr, and Olivetti 2024). Indeed, having even one child is associated with a 16 percent decrease in social security benefits among mothers (Rutledge, Zulkarnain, and King 2017).
Yet, despite the frequency of abortion and the powerful socioeconomic consequences of childbearing on women’s lives, sociologists have paid relatively little attention to the relationships between abortion and women’s economic lives. In this project, we ask: What are the socioeconomic outcomes of having an abortion or, conversely, of living in a restrictive abortion environment?
This project is especially important in the current abortion policy climate. In June 2022, in Dobbs v. Jackson Women’s Health Organization, the U.S. Supreme Court overturned almost 50 years of legal precedent, which mandated, at the federal level, the constitutional right to abortion. This case resulted in many states banning abortion and many others creating a legal context that makes abortion inaccessible or very difficult to obtain. As a result, over 21 million women of reproductive age live in states that ban abortion (Putterman 2024). As such, this massive policy shift has created a stratified system of access to abortion. This stratification of access to this common medical procedure—abortion—motivates our interest in the possible resultant socioeconomic stratification.
To assess long-term socioeconomic outcomes associated with abortion in the United States, we use a national, probability-based dataset—the National Longitudinal Study of Adolescent to Adult Health (Add Health). To address methodological challenges related to abortion research, we use two complementary approaches. The first approach assesses the relationship between the restrictiveness of abortion environments—at the state and county levels—and, later, individual-level socioeconomic outcomes. In the second approach, using counterfactual matching techniques, we compare midlife socioeconomic outcomes of teens who reported a live birth versus those who reported a pregnancy that ended in abortion (Becker and Ichino 2002; Rubin 1997).
In the first approach, we find that women who—when under age 20—lived in regions with less restrictive abortion environments were more likely to have graduated from college, to have higher income, and to score lower on multiple indicators of financial instability at various points between ages 22 to 43, than were comparable women who lived in more restrictive abortion environments. In the second approach, we find that women who had an abortion to avoid teen parenthood were more likely to have graduated from high school and college, to have a higher income, and to report lower economic instability at various points between ages 22 to 43, compared to women who had a live birth as a teen.
A Gap in the Sociological Study of Reproduction and Socioeconomic Outcomes
Control of reproductive capacity is a primary site of the production and maintenance of socioeconomic inequality and is especially central to gender, racial, and economic stratification (Beisel and Kay 2004; Calarco 2024; Dobbins-Harris 2017; Everett et al. 2022; Homan 2019; Littlejohn 2021; Luker 1985; Luna 2020; Roberts 2014; Ross and Solinger 2017; White 1999). For example, scholars and activists in the reproductive justice and Black feminism literatures have made explicit connections between the control of biological reproduction and socioeconomic outcomes for some time (Collins 2022; Roberts 2014; Ross and Solinger 2017). This work shows how restricting access to reproductive autonomy is a form of social control in general and that, more specifically, reproductive autonomy is a key determinant of access to educational and economic resources (Collins 2022; Goldin and Katz 2002; Roberts 2014; Ross and Solinger 2017; White 1999). Most relevant to our study, scholars in this tradition note that abortion is not just about individual “choice”; it is also a key component of understanding people’s lives more broadly—including the socioeconomic context that shapes their opportunities to make a range of choices (Ross and Solinger 2017).
In a related line of work, multiple studies have linked access to contraception to the reduction of gender inequality, especially in earnings, workforce participation, and education. For example, researchers have argued that U.S. Food and Drug Administration (FDA) approval of oral contraceptives in 1960 led to a transformation in the labor market by increasing levels of educational attainment for women, delaying the age at first marriage, and facilitating women’s greater workforce participation (Bailey 2006; Goldin and Katz 2002). In another example, two recent studies found that expanding contraceptive access in Colorado was associated with a statistically significant increase in the number of women graduating from high school (1.6 percent) (Stevenson et al. 2021) and graduating from college (1.8 to 3.5 percent) (Yeatman et al. 2022), compared to older cohorts without expanded access.
From the perspective of women seeking abortion, when women are asked, the most common reasons they cite are socioeconomic concerns (Finer et al. 2005). Such reasons often include concerns about how having another child will interfere with their ability to provide economically for the children they already have (Cohen and Joffe 2020; Finer et al. 2005; Foster 2020). In many cases, seeking an abortion is an attempt to avoid or mitigate poverty through continuing education or paid work (Borgmann and Weiss 2003; Cohen and Joffe 2020; Foster 2020). Relatedly, abortion access is a critical mechanism through which women delay marriage and childbearing to pursue educational and career goals (Myers 2012).
All of the research reviewed in this section is consistent with work by gender and family scholars documenting the effects of the transition to parenthood on women’s careers and income, referred to as the “motherhood penalty.” The transition to motherhood is a key economic inflection point for many women that results in fewer job offers and lower salaries (Correll et al. 2007; Jee, Misra, and Murray-Close 2019), particularly while their children are young (Kahn et al. 2014). However, the same results are not found among men. In fact, for some men, parenthood is associated with better career opportunities and salaries, resulting in a fatherhood bonus (Andersen 2018; Correll et al. 2007).
Despite this robust body of literature on the relationship between reproduction and socioeconomic inequality, sociology has not adequately centered the study of abortion as a fundamental issue related to socioeconomic inequality. Research and theory on abortion are often siloed in studies of gender or health, if addressed at all. We note four facts that inform our motivation to push sociological research on abortion into the center of the discipline’s study of socioeconomic inequality: (1) the causes and consequences of socioeconomic stratification are central topics in sociology (see, e.g., Collins 2022; Desmond 2016; Grusky and MacLean 2016; Killewald, Pfeffer, and Schachner 2017; Mize 2016; Phelan, Link, and Tehranifar 2010; Ridgeway 2011; Torche 2011; Zheng and Weeden 2023); (2) people use abortion to manage their socioeconomic position, and especially to protect themselves and their families from poverty (Cohen and Joffe 2020; Finer et al. 2005; Foster 2020; Watson 2018); (3) abortion is very common (Watson 2018); and (4) abortion is central to U.S. politics, with different policies across states leading to disparities in access to abortion (Watson 2018).
In summary, based on previous research, we theorize that understanding the socioeconomic consequences of abortion and its inequitable restriction is foundational to understanding the production of inequality because women’s control of their reproductive capacity is foundational to their socioeconomic status (Banerjee 2023; Beisel and Kay 2004; Borgmann and Weiss 2003; Dobbins-Harris 2017; Luker 1985; Roberts 2014; Ross and Solinger 2017; White 1999). Our project takes an important step toward empirically assessing the role of abortion in socioeconomic outcomes and stratification.
Abortion in the United States
U.S. policymakers treat abortion as medically exceptional: it is more difficult to obtain and more strictly regulated than other medical procedures that are comparable in complexity and safety (Cohen and Joffe 2020; Fisher, Royer, and White 2018). Treatment of abortion as medically exceptional limits access to abortion for many people (Cohen and Joffe 2020). It is difficult to succinctly summarize the many barriers to abortion access in the United States, as such barriers are extensive and changing rapidly. Barriers include, but are not limited to, abortion bans in many states, difficulties accessing accurate policy and medical information, travel distances and related costs, appointment scheduling in the context of mandatory wait periods and required counseling sessions, and harassment of clinics and abortion providers (Doran and Nancarrow 2015; Henshaw 1995; Janiak et al. 2014; Jerman et al. 2017; Kavanaugh, Jerman, and Frohwirth 2019; Margo et al. 2016; Smith and Cameron 2019; Upadhyay, Cartwright, and Grossman 2022). Even as we write this article, the state of abortion access in the United States continues to shift. We provide some examples of how abortion access was restricted as of spring 2023, but we urge readers to remember that, if anything, we have understated the barriers to abortion access.
Legal, Policy, and Bureaucratic Barriers
In June 2022, the Supreme Court overturned the nearly 50-year precedent of Roe v. Wade, which mandated, at the federal level, the constitutional right to abortion. Consequently, as of March 2023, abortion is illegal, or effectively illegal, in 10 to 12 states (Center for Reproductive Rights n.d.). In addition, as of March 2023, another 12 states had policies, legislators, or judges that were extremely hostile toward the legal right to an abortion (Center for Reproductive Rights n.d.), resulting in a total of 24 states where abortion is either illegal or laws are extremely hostile to abortion rights—almost half of the U.S. states.
