Medicaid Expansion Lowered Uninsurance Rates Among Nonelderly Adults In The Most Heavily Redlined Areas
By Jason Semprini, Abdinasir K. Ali and Gabriel A. Benavidez / October 2, 2023
Abstract
Medicaid expansion narrowed racial and ethnic disparities in health coverage, but few studies have explored differential impact by exposure to structural racism. We analyzed data on historical residential redlining in US metropolitan areas from the Mapping Inequality project, along with data on uninsurance from the American Community Survey, to test whether Medicaid expansion differentially reduced uninsurance rates among nonelderly adults exposed to historical redlining. Our difference-in-differences analysis compared uninsurance rates in Medicaid expansion and nonexpansion states both before (2009–13) and after (2015–19) the state option to expand Medicaid pursuant to the Affordable Care Act took effect in 2014. We found that Medicaid expansion had the greatest impact on lowering uninsurance rates in census tracts with the highest level of redlining. Within each redline category, there were no significant differences by race and ethnicity. Our results highlight the importance of considering contextual factors, such as structural racism, when evaluating health policies. States that opt not to expand Medicaid delay progress toward health equity in historically redlined communities.
Despite the significant advances made since the Civil Rights movement to outlaw racial discrimination and racist practices, the effects of structural racism remain persistent throughout the United States.1Structural racism is defined as the means by which a society perpetuates race-based discrimination through mutually enforcing systems, including systems related to housing, education, employment, credit, criminal justice, and health care policies and practices.2 One of the many ways in which structural racism manifests is in the form of health disparities among people who are members of minoritized racial and ethnic groups.3 Minoritized groups in the US, specifically Black, Hispanic, and Indigenous people, have higher rates of mortality and morbidity compared with their White counterparts.4 By systematically hindering access to resources necessary for achieving health and well-being, structural racism, rather than race, is the primary cause of racial health inequities in the US.1,2
A major barrier to accessing health care resources is the lack of health insurance coverage. Whether because of less access to employer-based coverage, less disposable income, or fewer available health insurance options, minoritized racial and ethnic groups are less likely than their White counterparts to be insured, and even when they are insured, they often face either higher out-of-pocket expenses proportional to their income or lower-quality coverage options.5 In 2010—the year the Affordable Care Act (ACA) was passed—the uninsurance rate for nonelderly, non-Hispanic White adults was 13 percent, whereas 20 percent of nonelderly non-Hispanic Black adults and 33 percent of nonelderly Hispanic adults were uninsured in that year.6
Background
Medicaid Expansion, Race, And Racism
The ACA contains many provisions aimed at increasing health care coverage. Arguably the most significant is expansion of Medicaid eligibility, which took effect as a state option in 2014.7,8 By 2016, Medicaid expansion had reduced uninsurance rates nationwide by 7 percent.7 Given the higher rates of uninsurance among nonelderly adults in minoritized racial and ethnic groups, Medicaid expansion has substantially narrowed racial and ethnic disparities in health insurance coverage in expansion states.9 There has been no shortage of research investigating differential effects of Medicaid expansion by individual characteristics such as race and ethnicity.10–12 Yet despite the growing consensus that structural racism, not race, causes disparities in health and health coverage, nearly all Medicaid expansion research has been based on self-reported individual race and ethnicity data. The mismatch between what investigators believe causes health inequities and the evidence base for policies aimed at advancing equity not only hinders progress toward this goal but also continues to perpetuate systems of structural racism.13,14
History Of Racist Redlining Practices
Our analysis focused on one measure of structural racism: the practice of residential redlining. In the 1930s the federal government established the Home Owners’ Loan Corporation (HOLC) to grade mortgage investment risk in neighborhoods across the country.15 The HOLC appraisal methods were explicitly racist and assigned investment risk on a four-point grade scale, as follows: best (A): “always upper or upper-middle class White;” desirable (B): “generally White, U.S. born neighborhoods;” declining (C): “areas with working-class or immigrants from Europe;” and hazardous (D): “infiltrated with undesirable populations such as Jewish, Asian, Mexican, and Black families.”15,16 Neighborhoods designated “hazardous” were depicted in red on HOLC maps. These grades affected access to and the terms of federally backed mortgages from financial institutions. Often, banks refused to lend to, or extorted excessive interest rates from, people in areas with the highest levels of redlining (HOLC grade D). Disproportionate access to credit for White middle- and upper-class families led to decades of wealth accumulation that was disproportionate to that of working-class and minoritized groups who lacked such access. Moreover, systematic disinvestment from minoritized communities prevented these communities’ residents from realizing the benefits of postwar economic prosperity experienced by White Americans.
