Lifetime abortion incidence when abortion care is covered by Medicaid: Maryland versus five comparison states

Abstract

Objective

To estimate the association of Medicaid coverage of abortion care with cumulative lifetime abortion incidence among women insured by Medicaid.

Data Sources and Study Setting

We use 2016–2019 (Pre‐Dobbs) data from the Survey of Women studies that represent women aged 18–44 living in six U.S. states. One state, Maryland, has a Medicaid program that has long covered the cost of abortion care. The other five states, Alabama, Delaware, Iowa, Ohio, and South Carolina, have Medicaid programs that do not cover the cost of abortion care. Our sample includes 8972 women residing in the study states.

Study Design

Our outcome, cumulative lifetime abortion incidence, is identified using an indirect survey method, the double list experiment. We use a multivariate regression of cumulative lifetime abortion on variables including whether women were Medicaid‐insured and whether they were residing in Maryland versus in one of the other five states.

Data Collection/Extraction Methods

This study used secondary survey data.

Principal Findings

We estimate that Medicaid coverage of abortion care in Maryland is associated with a 37.0 percentage‐point (95% CI: 12.3–61.4) higher cumulative lifetime abortion incidence among Medicaid‐insured women relative to women not insured by Medicaid compared with those differences by insurance status in states whose Medicaid programs do not cover the cost of abortion care.

Conclusions

We found that Medicaid coverage of abortion care is associated with a much higher lifetime incidence of abortion among individuals insured by Medicaid. We infer that Medicaid coverage of abortion care costs may have a very large impact on the accessibility of abortion care for low‐income women.

What is known on this topic

• Cost of care is a significant barrier to abortion access among low‐income women.

• At the population level, states whose Medicaid programs cover costs of abortion care may have higher short‐term abortion rates.

What this study adds

• This study adds to understanding of abortion incidence by individual Medicaid insurance status and state Medicaid coverage policies.

• This study is the first to use an indirect method, the list experiment, in survey data to estimate abortion differentials by state Medicaid coverage policy.

• Using the measure of ever having had an abortion, this study finds evidence suggesting a very large impact of Medicaid coverage for abortion care costs on abortion incidence.

1. INTRODUCTION

Although the Dobbs v. Jackson Women's Health Organization decision represents a new and significant restriction on abortion services in many U.S. states, ¹ federal laws around payment for abortion care have long limited abortion access. The Hyde amendment, introduced in 1976, severely limited federal funds that can be spent on abortion services with exceptions made only for rape, incest, and endangering the life of the mother. ² Individual state Medicaid programs must choose to provide funding for abortion care using state‐only funding sources. Only 17 states have opted to provide this funding to cover all or most medically necessary abortions, and these states differ in the accessibility of abortion services, whether the legislature or courts required that Medicaid funding may be used to pay for abortion care, the reimbursement rate for abortion services, and how long the policies have been in effect. ³ , ⁴ , ⁵ Some additional states, in theory, provide state funding for abortion care, but in practice, have restrictions such that state funds are rarely able to be accessed for payment of abortion services. For example, Iowa has, since 2013, allowed state funds to be used for abortion care but release of funds requires approval from the governor's office, and this state funding has very rarely been utilized. ⁶ , ⁷

In states that do not opt for state funds to cover abortion services, individuals insured by Medicaid must generally pay out‐of‐pocket for abortion services, creating a substantial barrier to care. ⁸ , ⁹ In 2022, the average cost for a medication abortion was $580 dollars and clinic first‐trimester abortions cost around $600. ¹⁰ Second‐trimester abortions are substantially more expensive, costing between $1500–$2000. ¹⁰ A 2020 study identified the out‐of‐pocket costs of a first trimester abortion to be at least 40% of median monthly non‐subsistence household income in 39 states. ¹¹

Existing literature has largely studied associations of public funding policy (and policy change) on abortion and birth rates using aggregate data. ¹² , ¹³ , ¹⁴ , ¹⁵ These studies have generally found small increases in abortion incidence, with larger effects observed among those at the highest risk for unintended pregnancy. A review of these studies ¹⁶ notes, however, substantial methodological challenges to assessing the impact of Medicaid funding for abortion care, as most studies do not have information on abortions by whether the woman was Medicaid‐insured. Population‐representative surveys, which can collect information on the individual's Medicaid insurance, have been rejected as a data source because of longstanding problems of underreporting of abortion in survey data sources asking direct questions about abortion experiences. ¹⁷ , ¹⁸ Few studies have therefore used population‐representative surveys to study abortion incidence and factors associated with abortion. ¹⁹ , ²⁰