Even in many states where abortion remains legal, access to abortion is subject to medically unnecessary administrative burdens that make the provision of abortion extremely bureaucratically and financially burdensome (Cohen and Joffe 2020; Dickman, White, and Grossman 2021; Kacanek et al. 2010; Moynihan, Herd, and Harvey 2015; Razon et al. 2022). Such policies are often referred to as Targeted Restrictions on Abortion Providers (TRAP) laws (Guttmacher Institute 2017). These include laws that force medical professionals to administer abortions in a manner that is not medically necessary and, in many cases, make abortion a higher-risk procedure than it otherwise would be (Cohen and Joffe 2020). Such laws also include requiring abortion providers to have medically unnecessary hospital admitting privileges and to adhere to ambulatory surgical center standards—even in clinics where only medication abortion is administered. In one example, Utah passed a bill requiring all abortion procedures to be completed in a hospital setting beginning May 2023, even though hospitals performed just 3 percent of all abortions in the state of Utah.
Such TRAP laws achieve their desired effect of limiting access to abortion. For example, in 2013 and 2014, a decade before the Dobbs decision, “over half of the abortion facilities in Texas closed due to implementation of a . . . TRAP law” (Bernstein and Jones 2019:16). Abortion bans and other anti-abortion laws, policies, and bureaucracies directly affect access to abortion, with many counties, states, and regions of the United States having no legal abortion providers at all (Banerjee 2023).
Legislators and policymakers have also imposed medically unnecessary regulations on the distribution of mifepristone, a commonly used medication for abortion that has passed rigorous FDA review and has been used safely millions of times (Razon et al. 2022; Thompson et al. 2021). Doubling down on this strategy, an anti-abortion group is attempting to revoke FDA approval for mifepristone through the courts, an outcome that would drastically reduce access to abortion in all 50 states (Leonard 2022; Sobel, Salganicoff, and Felix 2023; The Daily 2023). As of March 2023, some pharmacies have stated they will not dispense mifepristone—even though it is legal to do so—because of threats of legal action from anti-abortion state attorneys general (Ollstein 2023).
Numerous other policies designed to limit access to abortion have been enacted. Such policies include medically unadvised waiting periods, consent laws, and financial burdens that slow the process of obtaining an abortion—sometimes to the point where abortion is no longer a legal, medically advised, or desired option for the patient (Cohen and Joffe 2020).
Cost is also a barrier to access, as with any other type of healthcare (Cohen and Joffe 2020). However, unlike other kinds of healthcare, abortion is not covered by federally-funded health insurance, many state-funded health insurance programs, nor by some private insurance providers—making the cost of the procedure alone an insurmountable barrier for some patients (Cohen and Joffe 2020; Miller, Wherry, and Foster 2023). For patients who do have the ability to pay for the medical procedure out of pocket, the cost may still be a barrier if they live in a location where abortion is illegal or very difficult to obtain because such restrictions necessitate additional travel-related costs (Cohen and Joffe 2020). For example, sources estimate that out-of-state travel for abortion care costs between $1,200 and $30,000 (Jetelina 2023; McCann 2022; Planned Parenthood 2024).
These layers of administrative burden and a quickly shifting political landscape make it very difficult for patients and providers to keep track of what is legal and what is not (Banerjee 2023; Cohen and Joffe 2020; Miller et al. 2023). This murky and constantly changing legal landscape around abortion affects access even in locations where abortion is fully legal, and especially in the many states where abortion is legal but is not explicitly protected under the law (Cohen and Joffe 2020; Razon et al. 2022).
Abortion Stigma as a Barrier
Abortion is highly stigmatized in the United States (Baker et al. 2023; Cowan 2014; Watson 2018). Abortion stigma interacts with bureaucratic, legal, and financial restrictions, compounding issues of access. For example, some women who would otherwise have an abortion to end an unwanted pregnancy do not do so because they have internalized stigma against abortion (Cohen and Joffe 2020; Watson 2018). Additionally, abortion stigma means women who want an abortion cannot rely on their usual sources of social support in navigating the “obstacle course” to access abortion (Cohen and Joffe 2020). In many cases, women fear that asking for help will expose them to stigma or that their usual networks will not help them (Cowan 2014; Watson 2018). Women even report instances of friends or family trying to sabotage their attempt to access abortion (e.g., offering financial support that is not provided at the last minute in hopes the woman might be forced to forgo the abortion) (Cohen and Joffe 2010). As more states propose and pass bills that prosecute individuals who assist others in getting an abortion, those seeking abortions will likely become even more reticent to seek information and support from social networks. As such, people seeking abortions must rely on a smaller—and for some, a nonexistent—group of people they feel will help them (Cohen and Joffe 2020; Watson 2018). Moreover, because of rapid changes in legislation and misinformation related to stigma, some people who need or want an abortion do not seek one because they incorrectly believe the procedure is illegal where they live or is dangerous (Cohen and Joffe 20202).
A key mechanism through which administrative burdens operate is increasing the stigma associated with a policy target—such as abortion (Moynihan et al. 2015). For example, TRAP laws are effective at increasing the stigma against abortion. Through such laws, stigma contributes to the creation of large geographic regions where abortion is not available because some abortion providers are unwilling to work or to live in certain places due to concerns about vigilante violence or legal prosecution (Cohen and Joffe 2020). In this article, we investigate the socioeconomic consequences of the restriction and stigmatization of abortion because—despite the policies, stigma, and other factors that limit abortion access—abortion remains a very common medical procedure (Watson 2018).
Previous Research Linking Abortion and Socioeconomic Outcomes
Abortion and Financial Outcomes
There is good evidence that access to abortion is linked to positive financial outcomes for women (Banerjee 2023; Foster 2020, 2022; Foster, Biggs, Ralph, et al. 2018; Zabin, Hirsch, and Emerson 1989). Most notably, Foster’s (2020) groundbreaking “Turnaway” study sampled women seeking abortions from 30 abortion clinics and created a dataset that included women who received abortions just before their state’s gestational age limit, women who obtained abortions in the first trimester, and women who were “turned away” because they just missed gestational age limit. These data were collected between 2008 and 2010 (Foster, Biggs, Ralph, et al. 2018). This innovative design allowed the researchers to compare women who had an abortion to women who wanted an abortion but were turned away by the same abortion clinics. Using these data, Foster, Biggs, Ralph, and colleagues (2018) found that, by self-report, both six months and four years later, the women who were not turned away (i.e., who got the abortion) were less likely to be in poverty, less likely to receive public assistance, and more likely to be employed full-time, compared to those who had been turned away for a wanted abortion.
Using the same “Turnaway” data, Miller and colleagues (2023) found that women who were turned away from clinics and denied wanted abortions experienced an increase in financial distress, as measured by indicators on their later credit scores, compared to women who were not turned away. In addition, Foster, Biggs, Raifman, and colleagues (2018), again using the “Turnaway” data, show that children of mothers who were denied wanted abortions were more likely to be living below the federal poverty level and, in some cases, were more likely to live in households without enough money to pay for basic living expenses, compared to children whose mothers were not denied the wanted abortion. Data from the “Turnaway” study provide compelling evidence that abortion access is linked to positive financial outcomes for women and their children. In fact, almost all our recent knowledge about abortion access and its impact on women’s lives comes from this single dataset.
Outside of work using Foster’s (2020) “Turnaway” data, few studies address the question of how financial outcomes are related to abortion (Bernstein and Jones 2019). The studies that do exist find evidence of a relationship between policies that increase access to abortion at the state level, lower poverty rates, and higher rates of employment, especially for Black women (Bernstein and Jones 2019; Gruber, Levine, and Staiger 1999; Kalist 2004). These studies examined the effects of the liberalization of state-level abortion laws in the 1970s and Roe v. Wade, the 1973 U.S. Supreme Court case that legalized abortion nationwide (Bernstein and Jones 2019). This work examined the population-level outcomes associated with these state- and national-level abortion policy changes—this research did not examine individual-level socioeconomic outcomes.
Abortion and Educational Outcomes
The research linking abortion to educational outcomes is minimal. Angrist and Evans (1999) link state-level abortion reform in the 1970s to higher educational attainment for Black women in states with more access to abortion. Zabin and colleagues (1989), using a sample of 360 Black women in Baltimore, Maryland, seeking pregnancy tests, found that women who had an abortion were much more likely to be in school at the expected grade level for their age—or to have graduated from high school—than women who did not have an abortion. More recently, Foster (2020) and Ralph and colleagues (2019), using the “Turnaway” data, found some evidence that women who were not able to obtain wanted abortions were less likely to earn a postsecondary degree than those who received a wanted abortion. However, baseline differences in the educational attainment and ages of the two groups and limited sample size undermined the researchers’ ability to draw definite conclusions. As a result, the Foster (2020) and Ralph and colleagues (2019) results are inconclusive regarding educational attainment.
