Abortion Rate Increased And Birth Rate Decreased After Introduction Of Medicaid Abortion Coverage In Illinois

Abstract

The Hyde Amendment prohibits federal spending on abortion, including federal Medicaid dollars. Seventeen states cover abortion care in their Medicaid programs, using state funds, but causal evidence on how Medicaid coverage for abortion affects pregnancy outcomes is limited. Using a difference-in-differences design and pre-Dobbs 2014–21 birth and abortion data from the Centers for Disease Control and Prevention, we evaluated a 2018 Illinois policy introducing Medicaid coverage for abortion. Medicaid abortion coverage increased the number of abortions in Illinois by 2.43 per 1,000 reproductive-aged females, an 18.2 percent increase, and reduced births by 1.66 per 1,000 females, a 2.8 percent decrease, relative to twenty-nine comparison states that did not cover abortion during the study period. Subgroup analyses of birth rates suggested that decreases in birth rates were more pronounced among Black and Hispanic patients, residents in counties with higher poverty rates, and residents closer to an abortion facility. Our finding suggests that Medicaid can play an important role in abortion access.

Relative to people who are denied abortions, people who receive abortion care have fewer life-threatening medical complications,¹ fewer limitations in physical activity,¹ less financial distress,² and more aspirational life plans.³ There is no credible evidence that abortion has negative lasting impacts on the people who receive them.^4,5 Despite this evidence, abortion in the United States is increasingly difficult to access, with total bans in fourteen states and highly restrictive environments in seven additional states.^6,7

People with low incomes are more likely to have an unwanted pregnancy and to experience cost-related barriers to care.^8,9 Financial barriers to abortion care are, in part, a direct consequence of the federal Hyde Amendment, which prohibits federal spending on abortion.¹⁰ Although seventeen states voluntarily use state-only funding to provide abortion under Medicaid without federal matching,¹¹ 7.7 million people—more than half of reproductive-aged females with Medicaid—live in states that do not cover abortion.¹² The Hyde Amendment imposes disproportionate burdens on those living on low incomes and Black and Hispanic people, who are disproportionately served by Medicaid.¹³

In many settings, research demonstrates that Medicaid reduces financial barriers to care.¹⁴ Similarly, paying for abortion out-of-pocket can be cost-prohibitive. Medicaid coverage for abortion care may lead to changes in utilization and subsequent birth rates for Medicaid enrollees. It may also have spillover effects on populations without Medicaid coverage if it increases the number of providers offering abortion care or if it reduces abortion stigma.

Evidence on the effects of Medicaid abortion coverage is mixed. Most of this research uses data from before the 1992 Planned Parenthood vs. Casey Supreme Court case, which led to a dramatic increase in state-level abortion restrictions,^15–20 and recent research is largely cross-sectional.^21–26 As a consequence, there is considerable uncertainty about the causal effects of Medicaid abortion coverage in the post-Casey environment. Qualitative studies suggest that patients’ experiences with abortion are substantially affected by Medicaid coverage,^27,28 underscoring a need to quantify the effects. To address these gaps, we use quasi-experimental methods and evaluate the effects of introducing Medicaid abortion coverage in Illinois on abortion and birth rates of the state’s residents.²⁹

Study Data And Methods

Study Design

Illinois began covering abortion under Medicaid on January 1, 2018.²⁹ We used a difference-in-differences design that compared abortion and birth rates in Illinois with a set of comparison states in the four years before (2014–17) and after (2018–21) the Medicaid coverage change. We started the study period in 2014 so that all years are after coverage expansions under the Affordable Care Act. The study period ended one year before the Dobbs decision granted states the ability to ban abortion.

Measures And Data

We defined the abortion rate as the number of abortions per 1,000 females ages 15–44. Data on abortions came from the 2014–21 Centers for Disease Control and Prevention (CDC) Abortion Surveillance System, which captures annual abortion counts by state of residence, meaning that Illinois residents’ out-of-state abortions, which made up 1.1–1.5 percent of the total, were included.³⁰ State-by-year-level (referred to further on as “state-year”) population denominators were obtained from the Census Bureau. The birth rate was defined as the number of resident births per 1,000 reproductive-aged females. Birth counts were obtained from restricted use 2014–21 natality data from the National Center for Health Statistics, which captures more than 99 percent of births in the US.³¹ As with abortions, births were aggregated at the state-year level to obtain birth rates.