Recent work suggests that an indirect survey method, the abortion list experiment, may improve measurement of abortion within surveys in the U.S. context. ²¹ , ²² In the list experiment, respondents are randomly assigned to receive a treatment or control list and are asked how many, but not which, items on the list apply to them. A sensitive item, here having had an abortion, appears only in the treatment list. The difference in mean number of items reported between treatment and control lists estimates the incidence of the sensitive item. By asking respondents how many but not which items apply to them, the list experiment avoids revealing which individuals have had an abortion. Properly designed, the list experiment allows respondents to report on a sensitive item with less stigma and reduces privacy concerns of revealing sensitive information. ²³

To our knowledge, only two population‐representative studies have used the list experiment to estimate lifetime abortion incidence in the U.S. context. One study concluded positively about the utility of the method where the other found that the list experiment did not improve the measure of abortion incidence relative to direct questions. ²¹ , ²⁴ After many decades of failure of direct questions to measure abortion incidence, ¹⁷ , ¹⁸ , ²⁵ and the failure of other alternative approaches such as the Network Scale Up Method, ²⁶ the confidante method, ²⁷ and Respondent Driven Sampling ²⁸ to address underreporting of abortion in surveys, the positive early findings ²¹ , ²² from the abortion list experiment deserve further consideration.

In this study, we examine associations of women's current Medicaid‐insured status with cumulative lifetime abortion incidence by whether a state covers the cost of abortion care, using data from five states in which state Medicaid programs do not cover abortion care costs and one state, Maryland, whose state Medicaid program has long covered the cost of abortion services. In 1981, the Maryland Court of Appeals ruled in Kindley v. Governor of Maryland that the state legislature could, but was not obligated to, provide public funding to cover abortion costs, and that both therapeutic and non‐therapeutic abortion care costs could be covered. This case was decided in response to a bill of equity filed against state of Maryland officials who provided medical provider reimbursements for abortion services for indigent persons as part of the budget bill for fiscal year 1978. ²⁹ Since the early 1980s, the Maryland state government has generally provided public funding for abortion care. ³⁰ , ³¹ , ³² , ³³ , ³⁴ , ³⁵ , ³⁶

2. METHODS

2.1. Data sources

This study uses state‐level population‐representative survey data from the Statewide Survey of Women (SoW) in six diverse states, geographically and with respect to reproductive health policy: Alabama, Delaware, Iowa, Maryland, Ohio, and South Carolina ³⁷ , ³⁸ , ³⁹ with surveys conducted between 2016–2019 by NORC at the University of Chicago. The SoW recruited respondents from an address‐based sampling frame and is a multi‐mode study. Response rates, calculated using the American Association for Public Opinion Research (AAPOR) definition 3 with Council of American Survey Research Organizations (CASRO) assumptions, range from 22.9% in Maryland to 38.0% in Iowa and are listed for all study states in Appendix 1, Table A1. The majority of respondents (64%) completed an online questionnaire, and a sizable minority (36%) responded to a paper questionnaire mailed to their household; very rarely (<1%) computer‐assisted telephone interviewing was also conducted. The sampling design developed by NORC includes some oversampling by place of residence, race/ethnicity, and socioeconomic status which varied by state; however, across states, survey weights were developed so the collected sample can be re‐weighted to represent residents of a state who identify as women and are between the ages of 18 and 44. ³⁹ Although not all individuals who identify as women are at risk for pregnancy and not all individuals who obtain abortions are women, we refer to our respondents as women reflecting the survey recruitment criteria and weighting.

The SoW has previously been used to estimate abortion incidence by sociodemographic characteristics in Delaware and Maryland ²¹ and in Ohio. ²⁴ Of the 12,299 completed or partial interviews, we restrict our analytic sample to women who had a valid survey weight (N = 12,277), responded to the list experiment questions (N = 11,675), reported their age to be between 18 and 44 (N = 11,216), provided information on their Medicaid insurance status (N = 9465), and who reported not moving in the past year, for a total sample size of 8972. In analyses controlling for all sociodemographic characteristics, we use listwise deletion for cases missing values on any of these measures. In these analyses, our sample size is 8743.