Our Project
In two studies, we take innovative methodological approaches to build on extant work and address gaps in the current literature. We do so by using longitudinal data and two different statistical approaches to link (1) being in an environment with abortion restrictions and (2) having an abortion to multiple indicators of financial stability and educational attainment, at the individual level, up to 24 years later. These two studies have several strengths that address gaps in the current literature, including (1) an unprecedented follow-up period; (2) previously unstudied economic indicators; (3) samples derived from a population-based survey; (4) in Study 1, linking macro-level indicators of abortion restrictions to individual-level data; and (5) in Study 2, counterfactual matching techniques—with exact matching on poverty—to generate estimates from a pseudo-randomized experiment. Below, we detail the motivation and logic for each of our study approaches.
Study 1: Restrictive Abortion Environments
Research on the relationship between socioeconomic outcomes and abortion has been limited by methodological problems primarily related to documenting the effects of the highly stigmatized, sometimes illegal, medical procedure. For example, although abortion is common, it remains underreported in survey data (Desai et al. 2021; Lindberg et al. 2020; Maddow-Zimet, Lindberg, and Castle 2021; Tierney 2019). In our first study, we use national probability data to link state- and county-level indicators of abortion restriction to individual-level socioeconomic outcomes for women up to 24 years later, to evade the issue of underreporting. Our measure of “restrictive abortion environments” includes three measures of whether a respondent’s state has policies restricting abortion access and one measure of whether a respondent’s county has an abortion provider.
Roughly one-quarter of women will have an abortion (Jones and Jerman 2022). As such, variation in levels of abortion restrictions may influence roughly one-fourth of our sample, allowing us to make inferences about the role of abortion restrictions using high-quality, independently-coded data on state- and county-level abortion policies and abortion provider proximity on later socioeconomic outcomes. Importantly, we adjust for other state- and community-level variables that may correlate with socioeconomic status and abortion restriction, including state- and census-tract-level income and education, availability of family planning services in a respondent’s county, funding for education at the state and county levels, urbanicity, and percent Democratic voters in a respondent’s county.
Our approach is informed by recent work linking macro-level policy and social context indicators to individual-level health outcomes of respondents to assess the relationship between social contexts and individual-level health outcomes (Everett et al. 2022; Hatzenbuehler 2017; Homan 2019; Taylor 2019). Work using this method provides compelling evidence that macro-level policy and social contexts affect the health and well-being of individuals in affected geographic regions. We use a similar approach, in that we link access to a health procedure (i.e., abortion) at the macro level to individual-level socioeconomic outcomes.
Study 2: Using Abortion to Terminate Pregnancy
In the second study, we restrict our sample to women who report having had a pregnancy before age 20 that ended in either a live birth or an abortion. We use matching statistical techniques, based on a counterfactual statistical approach, which allows us to compare respondents who are similar across sociodemographic characteristics but differ in how their pregnancies ended (i.e., live birth versus abortion) to generate an average treatment effect (ATE). As such, our second study uses a counterfactual framework and matching statistical techniques that answer the question: Under similar circumstances and for people with very similar characteristics, what is the relationship between having an abortion and later socioeconomic status, compared to giving birth?
In this second study, we only assess outcomes for individuals with self-reported pregnancies before age 20. As such, we do not, and cannot, provide prevalence or population-based inferences because the sample excludes many women who had abortions but did not report them. Indeed, underreporting of abortion is a well-documented problem in this area of research (Desai et al. 2021; Lindberg et al. 2020; Tierney 2019). Moreover, our sample is restricted to respondents who reported a teen pregnancy—a sample that differs from the general population of adolescents. Finally, there is likely a selection effect into having an abortion versus a live birth that affects both the treatment effect (i.e., abortion) and our later-life socioeconomic outcomes, primarily related to respondents’ socioeconomic status at adolescence.
Our approach in Study 2 does not eliminate these concerns but does address them in several key ways. First, while our sample excludes persons who had abortions and did not report them, it is unlikely to include false negatives (i.e., individuals who did not have a teen pregnancy or abortion but reported they did on a survey). In this way, our sample is valid (i.e., reflects actual experiences) but not representative (i.e., excludes some unreported abortions). Second, and related to the first point, the results generated from the matching approach (i.e., the ATEs) are not intended to generate a prevalence-based estimate. Instead, the statistical approach is intended to mimic an experimental, randomized trial by comparing the socioeconomic outcomes of individuals, who, once pregnant, have similar probabilities of choosing to have an abortion—and the same poverty status—but where one did, and one did not, have an abortion. The estimate produced is an effect size and need not be generalizable.
Third, once in the study sample, the matching approach limits selection bias into abortion by ensuring the comparisons made between the two groups (abortion versus live birth) are not random. For example, a high-SES teenager who has an abortion will not be compared to a low-SES teenager who has a live birth. Only respondents with very similar backgrounds are compared to each other. As such, the methods and sample allow us to compare the future SES outcomes of respondents with very similar probabilities of having an abortion during adolescence, and the exact same poverty threshold, to each other. This approach thus addresses, while not eliminating, issues regarding selection bias in other probability-sampled data.
Data
We use data from Add Health, a large probability-based sample designed to track the health and well-being of U.S. individuals as they transition to adulthood (Harris et al. 2019). The initial sample was drawn in 1994 from 80 high schools and 52 middle schools throughout the United States, with an unequal probability of selection (Wave I; age 12 to 18). A subsample of students from these schools (n = 20,745) were asked to complete additional in-home interviews and were contacted for follow-up interviews in 2001–02 (Wave III; age 18 to 26), 2007–08 (Wave IV; age 24 to 32), and 2016–18 (Wave V; age 34 to 44). Response rates for these three waves were 77.4, 80.3, and 69.3 percent, respectively. Importantly, this dataset includes complete reproductive history rosters collected at Wave IV, macro-level indicators of abortion restrictions collected at Wave I (age 12 to 18), and multiple indicators of socioeconomic status at Waves IV (age 24 to 34) and V (age 33 to 44). Figure 1 depicts our analytic approaches for Studies 1 and 2.
Study Design
These data have several strengths for assessing our questions of interest. First, our analytic samples are drawn from a national probability-based sample rather than a clinical or convenience sample. Second, our data cover a 24-year period, allowing us to look at both short- and long-term socioeconomic outcomes. Third, this study uses a combination of state-, county-, and individual-level data. As a result, we adjust for several state- and county-level contextual variables while accounting for individual-level characteristics related to abortion and socioeconomic status. Finally, we can examine the relationships between abortion and multiple measures of socioeconomic status, allowing us to create a robust portrait of the role of abortion as related to socioeconomic stratification.
Study 1, Restrictive Abortion Environments: Methods
In Study 1, we use an index of state-level abortion policies and a county-level abortion provider indicator, all measured in 1994, when respondents were aged 12 to 19 years. We assess the relationship of this index to socioeconomic outcomes at ages 24 to 32 (2001–02) and again when respondents were aged 34 to 43 (2016–18).
Sample
For Study 1, we restrict the sample to women who (1) had a valid sample weight; (2) had their household geolocated and linked to abortion environment indicators at Wave I (i.e., when respondents were aged 12 to 18), and (3) participated in Wave IV and/or Wave V of the study. We began with 10,263 girls interviewed at Wave 1 and removed 2,395 participants who did not have valid population weights (n = 7,868). An additional 34 respondents were removed for not having abortion policy indicators at Wave I. Due to a lack of follow-up data, we removed 2,109 respondents in Wave IV and 2,137 in Wave V. Our analytic sample size varies based on response rates to measures of socioeconomic status, from 5,849 to 5,529 in Wave IV and from 5,731 to 4,848 in Wave V.