Racist redlining practices persisted for more than thirty years until they were outlawed by the Fair Housing Act, which is part of the Civil Rights Act of 1968.17 Still, the effects of residential redlining continue today,17–19 as it created a substantial racial and ethnic gap in home ownership. Given that home ownership is the largest source of wealth for most Americans, the gap not only destabilized short-term housing prospects for families of minoritized racial and ethnic groups but also exacerbated generational wealth inequality.20 Chronic disinvestment from minoritized communities restricted access to resources, such as health care, education, and employment opportunities, in these areas.19 Emerging evidence has begun to show that exposure to redlining (living in previously redlined neighborhoods) is consistently associated with poor health and higher mortality rates.21–27
Our study examined the impact of Medicaid expansion on population-level uninsurance rates across levels of exposure to historical redlining.
Study Data And Methods
Data And Sample
We obtained census tract–level uninsurance rate data from the American Community Survey (ACS).28 Conducted by the US Census Bureau, the ACS uses a mix of mail, telephone, and in-person questionnaires to survey 3.5 million people annually. We analyzed the five-year average uninsurance rate data file for 2009–13 and 2015–19, excluding 2014 as the Medicaid expansion year. We accessed the ACS data from the International Public Use Microdata Surveys (IPUMS) project National Health Geographical Information System (NHGIS). IPUMS NHGIS collects, cleans, and integrates longitudinal ACS data for public research purposes.29 Deidentified ACS data from IPUMS NHGIS are publicly available and do not meet the definition of human subjects research.
We examined aggregated uninsurance rates for non-Hispanic Black, non-Hispanic White, and Hispanic nonelderly adults (ages 18–64). We focused on nonelderly adults because they were the most likely to gain coverage from Medicaid expansion, given that before 2014, children younger than age eighteen had higher Medicaid income eligibility thresholds and thus greater access to Medicaid than nonelderly adults, and adults ages sixty-five and older generally had access to coverage through Medicare. The treatment exposure of interest was state Medicaid expansion. We obtained state Medicaid expansion data from KFF.30 All states that expanded Medicaid pursuant to the ACA in 2014 were treatment states, and all states that had not expanded Medicaid under the ACA by January 2015 were control states. Following established practice, we excluded all states that expanded Medicaid during 2015–19, and we defined states that fully expanded Medicaid before 2014 (New York and Massachusetts) as controls.10,31
In addition to stratifying the sample by race and ethnicity, we grouped census tracts into four categories on the basis of historical redlining patterns, with 1 indicating the least amount of redlining and 4 indicating the most. We obtained verified historical redlining data from the Inter-university Consortium for Political and Social Research.32 These data were derived from the Mapping Inequality project, which digitized and mapped historical redline appraisal data.15 Redline categories were based on a population-weighted average of all HOLC grades for residential mortgage investment risk within each census tract.32 We excluded all tracts that were not subject to historical residential appraisal redlining, as defined by the Inter-university Consortium for Political and Social Research32 and the Mapping Inequality project.15
Research Design
Our primary models were specified as linear regressions to estimate the average effect of Medicaid expansion on uninsurance rates. To account for unobserved temporal trends and heterogeneity in uninsurance rates, our difference-in-differences design controlled for time period–level and census tract–level fixed effects.33 For inference, we estimated standard errors robust to heteroskedasticity, clustered at the state level.34 See online appendix A for technical details on the primary linear regression models.35
We separately estimated Medicaid expansion’s effect on uninsurance rates in the aggregated sample, for each of the three primary groups (non-Hispanic Black, non-Hispanic White, and Hispanic), and by each redline category. We then conducted three sets of joint tests of statistical significance. The first was to determine whether the effect of Medicaid expansion within each racial and ethnic group differed across the four redline categories. The second separately tested whether the effects in redline categories 2–4 differed from the effect in redline category 1. The third compared the effect of Medicaid expansion across racial and ethnic groups within each redline category. To give more weight to larger census tracts, all regression models incorporated standard analytic weights, measured by the total nonelderly population of each racial and ethnic group in the census tract.36 Statistical significance was set at �<0.05. All analyses used Stata, version 17.