2.2. Outcome: cumulative lifetime abortion incidence

To measure our outcome, cumulative lifetime abortion incidence, we use the abortion list experiment. We estimate abortion by sociodemographic characteristics using SoW observations pooled across all six study states. To maximize statistical power, our study uses a double list experiment (shown in Table 1) such that each respondent receives one treatment list and a distinct control list with non‐sensitive items modeled after a previous pilot study. ⁴⁰ The difference in means between treatment and control lists represents the cumulative lifetime abortion incidence. The list experiment may improve reporting of sensitive behaviors, such as abortion, by asking how many but not which behaviors respondents experienced, provided that certain design assumptions are met. ⁴¹ , ⁴²

TABLE 1.

Double list experiment items administered in baseline survey across six states.

On the following list of health experiences, how many of these have you personally experienced? You don't need to say which ones, just how many
Version A‐ Control	Ever used or taken medication for which a prescription is needed Ever had a pap smear Diagnosed with breast cancer in the past 10 years
Version A‐ Treatment	Ever had an abortion (ended a pregnancy on purpose) Ever used or taken medication for which a prescription is needed Ever had a pap smear Diagnosed with breast cancer in the past 10 years
Version B‐ Control	Ever used a birth control method (such as: pills, an IUD or implant, condoms or the shot) Had a tubal or ectopic pregnancy in the past year Ever had your blood pressure measured
Version B‐ Treatment	Ever had an abortion (ended a pregnancy on purpose) Ever used a birth control method (such as: pills, an IUD or implant, condoms or the shot) Had a tubal or ectopic pregnancy in the past year Ever had your blood pressure measured

Note: IUD refers to an intrauterine device.

2.3. Measures

Our main explanatory variable, Medicaid‐insured, is asked directly in the survey using question text adapted from the American Community Survey. ³⁸ The question asks respondents whether they currently have employer sponsored, direct purchase, Medicare, Medicaid, military, Indian Health Service (IHS), or any other insurance coverage with a write‐in option provided. Respondents are classified as Medicaid‐insured if they checked that they have Medicaid insurance and are classified as uninsured if they responded no to all non‐write‐in insurance types. We excluded write‐in responses because rates of write‐in response were low and, consistent with prior research, ⁴³ write‐ins were generally duplicative of insurance type previously reported or did not contain sufficient information to identify an insurance type. We include respondents who report being insured by both Medicaid and another insurance type as Medicaid‐insured. In sensitivity analyses (described below), we show results are robust to excluding respondents who report concurrently having Medicaid and another insurance type. We also control for other insurance reported by respondents classifying insurance types as private (employer sponsored or direct purchase), another coverage type (Medicare, IHS, or military) or uninsured. Due to respondent reporting of dual coverage, insured respondents may be insured by multiple insurance types.

The survey instruments also feature common questions on sociodemographic characteristics that we use in our study. Respondents self‐report their race and ethnicity (Non‐Hispanic White, Non‐Hispanic Black, Hispanic or Other Race Ethnicity), education (Less than High School, High School, Some College, Bachelor's Degree or Higher), household income (Less than $25,000, $25,000–$49,999, Greater than or Equal to $50,000), parity (No Live Births, 1 Live Birth, 2 Live Births, 3 Live Births, 4+ Live Births), and marital status (Married, Cohabiting, Single). Finally, state of residence is included in all analytic models and is known from the survey's address‐based sampling frame.

2.4. Statistical analysis

We use the list experiment to estimate cumulative lifetime abortion differentials by Medicaid status and state of residence. Our analysis uses a multivariate regression that pools data across all study states and applies a linear estimator. ⁴⁴ Our first regression predicts cumulative lifetime abortion incidence as a function of single‐year age (centered at age 25), state of residence, Medicaid‐insured status, and the interaction of Medicaid‐insured status with Maryland residency. Our key parameters of interest are the main effect of being Medicaid‐insured on abortion incidence, which, without income controls, we expect to be positive given a long literature showing that women with lower socioeconomic status are more likely to have had an abortion, ⁴⁵ , ⁴⁶ and the interaction of being Medicaid‐insured and living in Maryland, which we also expect to be positive, as Maryland's Medicaid program has covered the cost of abortion care while other states included in our analysis have not.