Socioeconomic Outcomes
Wave IV socioeconomic outcomes
We examined five dimensions of socioeconomic status at Wave IV when respondents were 24 to 32 years old. First, high school graduation is a dichotomous measure of whether respondents indicated they had graduated from high school or completed a GED. Second, college graduation is a dichotomous measure of whether respondents had graduated with a bachelor’s degree. Third, self-reported household income is an ordinal variable that ranges from 1 (less than $5,000) to 12 ($150,000 or more). Fourth, we include a dichotomous variable of whether respondents report having ever worked full-time (1 = yes, 0 = no) for at least 35 hours a week at a paying job while not primarily a student. Respondents were instructed not to report summer work for this variable. Fifth, we created a scaled measure of economic insecurity that includes whether, in the past 12 months, respondents reported there was a time when “you/your household”: (1) “was without phone service because you didn’t have enough money”; (2) “didn’t pay the full amount of the rent or mortgage because you didn’t have enough money”; (3) “were evicted from your house or apartment for not paying the rent or mortgage”; (4) “didn’t pay the full amount of gas, electricity, or oil bill because you didn’t have enough money”; (5) “had the service turned off by the gas or electric company, or the company wouldn’t deliver because payments were not made”; or (6) “worried whether food would run out before you would get money to buy more.” This scale ranges from 0 to 6.
Wave V socioeconomic outcomes
We measured five dimensions of socioeconomic status at Wave V when respondents were aged 34 to 44 years. First, college graduation is a dichotomous measure of whether respondents had graduated with a bachelor’s degree. Second, self-reported income is measured at Wave V using an ordinal scale that ranges from 1 (less than $5,000) to 13 ($200,000) or more. Third, eviction is a dichotomous variable that measures whether, since 2008, the respondent has reported experiencing an eviction (1 = yes, 0 = no). Trouble paying bills measures whether, since 2008, respondents reported falling behind on paying bills (1 = yes, 0 = no). We also included a measure of debt that asked respondents: “Suppose you and others in your household were to sell all of your major possessions (including your home), turn all of your investments and other assets into cash, and pay off all of your debts. Would you have something left over, break even, or be in debt?” This dichotomous variable captures whether respondents would be in debt (1 = yes, 0 = no). Finally, we measured employment using a survey item that asked respondents, “Are you currently working for pay?” (1 = yes, 0 = no).
Measuring Restrictiveness of the Abortion Environment between Ages 12 and 18
Study 1 assesses the relationship between the restrictiveness of a respondent’s abortion environment and later socioeconomic outcomes. Our measure of the restrictiveness of the abortion environment is a scaled indicator of three abortion-related policies and one indicator of geographic access to abortion. All four items were measured at Wave I, when respondents were 12 to 18 years old, and were connected to respondents’ Wave I resident GPS location. To protect respondents’ anonymity, all indicators were coded by Add Health staff—not by the authors of this article.
The first item in the indicator is a measure of public funding for abortion, which captures whether the state of residence allowed Medicaid funds for abortion only in limited circumstances (e.g., rape, incest, or life endangerment) (yes = 1) or in all or most circumstances (yes = 0). The second indicator is a measure of parental consent requirements, which captures whether the state did not require parental or family consent before minors could receive abortions (= 0), had unenforced consent laws (e.g., laws passed but not enacted) (= 1), or had enforced consent laws (= 2). Third, we use a dichotomous indicator of whether a respondent’s home county had an abortion provider (0 = at least one abortion provider; 1 = no provider). The fourth indicator measures state-level mandatory waiting periods and informed consent; this indicates whether the state has no informed consent or mandatory waiting periods (= 0), informed consent (= 1), or informed consent and mandatory waiting periods (= 2). We combined these indicators into a scale ranging from 0 to 6, with higher scores representing more restrictive environments.
Additional Covariates
To better isolate the relationship between restrictive abortion environments and later socioeconomic outcomes, we include additional covariates in our models for respondents’ individual-, family-, and community-level characteristics at Wave I. We include covariates for variables that could plausibly be related to restrictive abortion environments, later socioeconomic status, or both (Banerjee 2023).
Characteristics of respondent
We add covariates for respondent race/ethnicity, which we treat as a factor variable. Non-Hispanic White is the referent category; the other categories are non-Hispanic Black, Hispanic, and other race/ethnicity. We also add a continuous age variable at Wave IV for Study 1 (age 24 to 32) and Wave I for Study 2 (12 to 20 years old).
Characteristics of respondent’s family at Wave I
We include two measures of family socioeconomic status at Wave I. The first measures whether the respondent’s parents graduated from high school (0 = no, 1 = yes), and the second whether, at Wave I, the respondent’s family income was below the federal poverty line (referent), was not below the federal poverty line, or was missing.
Characteristics of respondent’s community at Wave I
The Add Health contextual data files make several state- and county-level variables available to researchers, and we adjust for these variables in our models. We include a dichotomous variable of whether there were no family-planning service providers in a respondent’s county of residence at Wave I (1 = yes, 0 = no). We also use a continuous variable for the proportion of Democratic voters in a respondent’s state of residence at Wave I during the 1992 presidential election—the closest preceding presidential election to Wave I data collection. We include two measures of educational spending: the per capita local direct spending on education at the county and state levels. We include two census tract socioeconomic status variables that measure whether respondents lived in the top quartile of income across census tracts (1 = yes, 0 = no) and the top quartile of college graduates across census tracts (1 = yes, 0 = no). Finally, we adjust for whether a respondent lived in an urban area (1 = yes, 0 = no).
Analytic Approach
We use separate multivariate regression models to assess the relationship between (1) abortion restriction at Wave I and socioeconomic status at Wave IV, and (2) abortion restriction at Wave I and socioeconomic outcomes at Wave V. We use logistic regression for all dichotomous socioeconomic outcome variables (i.e., high school graduate, college graduate, employment, eviction, trouble paying bills, and debt). For income and economic instability, we use ordered logistic regression. Model 1 (Tables 2 and 3) for all outcomes shows the bivariate relationships between abortion restriction and our dependent variables. Model 2 (Tables 2 and 3) includes covariates. All models adjust for the Add Health complex sampling frame, which uses population weighting based on individual-level characteristics; we also adjust for the primary sampling unit (i.e., schools) and region using the “svy” commands in Stata 16.0.
Study 1, Restrictive Abortion Environments: Results
Descriptive Statistics
Table 1 shows the descriptive statistics for Study 1. The average level of abortion restriction was 3.55. At Wave IV, 93.21 percent of the sample had graduated from high school, and 36.10 percent had graduated from college. The average score on the economic instability scale was 0.54, suggesting low levels of instability among our sample at Wave IV (range: 0 to 6). The mean income score was 7.97, corresponding with $40,000 to $49,000 on the income scale. At Wave V, 42.04 percent of women in the sample had graduated from college. The average income score at Wave V was almost 9, corresponding to $50,000 to $75,000 on the income scale. Indicators of economic insecurity at Wave IV (2007–08) were relatively common. Just over half of women reported having had trouble paying bills, and almost one quarter reported that if their assets were liquified, they would be in debt. The average age at Wave IV was 28.62.
Descriptive Statistics for Study 1
Source: National Longitudinal Study of Adolescent Health.
Note: N = 5,848; M = mean.
The majority of the sample was non-Hispanic White (67.85 percent), 15.39 percent were non-Hispanic Black, and 11.96 percent were Hispanic. Among the sample, 20.69 percent had at least one parent who had not graduated from high school, and 13.45 percent lived below the federal poverty line at Wave I. At Wave I, just 2.86 percent of the sample lived in a county that did not have a single provider of family planning services, and the average proportion of Democratic voters in counties where our respondents resided was 0.42. Just under half of the sample lived in an urban setting.
Relationship between Restrictive Abortion Environments at Wave I and Socioeconomic Outcomes at Waves IV and V
Table 2 shows the results of regressing Wave IV socioeconomic outcomes on our measure of living in a restrictive abortion environment at Wave I using ordered logistic and logistic regression. The first set of models in Table 2 (Panel 1) shows the bivariate relationship. The second set of models (Panel 2) includes respondents’ characteristics at Wave 1, respondents’ families’ characteristics at Wave 1, and county- and state-level characteristics of where respondents lived at Wave 1 (see Table 1 for a complete list of covariates).
Ordered Logit and Logistic Regression of Wave IV Socioeconomic Outcomes on Level of Restricted Abortion Environment at Wave I
Source: National Longitudinal Study of Adolescent to Adult Health.
Note: Panel 1 shows bivariate associations. Models in Panel 2 adjust for all individual and additional community-level variables. OR = odds ratio; 95% CI = 95% confidence interval; P = probability value.