The introduction of abortion coverage in the Illinois Medicaid program in 2018 was the exposure of interest. Information on Medicaid abortion policies for 2014–21 was obtained from the Guttmacher Institute’s State Policies in Brief and supplemented by state legislative bills, judicial rulings, and legal reports from Reproductive Freedom for All and the Center for Reproductive Rights. Detailed citations for our policy sources can be found in the online appendix (section B).³²

The relevant period of exposure for each person spans from conception to the day of abortion or the last day of abortion availability, which we operationalized differently in the abortion and natality data. For abortions, we assumed that conception years align with calendar years in the abortion data, as most abortions occur early in pregnancy³³ and the abortion count data lacks conception dates. For births, we calculated approximate conception dates by assuming that each birth occurred on the fifteenth of the month (only birth month and year are available) and subtracting the obstetric estimate of gestation at delivery observed in the data.³⁴ If the obstetric estimated gestational age was not stated, it was imputed using the national average gestational age of other births in the same year (fewer than 0.1 percent of births; appendix exhibit A4).³²

Our analyses included twenty-nine comparison states that did not provide Medicaid coverage during the study period (Alabama, Arkansas, Colorado, Delaware, Georgia, Idaho, Indiana, Iowa, Kansas, Kentucky, Louisiana, Michigan, Mississippi, Missouri, Nebraska, Nevada, North Carolina, North Dakota, Ohio, Oklahoma, Pennsylvania, Rhode Island, South Carolina, South Dakota, Texas, Utah, Virginia, Wisconsin, Wyoming). To be consistent across analyses, states missing abortion data (California, Florida, Maryland, New Hampshire, New Jersey, Tennessee, Washington D.C.) (appendix exhibit A3)³² and states with inconsistent subgroup categories in birth data (Connecticut, New Jersey; described below) were excluded both in abortion and birth analyses. Maine and West Virginia were excluded because they changed Medicaid abortion coverage during the study period (in 2021 and 2019, respectively).^35,36 In addition, we constructed multiple alternative control groups; these are described below.

Consistent with Andersen’s Behavioral Model of health care use, we account for several social, structural, policy, and health care factors that have been found to influence unintended pregnancy and abortion.^{15,18,25,37–39} We derived state-year covariates from a variety of sources. Average sociodemographic characteristics of all reproductive-aged females in each state came from the American Community Survey (age, race and ethnicity, education, health insurance, median family income, percentage married, percentage noncitizen, percentage unemployed, percentage population that has income less than 100 percent of the federal poverty level). Provider availability variables were derived from the Health Resources and Services Administration Area Health Resources File (number of federally qualified health centers per capita, number of obstetrics-gynecology physicians per capita). We adjusted for Medicaid eligibility thresholds for parents, childless adults, and children (6–18 years old), using the Trends in Medicaid Income Eligibility Limits database from the Kaiser Family Foundation and the number of other abortion restrictions (trigger laws, parental involvement laws, mandated waiting times, in-person counseling mandates that necessitate two trips, inaccurate or misleading counseling, ultrasound mandate, private insurance coverage restrictions, and pre-Dobbs unconstitutional previability restrictions), using State Policies in Brief documentation from the Guttmacher Institute. Last, to account for the fact that different state experiences of the pandemic may have affected access to and use of abortion services, we included the average COVID-19 case rate per 100,000 population in 2020 and 2021 from the Opportunity Insights Economic Tracker. This variable was coded to 0 for prepandemic years. Detailed citations for these data sources can be found in the appendix (section C).³²

Statistical Analysis

We used linear difference-in-differences models, adjusting for the covariates, with state- and year-fixed effects and robust standard errors. The models included an interaction term between an indicator for residence in the treated state (Illinois) and an indicator for posttreatment years. The coefficients on the interaction term measured the average difference in outcomes between Illinois versus control states, before versus after introducing Medicaid coverage in Illinois. This approach allowed us to estimate changes in outcomes attributable to the introduction of Medicaid abortion coverage accounting for secular trends unrelated to the policy.