Sensitivity Analyses
We conducted several sensitivity analyses to assess whether the robustness of our findings was sensitive to analytical decisions. First, to examine whether our inference testing for differential effects across and within redline categories changed in all US census tracts, we included census tracts that were not subject to historical HOLC redlining in the difference-in-differences models.
Next, to ensure that our findings were robust to alternative analytical decisions, we implemented alternative specifications, including the use of time-variant, census tract–level controls (poverty, unemployment, and education rates) and a time-variant group-specific control variable (the proportion of a racial or ethnic group receiving Supplemental Nutrition Assistance Program [SNAP] benefits), and an analysis without adjusting for population weights.37 We also included models estimating unclustered robust standard errors and robust standard errors clustered at the census tract level.34 Then we estimated the effect of Medicaid expansion on uninsurance rates when Massachusetts and New York (early expanders) were included as treated states, and then again when they were among the excluded states.37 We also added a model in which the three late-expanding states (Indiana, Louisiana, and Pennsylvania) were included as treated states.30 To account for outliers, we conducted an unconditional quantile regression, which accounted for census-tract and time fixed effects, to estimate the effect of Medicaid expansion on median uninsurance rates.38
Finally, we included a set of alternative insurance rate outcomes in the difference-in-differences models. We first explored the external validity and potential mechanisms of our primary models by examining the impact of Medicaid expansion on rates of private insurance, any public insurance, Medicare for people ages 65 and older, Medicaid for people younger than age 18, Medicaid for people ages 18–64, total US uninsurance rates, and uninsurance rates for all adults reporting incomes less than 100 percent of the federal poverty level. As placebo tests, we evaluated the impact of Medicaid expansion on uninsurance rates for the population in each racial and ethnic group younger than age 18 and ages 65 and older.
Limitations
This study had several limitations. First, because we used census tracts as the unit of analysis, we were unable to detect heterogeneity in uninsurance rates within census tracts. Second, we had limited ability to select granular data on uninsurance rates at the census tract level. Although we would have liked to analyze uninsurance rates among additional racial and ethnic groups and to stratify by socioeconomic factors, we were unable to further disaggregate the ACS data.
Third, our results should be generalized only to census tracts subject to HOLC redlining practices. These census tracts were limited to metropolitan cores, defined by the Department of Agriculture Economic Research Service as census-tract equivalents of urbanized areas.39 Washington, D.C., was not included. We do not doubt that exposure to other forms of structural racism in census tracts outside of metropolitan core areas and in Washington, D.C., could have led to disparate impacts of the ACA’s Medicaid expansion, but that research question was outside the scope of our work. The limitations of using HOLC redlining data as a measure of contemporary exposure to structural racism have been well documented.40 Most of the debate about analyzing historical HOLC maps to understand contemporary disparities relates to how the impact of redlining is conceptualized.40 Should analysts conceptualize redlining as a mechanism for segregation or as disinvestment from already segregated areas? Were HOLC redlining practices a cause or a consequence of racist lending?40 This debate does not have major implications for our study, as we did not attempt to measure the effect of redlining on housing or investment outcomes. Rather, we used redlining grades—a validated measure of exposure to racist residential lending discrimination—as the basis for subgroup analyses on the impact of a major US health policy.