We found no relationship between a restrictive abortion environment at Wave I and high school graduation at Wave IV. However, a restrictive abortion environment at Wave I is associated with decreased odds of graduating from college without (Panel 1: OR = 0.91, p = 0.00) and with (Panel B, Model 2: OR = 0.94, p = 0.04) covariates. Restriction at Wave 1 is also associated with lower levels of income at Wave IV, without (Panel 1, Model 1: OR = 0.89, p = 0.00) and with (Panel 2, Model 2: OR = 0.92, p = 0.00) covariates. Restriction at Wave I is associated with increased economic instability at Wave IV (Panel 1: OR = 1.06, p = 0.02). This relationship attenuates with the inclusion of other covariates (Panel 2: OR = 1.04, p = 0.12). We see a marginally significant association between restriction at Wave I and employment in Wave IV in Panel 2 (the adjusted models), suggesting that restricted-access abortion environments may increase the likelihood of employment at Wave IV (OR = 1.09, p < 0.10).
In Figure 2, we plot predicted probabilities derived from adjusted models where the relationship between restrictive abortion environments and outcomes was at least marginally statistically significant (i.e., p < 0.10, Panel 2, Table 2) in the adjusted models. We use the recommended cutoff of 84 percent confidence intervals to account for the fact that overlapping 95 percent confidence intervals often do not demonstrate statistical significance (Payton, Greenstone, and Schenker 2003; Schenker and Gentleman 2001). Because the results for income were derived from ordered logistic regression models, they indicate the probability of moving to the next, higher income category. This figure illustrates that restrictive abortion environments during adolescence are associated with lower income, a lower likelihood of being a college graduate, and a higher likelihood of being employed at ages 24 to 32 (i.e., at Wave IV).
Study 1: Wave IV, Predicted Probabilities from Table 2 Adjusted Models
Table 3 presents the results from our models regressing socioeconomic outcomes at Wave V on restricted abortion environments at Wave 1. These results show that restriction at Wave I is associated with decreased odds of graduating from college at Wave V (Panel 2 OR = 0.95, p = 0.05) and increased odds of reporting having trouble paying bills since 2008 (Panel 2 OR = 1.07, p = 0.01). The results also show that restriction at Wave I is associated with being in debt if the respondent’s assets were liquified (Panel 2: OR = 1.07, p = 0.04) and having lower income (Panel 2: OR 0.92, p = 0.00), in adjusted models. Our results show no statistically significant relationship between restriction at Wave 1 and the likelihood of being employed at Wave V (Panel 2) and marginally significant increased odds of reporting having been evicted since 2008 (Panel 2: OR = 1.06, p = 0.08), again, in the adjusted models.
Ordered Logit and Logistic Regression of Wave V Socioeconomic Outcomes on Level of Restricted Abortion Environment at Wave I
Source: National Longitudinal Study of Adolescent to Adult Health.
Note: Panel 1 shows bivariate associations. Models in Panel 2 adjust for all individual and additional community-level variables. OR = odds ration; 95% CI = 95% confidence interval; P = probability value.
In Figure 3, we plotted predicted probabilities for the five adjusted models that reached marginal statistical significance (i.e., p < 0.10, Table 3, Panel 2) in the adjusted models. Figure 3, similar to Figure 2, uses the 84 percent confidence interval and demonstrates that restrictive abortion environments during adolescence are associated with a lower likelihood of being a college graduate, and an increased likelihood of trouble paying bills, being evicted, and being in debt, as well as a lower income between ages 34 and 44 (i.e., at Wave V).
Study 1: Wave V Predicted Probabilities from Table 3 Adjusted Models
Study 2, Using Abortion to Terminate Pregnancy: Methods
Sample
For Study 2, we restrict our sample to women who (1) have valid sample weights, (2) reported a pregnancy that ended in either a live birth or abortion before age 20 (n = 1,658), and (3) were followed up in Waves IV or V. Due to missing data, our analytic sample varies from 1,635 to 1,561 for Wave IV outcomes and 1,146 to 1,016 for Wave V outcomes.
Socioeconomic Outcomes
We use the same socioeconomic outcomes for Study 2 that we used in Study 1. See the “Socioeconomic Outcomes” section above for a detailed description of those variables.
Using an Abortion to Terminate Pregnancy
In Study 2, we are interested in assessing the relationship between having an abortion as a teenager and later socioeconomic outcomes. In Wave IV of the Add Health survey, respondents completed pregnancy history rosters that included information on when and how each reported pregnancy ended (i.e., abortion, miscarriage, live birth, and mode of delivery). We use Wave IV pregnancy histories because in Waves I and II, the majority of respondents were younger than age 20 (and therefore we did not have their complete adolescent reproductive history at the time), and because data accuracy issues have been raised for Wave III pregnancy rosters (Amato et al. 2008; Schoen, Landale, and Daniels 2007). In particular, the wording in the pregnancy roster survey items led to confusion and resulted in underestimates of pregnancy outcomes (Amato et al. 2008; Schoen et al. 2007).
Using a respondent’s date of birth and the date the reported pregnancy ended, we created a measure that captures whether respondents reported a pregnancy before age 20. We also created a dichotomous variable that captures whether a teen pregnancy ended in abortion (referent category = 1) or a live birth. We excluded respondents who reported a pregnancy that ended in miscarriage because it is unclear whether a reported miscarriage reflects a true miscarriage, a pregnancy “scare,” or a possible abortion that was reported as a “miscarriage” on the survey. No respondents in our sample reported both an abortion and a live birth before the age of 20.
Additional Covariates
To better isolate the relationship between abortion and later socioeconomic outcomes, we include covariates for individual- and family-level characteristics that could be related to the likelihood of reporting a teen pregnancy, having an abortion, or later socioeconomic status. We include respondent age at Wave I and respondent race/ethnicity (non-Hispanic White [referent category], non-Hispanic Black, Hispanic, other race/ethnicity). We include a measure of the respondent’s family structure at Wave I (two-biological-parent household [referent category], single mother, single father, other two-parent households, other family structure) and for whether the teen respondent’s household was below the federal poverty level (FPL). We calculated FPL using the total household income reported by the respondent’s parents at Wave I and household size (below 100 percent FPL [referent = 0]; > 100 percent FPL = 1). We include the respondent’s parents’ level of education measured as an ordinal variable that captures whether the respondent’s parents had less than a high school education (= 0), high school degree (= 1), some college (= 2), or graduated from college (= 3). We also include an indicator for “adolescent parent” if the difference between the age of the respondent at Wave I and the age of the parent filling out the parent survey indicated the parent was younger than age 20 at the time of the respondent’s birth.
Finally, we include depressive symptoms, binge drinking, and respondents’ college expectations. Depressive symptoms at Wave I were measured using the CES-D 10 scale, which ranges from 0 to 22. Binge drinking at Wave I was measured from an item that asks respondents if, in the past 12 months, they had consumed five or more alcoholic beverages in a single day (1 = yes, 0 = no). We measure college expectations from a survey item that asked respondents at Wave I, “On a scale of 1 to 5, where 1 is low, and 5 is high, how much do you want to go to college?”
Analytic Approach
We use matching analyses with a counterfactual framework using propensity score matching, with exact matching on adolescent poverty. In the ideal case, we would want to use a randomized experimental trial to assess the effects of abortion on socioeconomic outcomes. However, it is impossible and unethical to assign respondents to abortion randomly. Therefore, the counterfactual framework and matching techniques are ways to use statistical inference, rather than random assignment, to generate the effect of the “treatment” of abortion. Using this approach, we compare the socioeconomic outcomes of respondents who had similar propensities to experience a specific event—in this case, abortion—based on the covariates listed in Table 4, but had different “treatments” (i.e., abortion) versus the “control” group (i.e., live birth). We used the “Teffects” package in Stata 16.0.
Descriptive Statistics for Study 2 for the Full Sample and by Whether Respondent Had an Abortion
Source: National Longitudinal Study of Adolescent to Adult Health.
Note: m = mean; SE = standard error.
While not eliminating selection bias and underreporting issues, matching approaches attempt to address them directly and limit their effect on the results in several ways. First, selection bias is an important methodological consideration. Table 4 shows important sociodemographic differences between individuals who report having an abortion and those who report a live birth. Many of these differences are related to SES (e.g., respondents who have live births are more likely to live in poverty, have lower-educated parents, and have lower expectations for college). These differences are also related to future SES. This issue is one of the primary reasons we use matching techniques. The benefits of matching approaches lie in the fact that they restrict comparisons between respondents who had abortions and those who gave birth—to respondents with the same probability of reporting an abortion.