A difference-in-differences framework assumes that changes in the comparison group represented changes that would have occurred in the treatment group, had the policy not been adopted. To determine the plausibility of this assumption, we visually assessed whether trends in the outcomes were diverging in the treatment and control states before the coverage change, tested a preperiod regression of the outcomes on the interaction of year and the treatment state indicator, and examined event study models, which allow treatment-control differences to vary over each year, not just pre- versus posttreatment.

To assess heterogeneity in the effect of the policy change across sociodemographic groups, we conducted subgroup analyses on birth rates by maternal age (15–19, 20–24, 25–34, and 35+), race and ethnicity (Hispanic, non-Hispanic Asian, non-Hispanic Black, non-Hispanic other races, and non-Hispanic White), the resident county’s poverty rate (divided into within-state tertiles), and distance to the nearest abortion facility (divided into within-state tertiles). Data on county poverty rates in 2017 were obtained from the Department of Agriculture Economic Research Service. County-level distance to the nearest abortion facility as of December 2017 was obtained from the Myers Abortion Facility Database. Births with missing race and ethnicity were excluded from the subgroup analysis by race and ethnicity (appendix exhibit A5).³² Connecticut and New Jersey were excluded from the analysis because they had not adopted the 2003 revised birth certificate by the start of the study period, and thus had inconsistent measures of race and ethnicity. Abortion data do not include subgroup or county-level information, and thus we conducted subgroup analyses only on birth rates.

Sensitivity Analysis

We conducted several sensitivity analyses. First, it is typical in difference-in-differences studies to cluster the standard errors at the level of the policy to account for serial correlation. However, standard approaches to clustering fail when there is a single treated cluster, which motivated our decision to not cluster. We conducted a bootstrap analysis to examine how robust our preferred statistical inference strategy was.⁴⁰ Second, we examined unadjusted models to determine whether the results were consistent.⁴¹ Third, we repeated analyses across five groups of alternative comparison states: all available states (forty-one states for abortion and forty-eight states for birth); all states common to both outcomes (forty states); states that expanded Medicaid in 2014, such as Illinois (nineteen states); states with similar abortion regulations (nineteen states); and states with no other reproductive policy shock (changes to private coverage for abortion, Title X funding, or large contraceptive access initiatives; twenty-six states).

Limitations

This study had limitations. First, because of nonparticipation and inconsistent reporting requirements across states, abortion counts from the CDC might be underestimated. Specifically, medication abortion and abortions performed in physician’s offices (as opposed to hospitals and facilities) may be underreported.^42,43 However, if underreporting is consistent within states across years, it would not meaningfully change our findings. Birth rates are free from this kind of error.

Second, our findings were based on one treated state and might not be generalizable to other states. Illinois is a populous and racially diverse state and is generally supportive of abortion. The general abortion environment, as well as state-level socioeconomic characteristics, might influence the impact of Medicaid abortion coverage. In addition, having one treated state is an important source of uncertainty in our study’s statistical inference. Although we provide results using alternative strategies, our setting is one in which it is challenging to measure statistical significance.

Finally, it is possible that the implementation of abortion coverage in Illinois was correlated with other factors that also affected the outcomes. For example, we are unable to control for changes in the availability of alternative resources (that is, abortion funds) and unmeasured heterogeneity in the pandemic’s impact. To help mitigate some of these biases, we selected the study period (2014–21) to limit the possibility of confounding from the Affordable Care Act, Dobbs, or repeal of Illinois’ parental involvement law in 2022, and we controlled for COVID-19 case rates. We also conducted subgroup analyses to investigate whether effects were largest in areas that would be consistent with a true effect—communities with high poverty rates and those nearer to abortion facilities.