In our results, we report: (1) The difference in abortion incidence among women insured by Medicaid versus not insured by Medicaid residing in Maryland which is calculated as the main effect of being Medicaid‐insured plus the interaction of being Medicaid‐insured and living in Maryland; (2) the difference in abortion incidence among Medicaid‐insured women versus women not insured by Medicaid residing in other states which is calculated as the main effect of Medicaid; and (3) the difference in abortion incidence between women insured by Medicaid and women not insured by Medicaid residing in Maryland compared to the difference between women insured by Medicaid and not insured by Medicaid living in other study states which is calculated as the interaction of being Medicaid‐insured and living in Maryland. In Appendix 3, we show our regression equations and describe how these differences are calculated from regression coefficients. As this is a linear probability model, a coefficient of 0.01 corresponds to a 1 percentage‐point change in the probability of abortion during a woman's lifetime.

In our main analysis, we include all women in the analytic sample regardless of whether, based on income, they are likely Medicaid eligible or not. We do this because the list experiment has larger standard errors than analysis of direct questions, and so we aim to retain cases in order to ensure sufficient statistical power. ⁴⁷ In a sensitivity analysis, we restrict our sample to individuals living in households with incomes below $50,000 (approximately 200% of the federal poverty level for a family of 4 in 2018 ⁴⁸ ) who we expect may be least likely to afford abortion services if Medicaid does not cover care costs.

A concern is that we are comparing across states that differ in the availability of reproductive health services and in population demographics and socioeconomic status. Thus, we include a fixed effect for state of residence to account for time invariant features of states. To account for potential demographic and socio‐economic differences, we also run a second model that controls for race and ethnicity, education, household income, parity, marital status, and health insurance in addition to age controls and the state of residence fixed effect. In another sensitivity analysis, we restrict the sample to women living in Maryland and Delaware. These two states have historically had high abortion rates ⁴⁹ and populations with similar demographic characteristics (see Appendix 1, Table A1).

For each of our control measures, age, race/ethnicity, education, income, parity and marital status, we used Wald tests to check for significant interactions between the characteristic and state of residence. We conducted this testing to determine if factors associated with abortion incidence differed across states. Failing to find any significant interaction between control measures and state of residence, we include only the interaction term for Medicaid‐insured by Maryland residence in our final models.

Finally, the list experiment requires design assumptions to be met, specifically that respondents are randomized across treatment and control lists and that responses to the non‐sensitive items are not affected by the inclusion of the sensitive item in the list. We test the first assumption by examining the race and ethnicity, age, and socio‐economic characteristics by list assignment using chi squared testing and evaluating the design‐based F‐statistics. We assess the second assumption using the Blair and Imai design effect test ⁴¹ and report both sets of results in Appendix 2.

All analyses are done in Stata 18. We use survey weights normalized by state of residence in all regression and descriptive analyses. The Blair‐Imai design effect test does not use survey weights as it examines responses of the sample rather than generalizing to a population. The Institutional Review Board at the University of Maryland reviewed this study and determined it exempt as data is deidentified and only secondary analysis was conducted.

3. RESULTS

Across the six study states, approximately 20% of women are insured by Medicaid. We do not find that levels of Medicaid insurance differ significantly between Maryland and the comparison study states in 2016–2019 (Table 2); however, there is variation in percent insured by Medicaid across comparison states with some states (Ohio) having a larger share of women insured by Medicaid and other states (Alabama and South Carolina) having a smaller share of women insured by Medicaid (See Appendix 1, Table A1). We find differences between Maryland and the other five states in terms of private insurance, proportion uninsured, race and ethnicity, education, income, parity, and marital status, and again there is variation across comparison states on demographic and socioeconomic characteristics (See Appendix 1, Table A1). Relative to the average of comparison states, women in Maryland were more likely to be covered by employer‐sponsored or direct‐purchase insurance, less likely to be uninsured, less likely to be non‐Hispanic White, more likely to have a four‐year college degree, more likely to be living in a household with an income above $50,000, more likely to be nulliparous, and more likely to be single.

TABLE 2.

Demographic and socioeconomic characteristics of women in six states, 2016–2019.