In Step 1 of the approach, propensity scores are developed using sociodemographic characteristics that generate a probability score of reporting an abortion. In this step, we entered all the covariates listed in Table 4 into an equation that predicts the probability of having an abortion. In Step 2, average treatment effects (ATEs) are generated—an average difference in our outcome variables between matched pairs, based on the propensity scores.
Because adolescent SES has such a powerful influence on future SES and is related to the use of abortion to end a pregnancy, we use exact matching on adolescent poverty. That is, adolescent poverty does not feed into the propensity score. Rather, abortion–live birth pairs must share the same adolescent poverty status. As such, the estimate produced from these models, the ATE, tells us, on average, the effect of having an abortion on future SES—compared to someone who has a statistically similar probability of having an abortion and has the same adolescent poverty status—but who had a live birth.
Second, abortion underreporting plagues this area of research. The matching approach is a suitable way to address underreporting insofar as the results generated from matching are not intended to generate data on the prevalence of abortion or its effects at the population level. Rather, we are interested in the specific treatment effect of abortion. That is, among individuals who do report pregnancies (and who have similar propensity scores to have an abortion and the same adolescent poverty status), what is the effect of abortion on future SES compared to a live birth (i.e., the counterfactual outcome)?
Our approach means that individuals who had abortions but did not report them are excluded from our sample. It is thus important to consider who these individuals might be and how they would differ from our sample. An analysis of abortion underreporting in Add Health showed no differences in underreporting by race/ethnicity or age at which the abortion happened (Tierney 2019). However, respondents who have internalized stigma around abortion may be less likely to report it on a survey. Such respondents might be highly religious or politically conservative, or may have grown up in religious or conservative households. Additionally, individuals with higher levels of education and income are more likely to have supportive abortion attitudes (McCall and Manza 2011), perhaps leading to less stigma in these households and an increased likelihood of reporting abortion on a survey. Respondents in our survey who report abortion might thus have higher SES than those who had abortions who are not in our sample. Related, see our Supplementary Analyses section for a discussion of models in which we statistically account for respondents’ religiosity in adolescence.
Taken together, these factors may bias who reports having had an abortion in our sample. Again, however, our goal is not to generate population-level inferences, despite the fact that we use population-based data. Rather, we are interested in the effect of abortion compared to the counterfactual of a live birth, matching on predisposing characteristics related to both the likelihood of having an abortion and future SES. While we cannot entirely deal with selection bias in the sample, we are able to account for factors that may influence selection into having an abortion versus a live birth among respondents who report a teen pregnancy (i.e., our sample), resulting in internal study validity.
Study 2, Using Abortion to Terminate a Pregnancy: Results
Descriptive Statistics
Table 4 presents the descriptive statistics for our full Study 2 sample and stratified by whether respondents reported an abortion or live birth. Among respondents in Study 2, 19.57 percent reported that their pregnancy ended in abortion. Among the total sample, 87.26 percent of respondents had graduated from high school, and 12.68 percent had graduated from college by Wave IV. The mean income level at Wave IV was 7.13, which corresponds with $30,000 to $39,999 per year, and 93.54 percent reported being employed. The mean score on the economic instability scale was 0.94. The descriptive results, stratified by how the pregnancy ended, suggest better economic outcomes among women whose pregnancies ended in abortion. For example, 96.57 percent of women who had an abortion as an end to their teen pregnancy graduated high school, compared to 84.99 percent of those who had live births, and 27.81 percent of women who had an abortion had graduated from college at Wave IV, compared to just 8.99 percent of women who had live births.
We also see differences in socioeconomic outcomes at Wave V. Among women who had an abortion, 40.98 percent had completed college, compared to 16.30 percent of those who had a live birth. The mean income was 9.12 for women who had an abortion ($50,000 to $74,999) compared to 7.42 ($30,000 to $39,999). In addition, 70.51 percent of the total sample reported being late paying bills since 2008, 29.15 percent reported an eviction since 2008, and 31.63 percent reported they would be in debt if all their assets were liquified.
The distribution of race/ethnicity, family structure, having a teen parent, depressive symptoms, and binge drinking were similar for women who had an abortion and those who had a live birth. However, we see differences in family socioeconomic status and college expectations. For example, among those who had a live birth, 21.30 percent lived in households below the federal poverty level, compared to 8.09 percent of those who had an abortion.
Using Matching Techniques to Assess the Relationship between Using an Abortion to Terminate Pregnancy at Wave I and Socioeconomic Outcomes at Waves IV and V
Table 5 shows the results from our matching analyses. We show the ATE from propensity score matching results, with exact matching on adolescent poverty. Results for Wave IV outcomes show that a teenage pregnancy ending in abortion (versus a live birth) is associated with an 11 percent probability increase in graduating from high school (ATE = 0.11, p < 0.001) and a 15 percent probability increase in graduating from college (ATE = 0.15, p < 0.001). Having an abortion versus a live birth in adolescence is also associated with over a 1-point increase in the income scale (p < 0.001). This increase equates to approximately $5,000 more in income per year. Furthermore, in our sample, having an abortion versus a live birth is associated with a 0.30-point lower score on the economic instability scale (p < 0.0001). Finally, we do not find a significant association between abortion and employment.
Average Treatment Effects from Propensity Score Models Comparing Socioeconomic Outcomes among Women Who Had Abortions and Those Who Had Live Births during Adolescence
Source: National Longitudinal Study of Adolescent to Adult Health.
Note: Propensity score models adjust for all individual- and family-level characteristics and use exact matching on poverty. ATE = average treatment effect; 95% CI = 95% confidence interval; P = probability level.
Turning to our Wave V outcomes, using abortion to end a pregnancy continues to be associated with an increased probability of graduating from college in Wave V compared to women who had a live birth (ATE = 0.20, p < 0.001). Abortion use is also associated with statistically significant (ATE = 1.32; p < 0.001) higher scores on the income scale. We do not find a statistically significant relationship between abortion use and being late to pay bills or eviction, although the ATE estimates trend in the direction of economic instability. In contrast to the rest of our results, abortion use is associated with a decreased probability of being in debt (ATE = −0.16; p < 0.05). That is, abortion use is associated with less economic stability, rather than more, on this one measure (i.e., probability of being in debt), in our most conservative matching models. As with the Wave IV results above, we do not observe a statistically significant relationship between abortion use and employment at Wave V.
Supplementary Analyses
Studies 1 and 2: Race and Ethnicity
We conducted supplementary analyses examining whether our results varied by race or ethnicity of the respondent because previous work shows that the socioeconomic effects of abortion differ by race and ethnicity (Bernstein and Jones 2019; Ross and Solinger 2017). We did not find consistent results suggesting that the relationship between our abortion measures and later socioeconomic status was moderated by race or ethnicity, in either Study 1 or Study 2.
Study 1: Operationalizing Policy Indicators
We conducted analyses with an alternative scale for our indicator of a restrictive abortion environment. Given that the parental consent policy, informed consent, and mandatory waiting periods ranged from 0 to 2, and public funding and having an abortion provider in the county ranged from 0 to 1, we tested an alternative coding of the scale. We recoded the scale such that parental consent was rescaled to no consent required (= 0), unenforced consent laws (e.g., laws passed but not enacted) (= 0.5), and had enforced consent laws (= 1). The informed consent/waiting periods measure was rescaled to no informed consent or mandatory waiting periods (= 0), informed consent (= 0.5), and informed consent and mandatory waiting periods (= 1). The recoded scale ranged from 0 to 4, with greater scores representing environments with lower levels of abortion access. Our results are almost identical using this different scale specification, with all reported effect sizes being quite close and no changes in the significance level of any of our results.
Study 1: Geographic Mobility of Respondents
We ran our Study 1 analysis, restricting our sample to women who did not move more than 50 miles from where they lived in Wave I, which was 71 percent of the sample. All our results are robust to this specification.
Study 2: Additional Covariates and Matching Specifications
We ran a series of matching models that also included respondent religiosity at Wave I, because religiosity may affect the likelihood of obtaining an abortion to end a pregnancy. Religiosity did not affect our estimates, so we did not include it as a measure in our main models.
We also conducted a series of matching results using different specifications, including removing the exact matching on poverty, different specifications for nearest neighbor matching, and assessing the average treatment effect on the treated (ATET) (results available upon request). The exact estimates varied slightly across these different specifications, but they consistently told the same story: compared to teenagers who had live births, teenagers who had abortions had significantly better SES outcomes in young adulthood and midlife.