Study Results

Descriptive Statistics

Average population and state characteristics of Illinois versus control states are presented in appendix exhibit A1.³² Overall, Illinois was generally similar to the control states. Illinois had higher proportions of non-Hispanic Black and Hispanic populations and people with Bachelor’s degrees or higher education. Illinois also had fewer abortion restrictions than the typical control state. State fixed effects in our regressions control for any observed or unobserved imbalance across groups that is consistent over time.

Unadjusted trends in abortion and birth rates in Illinois and the control states are reported in appendix exhibit A2.³² The figures do not suggest that pre period trends in Illinois diverged from trends in the control states, supporting the assumption of the design. Our formal statistical test of this assumption is presented in the appendix (exhibit A6).³²

Difference-In-Differences And Event Study Estimates

The difference-in-differences estimates are reported in exhibit 1. Introducing Medicaid abortion coverage in Illinois significantly increased the abortion rate by 2.43 per 1,000 females ages 15–44 (95% confidence interval [CI]: 1.31, 3.55) and decreased the birth rate by 1.66 per 1,000 (95% CI: −3.18, −0.15).

Exhibit 1:

Difference-in-differences and event study regression results, Illinois versus control states, 2014–21

	Unadjusted rates per 1,000 females ages 15–44				Adjusted DID estimates	95% CI
	Prepolicya (2014–17)		Postpolicya (2018–21)
	Illinois	Control states	Illinois	Control states
Abortion	13.3	8.5	15.3	8.6	2.43***	1.31 to 3.55
Birth	59.6	64.5	54.3	60.1	−1.66**	−3.18 to −0.15

SOURCE Centers for Disease Control and Prevention Abortion Surveillance and Vital Statistics, 2014–21. NOTES The analyses included 29 comparison states that did not provide Medicaid coverage during the study period (Alabama, Arkansas, Colorado, Delaware, Georgia, Idaho, Indiana, Iowa, Kansas, Kentucky, Louisiana, Michigan, Mississippi, Missouri, Nebraska, Nevada, North Carolina, North Dakota, Ohio, Oklahoma, Pennsylvania, Rhode Island, South Carolina, South Dakota, Texas, Utah, Virginia, Wisconsin, Wyoming). To be consistent across analyses, states missing abortion data (California, Florida, Maryland, New Hampshire, New Jersey, Tennessee, Washington, D.C.) (see online appendix exhibit A3, see note 32 in text) and states with inconsistent subgroup categories in birth data (Connecticut, New Jersey; described below) were excluded both in abortion and birth analyses. Maine and West Virginia were excluded because they changed Medicaid abortion coverage during the study period (2021 and 2019, respectively). Models were controlled for state-year-level age; race and ethnicity; education; health insurance; median family income; percent married; percent noncitizen; percent unemployed; percent under 100 percent of the federal poverty level of females ages 15–44; number of other abortion restrictions; number of federally qualified health centers per 1,000 females ages 15–44; number of obstetrics-gynecology providers per 1,000 females; Medicaid eligibility criteria for parents, childless adults, and children; and COVID-19 cases per 100,000 people. Difference-in-differences (DID) models included state and year fixed effects. Confidence intervals (CIs) are based on robust standard errors.

^aIllinois started to provide Medicaid coverage for abortions January 1, 2018.

^*p < 0.1

^**p < 0.05

^***p < 0.01

The event study results showed that abortion rates increased and birth rates decreased in all post period years (exhibit 2). The pre period coefficients do not suggest that there were diverging trends in either abortion rates or birth rates between Illinois and the control states.

Exhibit 2:

Difference-in-differences and event study regression results, Illinois versus control states, 2014–21

SOURCE Centers for Disease Control and Prevention Abortion Surveillance and Vital Statistics, 2014–21. NOTES The figure presents results of the authors’ event study analysis (described in the text). A description and list of the 29 states in the control group and 11 states that were excluded are in the text and the exhibit 1 notes. Models were controlled as described in the exhibit 1 notes. Difference-in-differences models included state and year fixed effects. 95% confidence intervals (indicated by whiskers) were based on robust standard errors.