	Alabama, Delaware, Iowa, Ohio, South Carolina N = 8043	Maryland N = 929
	Proportion	Proportion
Age	p = 0.757
18–24	0.232	0.251
25–29	0.198	0.205
30–34	0.215	0.211
35–39	0.170	0.164
40–44	0.185	0.167
Race/Ethnicity	p < 0.001
Non‐Hispanic White	0.719	0.499
Non‐Hispanic Black	0.177	0.309
Hispanic or other race and ethnicity	0.104	0.192
Education	p = 0.013
Less than high school	0.039	0.030
High school	0.152	0.104
Some college	0.429	0.411
Bachelors degree or higher	0.380	0.455
Household income	p < 0.001
Below $25,000	0.243	0.141
Between $25,000 & $49,999	0.232	0.132
At or Above $50,000	0.525	0.728
Marital status	p = 0.002
Married	0.465	0.394
Cohabiting	0.189	0.179
Single	0.346	0.427
Parity	p = 0.024
0 Live births	0.463	0.511
1 Live births	0.176	0.159
2 Live Births	0.206	0.225
3 Live births	0.108	0.072
4+ Live births	0.047	0.033
Health insurance a
	p = 0.538
Medicaid	0.204	0.191
	p = 0.009
Employer sponsored or direct purchase	0.705	0.765
	p = 0.1538
Other public or military	0.046	0.060
	p < 0.001
Uninsured	0.121	0.048

Note: Baseline data collection occurred between 2016and 2019 (See Appendix 1, Table A1 for dates of administration by state). Sample restricted to respondents reporting on the list experiment, age and Medicaid coverage status and who did not move between states in the past year. Estimates use survey weights normalized by state. p values calculated from corrected F‐statistic of no difference in distribution between Maryland and the other five states.

^a Medicaid, Employer Sponsored or Direct Purchase, and Other Public or Military insurance types are not mutually exclusive due to dual coverage.

In Table 3, we show calculated difference terms from two regression models examining associations of Medicaid with cumulative lifetime abortion incidence. In our base model adjusting for only age and state of residence (Model 1), women insured by Medicaid and not residing in Maryland have a 6.0 percentage‐point greater probability (CI: 0.1–11.8) of having had an abortion during their lifetimes compared with women with another insurance status not residing in Maryland (p < 0.05). This itself is substantively significant and consistent with prior studies showing higher abortion incidence among women with low socioeconomic status. ⁴⁵ However, in Maryland, women insured by Medicaid have a 47.9 percentage point higher probability (CI: 24.3–71.5) of having had an abortion compared with women not insured by Medicaid. We therefore find that in Maryland, which provides state funds to cover abortion costs for the Medicaid‐insured, Medicaid‐insured women are 41.9 percentage points (CI: 17.6–66.2) more likely than women not insured by Medicaid to have had an abortion during their lifetimes relative to the difference between women insured by Medicaid versus not insured by Medicaid residing in other states.

TABLE 3.

Association of Medicaid with cumulative lifetime abortion incidence in six states, 2016–2019.

	(1) Difference between Medicaid and not Medicaid‐insured residing in Maryland	(2) Difference between Medicaid and not Medicaid‐insured residing in non‐Maryland state	(3) Difference between Medicaid‐insured and not Medicaid‐insured residing in Maryland versus difference between Medicaid‐insured and not Medicaid‐insured residing in non‐Maryland state
Model 1: Adjusting for age and state of residence N = 8972	0.479*** (0.243–0.715)	0.060* (0.001–0.118)	0.419*** (0.176–0.662)
Model 2: Adjusting for age, state of residence, race, income, education, marital status, parity, and health insurance status N = 8743	0.377** (0.134–0.621)	0.008 (−0.067–0.082)	0.370** (0.123–0.614)

Note: Estimated differences are calculated from a linear probability model with survey weights and multiplied by 100 represent the percentage point difference in the outcome. Column 2 corresponds to the coefficient for Medicaid; Column 3 corresponds to the interaction of Medicaid and Maryland state of residence; and Column 1 corresponds to the coefficient for Medicaid plus the interaction of Medicaid with Maryland state of residence with uncertainty calculated using the lincomest command in Stata 18. Full set of regression coefficients are shown in Appendix 3, Table A5. Survey weights are normalized by state. 95 percent confidence intervals in parentheses; +p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Source: Survey of Women 2016–2019 Baseline Surveys.