Conclusions
Abortion remains one of the most polarizing topics in U.S. life, despite its ubiquity (Watson 2018). In 1973, in Roe v. Wade, the U.S. Supreme Court affirmed the constitutional right to abortion (Guttmacher Institute 2022). Since that time, there has been strong and organized opposition to access to abortion that, for many people, has made abortion extremely difficult, or impossible, to access—even, in many cases, where it remains legal (Cohen and Joffe 2020; Luker 1985). The repeal of Roe v. Wade, in Dobbs v. Jackson Women’s Health Organization, in June 2022 represented a turning point in the history of abortion in the United States: it eliminated the constitutional right to abortion, making abortion even more difficult to access (Center for Reproductive Rights n.d.).
Since the repeal of Roe, many states have made abortion illegal or effectively inaccessible (Center for Reproductive Rights n.d.). Other states have passed ballot initiatives to codify access to abortion in their state (Society of Family Planning 2023). States that had trigger laws in place to render abortion illegal once Roe v. Wade was overturned have been caught up in lawsuits and legal stays that have rendered these policies unenforceable, at least temporarily (Society of Family Planning 2023). The full economic implications of abortion policy changes since the repeal of Roe will likely not be fully observable for some time. However, our findings yield important insights into the implications of increased abortion restrictions for women’s socioeconomic outcomes.
Using two different approaches, one that focuses on restrictive abortion environments and a second that focuses on having an abortion, we found that both living in a relatively less abortion-restricted environment and having an abortion during adolescence are associated with better socioeconomic outcomes for women up to 24 years later. We found that women who lived, as teenagers, in locations where abortion was less restricted were more likely, when they were 34 to 43 years old, to have graduated from college and to score lower on multiple indicators of poverty and economic insecurity than those living in states and counties with higher abortion restriction. This means that women with difficulty accessing abortion are more likely to experience trouble paying their bills, be in debt, and face eviction from their homes.
We also found that—among women who self-reported pregnancies under age 20—those who reported the pregnancy ended in abortion were more likely to have graduated from college and have higher incomes, and were less likely to experience economic challenges, including having their utilities (e.g., phone, electricity, and gas) turned off, eviction, and food insecurity, than were those who reported the pregnancy resulted in a live birth. Finally, we found that the relationships between our measures of abortion—in Studies 1 and 2—were less consistently associated with employment outcomes compared to other measures of economic security, implying that women who do not have access to abortion may not be less likely to be employed, but are perhaps more likely to be employed in low-wage jobs. Our results, combined with previous research, make several important contributions to sociological literature by documenting how abortion is a mechanism of socioeconomic stratification in the United States
Contributions to the Literature
Similar to our Study 1, previous research has attempted to address self-report issues endemic to research on abortion by assessing the effects of national and state abortion policies on socioeconomic outcomes for women living in those locations. One limitation of previous research taking this approach is that it assesses outcomes at the aggregate level because these researchers did not have information on the outcomes of individuals living in a given abortion policy climate (Bernstein and Jones 2019). Nevertheless, examining state-level, aggregated outcomes in this way is critical and provides compelling evidence documenting the adverse socioeconomic outcomes associated with denying people access to abortion (Bernstein and Jones 2019). Our results from Study 1 build on this important research by using unique, probability-based, national data that contain abortion restrictions at the state and county levels and individual respondent outcomes at two times points over more than 20 years. These innovations allow us to adjust for respondents’ individual- and family-level characteristics, and to assess individual-level outcomes, at multiple times, over an extended period. This approach, paired with previous work, provides strong evidence that abortion restrictions may shape later socioeconomic outcomes—even accounting for respondents’ baseline personal and family characteristics.
Our results from Study 2 make another important methodological contribution by using a counterfactual framework—using matching statistical techniques and longitudinal data—to assess the socioeconomic effects of abortion use. In Study 2, we restricted our sample to women who reported having had a pregnancy before age 20 that ended in a live birth or abortion. Using a counterfactual approach allows us to simulate experimental data that approximate, under similar conditions, the effect of impossible and unethical experimental “treatments,” in this case abortion versus teenage parenthood, on long-term socioeconomic outcomes. This approach has previously been used on studies about men and their sexual partner’s abortions (Everett et al. 2019), but to our knowledge, no other studies on the economic effects of abortion have used this approach among women.
Another strength of the Study 2 sample is that, although it is not representative, it is drawn from a national probability sample rather than a clinical sample. Study 2’s probability sample represents multiple U.S. state and local contexts and may include adolescents who self-managed abortions, as well as a broad comparison of who became teen parents. Combining the findings from Foster’s (2020) quasi-experimental clinic data (Foster, Biggs, Ralph, et al. 2018; Miller et al. 2023) with our counterfactual study with a larger sample that is more diverse on gestational ages and parenthood intentions provides strong evidence that abortion use, and access, are related to later socioeconomic outcomes. Without a true experiment—involving unethical and impossible random assignment to abortion, pregnancy, and live birth—we cannot make the claim that the relationship between abortion use, access, and later socioeconomic outcomes is causal; nevertheless, when combined with Foster’s (2020) “Turnaway” findings, our findings from Studies 1 and 2 suggest this relationship is likely to be causal.
Limitations
Despite these contributions, there are limitations to our approach. One limitation, in both Studies 1 and 2, is that our analyses focus on abortion during adolescence, yet people have abortions throughout the life course. Our data are not well-suited to assess whether abortion later in life is associated with even later socioeconomic outcomes. By focusing on adolescent experiences with abortion, we leverage the strengths of our data to (1) time-order the exposure variable (i.e., abortion) on our socioeconomic outcomes and (2) take advantage of the full time-horizon of adulthood to document whether and when early abortion is associated with socioeconomic outcomes. Future work could assess whether abortion later in life is related to even later socioeconomic outcomes, including age at retirement and social security benefits. Based on our findings, and other research, we expect these effects to persist (Foster 2020, 2022; Foster, Biggs, Raifman, et al. 2018; Foster, Biggs, Ralph, et al. 2018).
A second, and related, limitation of our data is that the respondents whose trajectories we follow were all adolescents in the 1990s. As such, they were contending with sexual activity, pregnancy, and abortion in a very different policy and cultural context than today. For this reason, there are limitations to our ability to extrapolate our findings. That said, the effects we documented are likely even more widespread now, as access to abortion has become much more restricted since the 1990s.
In general, this limitation of our data is a limitation of any longitudinal dataset. We cannot, right now, measure the effects of the current policy climate on later economic outcomes. In some sense, this limitation emerges from what is actually a strength of our data—its longitudinal nature. By definition, we can never know what will happen in the future based on current policy. Other cross-sectional study designs could match current policies to women’s current experiences, but they would then not meet the first requirement of causality—that is, the dependent variable occurs subsequently to the independent variable.
A third limitation is that our sample is restricted to women who did not already have a child. Given that previous research has shown the negative effects of denied abortions on existing children, more research is needed to examine the socioeconomic effects of abortion on the children of women seeking abortions (Foster 2020; Foster, Biggs, Raifman, et al. 2018; Foster, Biggs, Ralph, et al. 2018).
A fourth limitation is that abortion research is affected by underreporting because abortion is highly stigmatized and, in many places, illegal (Tierney 2019; Watson 2018). Our data are no different in this respect. One study suggests that in Add Health, our dataset, only 35 percent of expected abortions were reported (Tierney 2019). Our study uses unique features of the Add Heath data to address this limitation in two ways.
First, in Study 1, we used measures of abortion restriction at the state and county levels as a proxy for having an abortion at the individual level. Many recent studies investigating topics other than abortion have used this approach and found robust effects of state- and county-level cultural and policy climates on individual-level outcomes (Budge 2023; Everett et al. 2022; Hatzenbuehler 2017; Homan 2019; Taylor 2020). The robust findings from prior research using this approach give us confidence that such macro-level measures can help us skirt self-report issues in our work. Moreover, abortion is common, especially in the case of adolescent pregnancies, so our measures of abortion restriction likely affect enough people to give us robust evidence regarding the effects of abortion on socioeconomic outcomes (Kost, Maddow-Zimet, and Arpaia 2017; U.S. Department of Health & Human Services n.d.).