Subgroup Analyses on Birth Rates

Birth rates for 15–19- and 25–34-year-olds decreased by 1.58 per 1,000 (95% CI: −3.14, −0.03) and 2.41 (95% CI: −4.37, −0.45), respectively (exhibit 3). The decrease among 20–24-year-olds was not statistically significant. Birth rates decreased by 4.48 (95% CI: −8.27, −0.70) among Hispanic and 2.93 among non-Hispanic Black (95% CI: −5.78, −0.07) people. The confidence interval was wide for the non-Hispanic Asian subgroup. Counties with higher poverty rates (−3.22; 95% CI: −5.63, −1.00) and closer abortion facilities (−2.11; 95% CI: −3.75, −0.47) had significant birth rate reductions (exhibit 3), whereas less-poor counties and counties with less geographic access did not.

Exhibit 3:

Changes in birth rates by subgroup in Illinois relative to control states before and after the introduction of Medicaid coverage for abortion, 2014–21

SOURCES Centers for Disease Control and Prevention Abortion Surveillance System and vital statistics data, 2014–21. NOTES The figure presents results of the authors’ difference-in-differences analysis (described in the text). A description and list of the 29 states in the control group and 11 states that were excluded are in the text and the exhibit 1 notes. Models were controlled as described in the exhibit 1 notes. Difference-in-differences models included state and year fixed effects. 95% confidence intervals (indicated by whiskers) were based on robust standard errors.

Sensitivity Analyses

The difference-in-differences results using alternative comparison states were all qualitatively similar to the main model (exhibit 4). More detailed results using the alternative comparison states are reported in the appendix (exhibit D1–D4).³² Subgroup results by race were generally consistent, but did moderately vary with alternative control groups (appendix exhibit D4).³²

Exhibit 4:

Changes in abortion and birth rates in Illinois relative to control states before and after the introduction of Medicaid coverage for abortion, using alternative control groups, 2014–21

SOURCES Centers for Disease Control and Prevention Abortion Surveillance System and vital statistics data, 2014–21. NOTES The figure presents results of the authors’ difference-in-differences analysis (described in the text). A description and list of included control states are in the text and in the exhibit 1 notes. Models were controlled as described in the exhibit 1 notes. Difference-in-differences models included state and year fixed effects. 95% confidence intervals (indicated by whiskers) were based on robust standard errors.

We also found that excluding control variables did not alter the findings (appendix exhibits D5 and D6),³² and the bootstrap method had similar results for both abortion and birth rates, with the exception of 2018 birth rates (appendix exhibit D7).³²

Discussion

In this quasi-experimental study, we found that introducing Medicaid coverage for abortion care in Illinois increased the abortion rate by 2.43 abortions per 1,000 females, on average, over the course of five years. This is equivalent to an 18.3 percent increase from baseline, or 6,093 more abortions by Illinois residents per year. This is similar to the expected change when distance to the nearest abortion provider decreases from 100 to 0 miles.^44,45 Our results suggest that in Illinois, Medicaid coverage is an especially important factor that is at least as important as geographic availability in accessing abortion care.

Our event-study estimates demonstrated that the effects on abortion appeared immediately after the new policy took effect in 2018 and generally increased over time, with a downward bump in 2020, likely due to COVID-19. The observed increase over time may be attributable to increased awareness of abortion coverage in Medicaid by the public, increased enrollment in Medicaid, or increased clinic capacity and familiarity with Medicaid reimbursement processes.⁴⁶ Medicaid enrollment grew after the COVID-19 public health emergency that required states to continue enrollment from March 2020.⁴⁷ From 2017 to 2020, Illinois saw a 20 percent increase in abortion clinics from twenty-five clinics to thirty clinics, whereas the control states had a 4 percent increase, on average.⁴⁸

Our finding that Medicaid abortion coverage decreased birth rates by 1.66 births per 1,000 females is equivalent to a 2.8 percent decrease from baseline, or 4,162 fewer births by Illinois residents per year. Much like our abortion findings, the birth rate results closely mirror the expected change observed when the distance to the nearest abortion provider decreases from 100 to 0 miles, and interestingly is similar in magnitude to recent estimates of the effect of state-level abortion bans under Dobbs.^44,49 The birth results are particularly important because unplanned births have consequences for health and well-being.^4,50,51 The risk for death is thirty-five times higher for people who remain pregnant relative to those who obtain an abortion.⁵²