After controlling for race, education, income, parity, marital status, and health insurance (Model 2), women outside the state of Maryland no longer have a higher abortion incidence relative to women with another insurance status. However, in Maryland, Medicaid‐insured women are 37.7 percentage points (CI: 13.4–62.1) more likely to have had an abortion during their lifetimes relative to women with another insurance status. In this adjusted model, Medicaid‐insured women in Maryland are 37.0 percentage points (CI: 12.3–61.4) more likely than women not insured by Medicaid to have had an abortion compared to the difference in abortion incidence by Medicaid insurance status for women living in another study state.

In Appendix 3 (Table A5), we present all regression coefficients from our two model specifications. Coefficients show that abortion incidence differs by state of residence as well as race and ethnicity and socioeconomic status. Non‐Hispanic Black women have higher lifetime abortion incidence than non‐Hispanic White women, cohabiting women have higher lifetime incidence than married women, and women with some college have higher lifetime abortion incidence than college‐graduate women. These findings are in keeping with other studies which have found that Non‐Hispanic Black and women of lower socioeconomic status have a higher risk of abortion. ⁴⁶

Critical to interpretation of results, we find that list experiment assumptions of randomization and response consistency are met across study states. We detail statistical testing of these assumptions in Appendix 2.

3.1. Sensitivity analyses results

In our main analysis, we use Medicaid as self‐reported by survey respondents. A long body of research has found some misreporting of Medicaid insurance status and a tendency of survey respondents to underreport Medicaid coverage. ⁵⁰ , ⁵¹ , ⁵² , ⁵³ In a sensitivity analysis, we run our same model specifications but exclude respondents who report being currently insured by Medicaid and some other insurance. These results are reported in Table 4 and are consistent with previously described results.

TABLE 4.

Sensitivity analyses of association of Medicaid with cumulative lifetime abortion incidence in six states, 2016–2019.

	(1) Difference between Medicaid and not Medicaid‐insured residing in Maryland	(2) Difference between Medicaid and not Medicaid‐insured residing in non‐Maryland state	(3) Difference between Medicaid‐insured and not Medicaid‐insured residing in Maryland versus difference between Medicaid‐insured and not Medicaid‐insured residing in non‐Maryland state
Sensitivity analysis 1: excludes women reporting Medicaid and another insurance type
Model 1 adjusting for age and state of residence N = 7992	0.515*** (0.269–0.760)	0.064+ (−0.001–0.127)	0.451** (0.198–0.704)
Model 2 adjusting for age, state of residence, race, income, education, marital status, parity, and health insurance status N = 7782	0.422** (0.171–0.627)	−0.010 (−0.099–0.079)	0.432** (0.181–0.682)
Sensitivity analysis 2: excludes women living in households with incomes over $50,000
Model 1 adjusting for age and state of residence N = 3518	0.691*** (0.332–1.050)	0.023 (−0.049–0.094)	0.669*** (0.303–1.034)
Model 2 adjusting for age, state of residence, race, income, education, marital status, parity and private health insurance a N = 3399	0.607*** (0.239–0.975)	−0.018 (−0.106–0.071)	0.625*** (0.259–0.990
Sensitivity analysis 3: excludes women not residing in Maryland or Delaware
Model 1 adjusting for age and state of residence N = 1807	0.484*** (0.250–0.720)	0.219+ (−0.008–0.446)	0.266 (−0.055–0.587)
Sensitivity analysis 4: excludes women not residing in Maryland or Delaware and in households with incomes above $50,000
Model 1 adjusting for age and state of residence N = 558	0.696*** (0.337–1.055)	0.200 (−0.097–0.498)	0.496* (0.035–0.956)

Note: Estimated differences are calculated from a linear probability model with survey weights and multiplied by 100 represent the percentage point difference in the outcome. Column 2 corresponds to the coefficient for Medicaid; Column 3 corresponds to the interaction of Medicaid and Maryland state of residence; and Column 1 corresponds to the coefficient for Medicaid plus the interaction of Medicaid with Maryland state of residence with uncertainty calculated using the lincomest command in Stata 18. 95% confidence intervals in parentheses; +p < 0.10; *p < 0.05; **p < 0.01; ***p < 0.001.

Source: Survey of Women 2016–2019 Baseline Surveys.