Our second study does use self-reports of having an abortion and, as such, the number of women who report abortions is an underestimate. However, we are not attempting to make population-level inferences from these results. Instead, we are leveraging other strengths of these data. For one, we can compare the outcomes between women who report pregnancies as teens and compare two counterfactual conditions: a live birth and a reported abortion. We leverage the fact that we know who did have an abortion—even if that number is an undercount—and we can compare the later socioeconomic outcomes of this group to another group that is otherwise very similar to those who had an abortion. In summary, despite the likely undercount of abortion in Study 2, our approach still provides new evidence linking having an abortion to later socioeconomic outcomes—especially in combination with Study 1, which does not rely on self-report of abortion.
Another limitation of Study 1, in particular, is that we could not include some important measures of access to abortion restriction, such as distance to an abortion provider and other TRAP laws (e.g., hospital admitting privileges, fetal pain bills, and gestational age limits). However, such measures are strongly correlated with the measures included in our abortion restriction measure, and their inclusion would likely make our results more robust. Other contextual variables may be mechanisms that link abortion restriction and socioeconomic outcomes. We adjust for several of these possibilities, including educational spending, the proportion of Democratic voters, and the economic makeup of respondents’ census tracts, however, this is not a definitive list of all potential confounders. The matching analyses in Study 2 address some of these limitations, in that they compare very similar pregnant respondents to each other. As such, by pairing these two studies, we address some of the limitations of each approach and provide robust evidence linking abortion restriction and having an abortion to later socioeconomic outcomes.
Implications
Overall, our approach builds on previous work that emphasizes how having an abortion and abortion restrictions are embedded in social structures and have social consequences, rather than conceptualizing abortion as an individual-level decision with individual-level repercussions (Banerjee 2023; Beisel and Kay 2004; Homan 2019; Ross and Solinger 2017). Our sociological approach to abortion suggests several implications.
For one, our findings relate to inequalities in women’s access to the public sphere. As the Supreme Court decision in Planned Parenthood v. Casey notes: “The ability of women to participate equally in the economic and social life of the Nation has been facilitated by their ability to control their reproductive lives” (Robbins, Goodman, and Klein 2023). In recent U.S. history, access to abortion, and more broadly access to contraception, fundamentally changed economic and educational opportunities for women (Ananat and Hungerman 2012; Bailey 2006; Myers 2012). For example, access to control over reproduction resulted in later ages of first marriage, higher levels of degree completion, and later ages of first birth for women (Goldin and Katz 2002). These demographic changes reflected women’s new positions, opportunities, and expectations in the economy and reduction of gender inequality (Goldin 2006; Goldin and Katz 2000, 2002). However, even against this backdrop of gains in gender equality, policies in the United States over the past several decades have created more barriers to abortion access, culminating with the overturning of the right to abortion at the federal level in 2022 with Dobbs v. Jackson Women’s Health Organization (Cohen and Joffe 2020). Our analyses show that with these increasing barriers to abortion, gains in women’s access to the public sphere may be curtailed as women’s access to education and financial stability decline.
In a related implication, our findings are important because denying women access to education and financial stability may further exacerbate the gender pay gap and the feminization of poverty. While we found associations between abortion and lower levels of education and financial stability, we found a less robust relationship between abortion and later employment. This combination of results provides suggestive evidence that denying women access to abortion may push them into poverty and the labor force, presumably at low wages or in poor working conditions. Future research could establish whether and how denying women access to abortion might force them to enter the labor force for subsistence wages and make them vulnerable to exploitation in the workplace. In summary, abortion restrictions may curtail some women’s access to the public sphere while pushing other women into the public sphere—but, in either case, not on their own terms.
Another implication of our findings is that the effects of eviction, poverty, food insecurity, and utilities being turned off are likely to affect families (Everett et al. 2019; Everett, Sanders, and Higgins 2023; Foster 2020; Foster, Biggs, Raifman, et al. 2018). Limiting access to abortion can transform family dynamics by increasing family size and increasing economic demands associated with more children (Foster, Biggs, Raifman, et al. 2018). Such transformed family dynamics can lead women to exit, or enter, the labor force, increase their reliance on male romantic partners for financial stability, and push women and their families into poverty (Foster 2020; Foster, Biggs, Ralph, et al. 2018; Miller et al. 2023; Roberts et al. 2014). One study, using the “Turnaway” study data, found that children born when their mothers wanted an abortion were more likely to live below the federal poverty level and more likely to live in a household without enough money to pay for basic living expenses, in comparison to children whose mothers obtained a wanted abortion and then had a child within five years (Foster, Biggs, Raifman, et al. 2018). Another study, using the same data, found that women who were denied wanted abortions were nearly four times more likely subsequently to be below the federal poverty line, along with one or more of their children, and were also more likely to live alone, with one or more of their children, five years after the birth of the child (i.e., without the support of a partner, extended family, or roommates) (Foster, Biggs, Ralph, et al. 2018). Children in such families, headed by single mothers without adequate financial and social resources, are disadvantaged in multiple ways (Brown 2004; Jackson et al. 2000; McLanahan and Sandefur 2009). Another study found that men, too, can be “abortion beneficiaries,” with men who, in adolescence, had a partner whose pregnancy ended in abortion having a higher probability of attaining higher education than comparable men whose partner’s pregnancy ended in a live birth (Everett et al. 2019). In all, the socioeconomic outcomes we documented here are likely to spill over to socioeconomic outcomes of families and sexual partners—with sometimes dire consequences for families in which women are denied access to abortion, including eviction, debt, food insecurity, and long-term poverty.
Moreover, the effects of policies that limit or prohibit access to abortion differ by race and class, as well as other social categories (Rader et al. 2022; Robbins et al. 2023; Ross and Solinger 2017). This is because such social categories stratify access to the power and resources needed to get an abortion, and especially so in the many places in the United States where abortion is prohibited or very difficult to obtain (Rader et al. 2022; Robbins et al. 2023; Ross and Solinger 2017). This inequality in access disproportionality harms women of color, women with disabilities, poor women, queer women, non-binary people, transmen, and people at the intersection of these categories (Rader et al. 2022; Robbins et al. 2023; Society of Family Planning 2023). For example, states where abortion access is most curtailed are also the states with some of the highest racial inequities in socioeconomic status and maternal and infant health (Society of Family Planning 2023). Women of color and women in poverty living in those states will disproportionately feel the impact of abortion restrictions.
Conclusion
Our findings add valuable insights to the literature on the socioeconomic effects of abortion restriction and of having an abortion. We find that women under age 20 who faced fewer abortion restrictions or had an abortion are more likely to have graduated from college, have higher income, and score lower on multiple indicators of financial instability—at two time points—over an almost 25-year period, than comparable women who experienced a more restrictive abortion environment or had a pregnancy end in a live birth. Our results provide evidence that abortion is likely an important axis along which women’s social and economic lives are shaped.
We addressed some of the issues of self-report endemic to the study of abortion by (1) using national probability data, collected over multiple waves, for decades; (2) assessing individual respondents’ socioeconomic outcomes, based on the level of abortion restriction in their geographic area; and (3) applying counterfactual matching techniques. Together with previous research on the topic, our results provide strong evidence that abortion access is linked to better socioeconomic outcomes for women, and their families, across the life course. Furthermore, our project is innovative and should be of general interest to sociologists: our results suggest that abortion restriction and having an abortion are likely axes of socioeconomic stratification in the United States. Future work should do more to highlight the central role of access to abortion and other types of birth control—as well as control over reproductive capacity, more generally—to better understand inequalities by race, class, and gender.
Footnotes
Acknowledgements
We thank Zoë Bergman, Daniela Cano, and Aoife Hernon for their capable research assistance.
Authors’ Note
Both authors contributed equally to the work.
Funding
Research reported in this publication was supported by the Eunice Kennedy Shriver National Institute of Child Health & Human Development of the National Institutes of Health under Award Number R01HD091405 and by the University of Colorado Population Center (grant R24 HD066613) through administrative and computing support. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Data Note
This research uses data from Add Health, funded by grant P01 HD31921 (Harris) from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), with cooperative funding from 23 other federal agencies and foundations. Add Health is currently directed by Robert A. Hummer and funded by the National Institute on Aging cooperative agreements U01 AG071448 (Hummer) and U01AG071450 (Aiello and Hummer) at the University of North Carolina at Chapel Hill. Add Health was designed by J. Richard Udry, Peter S. Bearman, and Kathleen Mullan Harris at the University of North Carolina at Chapel Hill.