Our estimates on birth rates are not aligned with our estimates on abortion rates—the effect we observe on abortions is 46 percent larger than the effect we observe on births. However, the two estimates are not fully comparable. The abortion effects grow over time, but our birth analysis ended a year before the last year of available abortion counts because of the nine-month pregnancy lag. It is also possible that there are different measurement error patterns in the abortion and birth data. Nonetheless, the discrepancy between the observed abortion and birth effects suggest caution when interpreting our results.

Among the thirty-four states that do not provide Medicaid coverage for abortion, twenty states remain without a total ban under Dobbs and are generally supportive of abortion.^11,53 There are a few possible explanations for why these twenty states have not opted to cover abortion. First, states that are generally supportive of abortion might not offer Medicaid abortion coverage because of budget concerns. Second, there might not be sufficient political will to overcome the default arrangements under Hyde. Policy preferences in these states might value freedom from abortion restrictions, but not the proactive funding of abortion. Indeed, it has often been assumed—and used to support Hyde restrictions—that there is a broad consensus against taxpayer funding for abortion, regardless of one’s stance on abortion legality, although there is limited empirical evidence supporting that claim.⁵⁴ Third, support for Medicaid abortion coverage is likely related to the perceived “deservingness” of people who need such coverage.^55,56 Fourth, the unequal political influence of low-income people may also contribute to an absence of action.⁵⁷ However, one unintended effect of Dobbs may be that it redirects public opinion toward more proactive access policies in states that are otherwise supportive of abortion.

Our results are consistent with previous work that shows that cost is an important determinant of abortion access. In Louisiana, which restricted Medicaid abortion coverage before its total ban, people with low incomes reported cost-related hardships in their abortion care because of lack of public funding, whereas in Massachusetts, where Medicaid abortion coverage is provided, abortion care was reported as affordable and accessible.^27,28 Alongside prior research, our results suggest that Medicaid abortion coverage improves reproductive equity, as Medicaid enrollees are low-income by definition and disproportionately Black and Hispanic. In the subgroup analyses on birth rates, we observed larger effects for Black and Hispanic populations and people living in higher-poverty areas.

Our results are consistent with previous findings from Illinois. Another recent study also found evidence that the introduction of Medicaid abortion coverage was followed by increased procedure volume in study-participating clinics’ medical records.⁵⁸ A qualitative study of stakeholders in Illinois found that the introduction of Medicaid abortion coverage substantially reduced financial burdens, improved access and quality of abortion care, and reduced emotional stress.⁴⁶ A cross-sectional study found that introducing Medicaid abortion coverage in Illinois was associated with reduced out-of-pocket spending for abortion care.²¹ However, research also underscores barriers to providing Medicaid-reimbursed abortion in Illinois, where providers report concerns about low reimbursement rates and implementation difficulties.^46,59 More work is needed on how these issues affect providers and patients.

Our findings also have implications for abortion access in the era of Dobbs. Evidence from the post-Dobbs period suggests that abortions increased in states where abortion is permitted, especially when they neighbor banned states.⁶⁰ This increase is likely from residents of banned states traveling to obtain a legal abortion. Medicaid abortion coverage may help providers meet the increased demand, as it will relieve financial strain on the system by covering the care of in-state residents.^46,61

This study suggests that Medicaid can play an important role in helping people access abortion care, especially those at increased social vulnerability. As state policies evolve in the post-Dobbs context, Medicaid abortion coverage will continue to be a vital factor in reproductive access.

Biographies

Bio1: Taehyun Kim (taehyun@wisc.edu), University of Wisconsin–Madison, Madison, Wisconsin.

Bio2: Jacqueline Ellison, University of Pittsburgh, Pittsburgh, Pennsylvania.

Bio3: Julia Renee Steinberg, University of Maryland, College Park, Maryland.

Bio4: Michel H. Boudreaux, University of Maryland, College Park.