^a Due to small sample size, we pool uninsured individuals and individuals insured by other coverage types in this model. Full set of regression coefficients are shown in Appendix 4, Tables A6–A8. Survey weights are normalized by state.

Our main analysis compares Medicaid‐insured women to all women not covered by Medicaid insurance because we want to maximize available sample size due to the statistical inefficiency of the list experiment ⁴⁷ ; however, we expect that the largest differences may be observed in low‐income households that might not otherwise afford the cost of abortion care. In a second sensitivity analysis, we restrict our sample to households with incomes under $50,000. Adjusting for demographic and socioeconomic controls, Medicaid‐insured women outside of the state of Maryland do not differ from women of other insurance statuses in the probability of having had an abortion. In Maryland, Medicaid‐insured women are 60.7 percentage points (CI: 23.9–97.5) more likely to have had an abortion compared with women not insured by Medicaid but living in Maryland. The difference between women insured by Medicaid and not insured by Medicaid by Maryland state of residency is 62.5 percentage points (CI: 25.9–99.0) (see Table 4).

A final sensitivity analysis restricts our sample to women residing in Maryland or Delaware, two states that have a similar abortion incidence and demographic profile but differ in state funding for abortion care. While the sample size for this analysis is quite small, we find that, in Maryland, Medicaid‐insured women are 48.4 percentage points (CI: 25.0–72.0) more likely to have had an abortion compared with Maryland residents with a different insurance status. In Delaware, Medicaid‐insured women are 21.9 percentage points (CI: −0.8 – 44.6) to have had an abortion compared with women of a different insurance status. The difference in abortion incidence between Medicaid‐insured and not Medicaid‐insured women in Maryland versus Delaware is not statistically significant, but remains substantively quite large, 26.6 percentage points (CI: −5.5 to 58.7). When we restrict the Maryland and Delaware sample to women living in households with incomes under $50,000, we find that the difference in abortion incidence between women insured by Medicaid and not insured by Medicaid is 49.6 percentage points (3.5–95.6) greater in Maryland than it is in Delaware. This difference is statistically significant as shown in Table 4. Coefficients for all sensitivity analyses are shown in Appendix 4 (Table A6–A8).

4. DISCUSSION

We find evidence that in Maryland, a state that has long provided Medicaid funding for low‐income women's abortion care, women insured by Medicaid are much more likely to have had an abortion over their reproductive lifetimes than are Maryland residents not insured by Medicaid, relative to the difference in abortion incidence by Medicaid‐insurance status in other states. This evidence is conceptually a “difference in difference” result, represented by an interaction coefficient for Maryland‐resident by Medicaid‐insured. This interaction coefficient is very large before (41.9% greater cumulative lifetime incidence) and after (37.0% greater cumulative lifetime incidence) controlling for sociodemographic differences between women in Maryland and the five comparison states. We do not find evidence that being Medicaid‐insured is itself associated with a greater likelihood of having an abortion after controlling for sociodemographic characteristics associated with the likelihood of being Medicaid‐insured, including marital status, race and ethnicity, household income, parity and educational attainment. In supplemental analyses, we find that these conclusions are robust to how we specify Medicaid‐insured and restricting the sample to households with incomes below $50,000.

These findings, while observational, strongly suggest that Maryland's public funding for low‐income women's abortion care is associated with increased utilization of abortion care. This also suggests that in states that do not provide Medicaid coverage for abortion care, there may be a large unmet need for abortion care among low‐income women. For states seeking to protect access to abortion care in the Post‐Dobbs environment, allowing state funds to pay for abortion services for Medicaid‐insured persons may address socioeconomic disparities in abortion access. However, other factors, such as levels of abortion access in a state, distance from abortion providers, and Medicaid reimbursement rates may influence the extent to which state funding of Medicaid enables access to wanted abortion care. ⁵⁴ Even in Maryland, a state which has long covered the cost of abortion services, qualitative work has documented that abortion providers may have difficulties being reimbursed by Medicaid and receiving payment for the full cost of services. ³²

These findings are in keeping with previous studies that have identified costs as a substantial barrier to low‐income individuals receiving abortion care. ⁸ , ⁹ In addition to medical costs which are substantial for patients paying out‐of‐pocket, abortion patients are often faced with transportation, childcare, and lost work costs. ⁵⁵ Even prior to the Dobbs decision, analyses from the Turnaway study found that more than half of interviewed abortion patients paid at least one third of their personal monthly income in order to obtain an abortion procedure. ⁹ The recent Dobbs v. Jackson decision has likely increased the non‐health‐care costs for obtaining abortion care for patients in states that have chosen to criminalize or greatly restrict abortion services as patients have to travel significantly greater distances to receive these services. ⁵⁶

Our findings also align with a body of research that has used primarily administrative data to examine associations between Medicaid coverage for abortion care and abortion rates at the aggregate and over short periods. One recent study ¹⁵ found that discontinuation of public funds for abortion services decreased abortions among those insured by Medicaid while the introduction of public fund programs to cover abortion costs increased abortions among those insured by Medicaid, although not to levels observed in states with longstanding Medicaid programs covering abortion care. Leveraging variation over time in North Carolina policy, Cook and colleagues ¹² found that more abortions occurred when state funding of abortion care was available, with the largest associations for Black women and women aged 18–29. Using abortion counts data from Centers for Disease Control reports compiling state health department statistics and Guttmacher‐compiled information on abortion providing and funding, Haas‐Wilson ¹³ found that restrictions on Medicaid funding for abortion reduced abortion demand for minors. Similarly, analyzing Guttmacher Abortion Provider Census and National Center for Health Statistics data from the 1980s, Meier and McFarlane ¹⁴ found that public funding for abortion care significantly increased the number of abortions obtained and reduced the number of births to teenage mothers, low birth‐weight births, premature births, and women receiving late or no prenatal care. ¹⁴ Our study using survey data has the major advantage over the above studies using aggregate data by our having information on women's insurance coverage, and thus we are able to directly estimate differences in abortion‐incidence for Medicaid‐insured women between states with and without public funding for abortion care.

Administrative data sources have long been known to be limited by capturing only a few characteristics of abortion patients and not containing information on all states. ⁵⁷ , ⁵⁸ Surveys, which could be fielded in contexts without available administrative records, have not been used due to direct questions producing large underestimates of abortion incidence ¹⁸ and failing to capture abortion differentials. ⁵⁹ However, using the list experiment, we are able to obtain plausible estimates of abortion incidence, and capture expected abortion differentials by sociodemographic variables. These findings are consistent with a study using Survey of Women data in Maryland and Delaware that found expected differentials in abortion incidence by race, age, income, and parity. ²¹

Still, several limitations are of note. Because we are asking about abortion retrospectively, we do not observe a woman's insurance status or state of residence for the reproductive years in which she was exposed to or obtained an abortion. While we partially address this limitation by examining abortion incidence only among women who did not change state of residence in the most recent year, we do not have a full accounting of either women's interstate migration nor health insurance status through their reproductive lifetimes. Additionally, we use respondent self‐reports of Medicaid coverage. Prior literature has shown that survey respondents may not accurately know their insurance coverage and that some underreporting of Medicaid coverage is likely. ⁵⁰ , ⁵² , ⁶⁰ While our study suggests that large Medicaid differentials in abortion care are likely in states that provide public funding, there is a high sampling variance, and therefore wide confidence intervals, around our survey estimates. This limitation is intrinsic to the list experiment method which has substantially higher variance than direct question methods. ⁴⁷ Future studies could aim to improve the precision around estimates through surveys with larger sample sizes or by using combined data techniques. ⁶¹ Finally, our study is inherently observational in nature. Maryland consistently covered abortion care costs while other states did not. We are therefore not able to evaluate a difference in abortion incidence as a consequence of a policy change. Further, during the reproductive lifetime when women were at risk for abortion, other contextual factors like access to telehealth services for abortion care, varied across states and time ⁶² , ⁶³ and may have influenced abortion differentials. As such, factors other than Medicaid coverage may also have increased abortion care utilization for Medicaid‐insured individuals in Maryland relative to the study's comparison states. Future studies may consider fielding repeated cross‐sectional surveys with the list experiment so as to capture changes in abortion experiences before and after major policy changes in order to build this causal evidence.

Supporting information

Appendix S1. Supporting Information.

Jackson HM, Rendall MS. Lifetime abortion incidence when abortion care is covered by Medicaid: Maryland versus five comparison states. Health Serv Res. 2024;59(5):e14358. doi: 10.1111/1475-6773.14358