Finally, although the main results extend only to nonelderly non-Hispanic Black, non-Hispanic White, and Hispanic populations, our alternative outcomes help assess the extent to which our findings can be extended to other subpopulations by income, age, and insurance status.
Study Results
Summary Statistics And Pre-ACA Uninsurance Rates
Our analyses of the 11,643 census tracts subject to historical redlining included areas that accounted for 25 percent of the non-Hispanic Black nonelderly adult US population (appendix exhibit 1).35 Of the most-redlined tracts (category 4), 1,928 were in Medicaid expansion states, and 1,793 were in nonexpansion states (appendix exhibit 2).35
Before 2014, in both Medicaid expansion and nonexpansion states, average uninsurance rates were lowest in census tracts in redline category 1 (11.0 percent in expansion states; 12.7 percent in nonexpansion states) and highest in category 4 (30.0 percent in Medicaid expansion states; 26.1 percent in nonexpansion states) (exhibit 1, appendix exhibit 3).35 Average uninsurance rates in the preexpansion period (2009–13) ranged from 9.9 percent for non-Hispanic White nonelderly adults in the lowest redline category (category 1) of census tracts in states that subsequently expanded Medicaid to 37.8 percent for Hispanic nonelderly adults in the highest redline category (category 4) of census tracts in such states (exhibit 2, appendix exhibit 3).35 In both Medicaid expansion and nonexpansion states, and across all racial and ethnic groups, pre-ACA uninsurance rates were highest in the highest redline categories (exhibit 2).
We also observed skewness and a positive association between redline categories and uninsurance rates across quantiles of uninsurance rates in 2009–13 (appendix exhibit 4).35 Similarly, higher-redline census tracts had higher rates of low-socioeconomic-status indicators (percentage living in poverty, percentage receiving SNAP benefits, and percentage unemployed), lower rates of adults with a bachelor’s degree, and larger populations of non-Hispanic Black and Hispanic adults during this period (appendix exhibit 5).35
Changes In Average Uninsurance Rates
Relative to the change in nonexpansion states, uninsurance rates among nonelderly adults in Medicaid expansion states in the aggregate fell by 6.2 percentage points in census tracts in redline category 4 (�<0.05) (exhibit 3). Conversely, we found no statistically significant effect of Medicaid expansion on average uninsurance rates among nonelderly adults in the aggregate in census tracts with redline scores 1–3.
Compared with the change in nonexpansion states, average uninsurance rates among nonelderly adults in redline category 4 census tracts declined from the pre period to the post period in expansion states for non-Hispanic Black adults (−5.7 percentage points; �<0.001), non-Hispanic White adults (−3.9 percentage points; �<0.01), and Hispanic adults (−7.9 percentage points; �<0.05).
Although the statistically significant percentage-point estimates suggest that Medicaid expansion lowered uninsurance rates in the highest redline category in the aggregate as well as for all racial and ethnic groups, our joint tests of significance within each racial and ethnic group across redline categories found that Medicaid expansion’s effects on uninsurance rates differed across all redline categories in the aggregate (�<0.01) and in the non-Hispanic Black population (�<0.05) (appendix exhibit 6).35 Our joint tests comparing the estimates for each redline category against the estimate for redline category 1 found significantly larger effects of Medicaid expansion in redline categories 3 and 4 among non-Hispanic White nonelderly adults and in redline category 4 among Hispanic nonelderly adults (�<0.05) (appendix exhibit 6).35 None of our joint tests of significance within each redline category found significant differences in the impact of Medicaid expansion across racial and ethnic groups (appendix exhibit 7).35
Sensitivity Analyses
When we examined the effect of Medicaid expansion across all census tracts, including those not subject to redlining, only in non-Hispanic White nonelderly adults did we find a significant decrease in uninsurance rates in census tracts not subject to redlining (appendix exhibit 8).35 Similar to our main results, however, Medicaid expansion’s impact on uninsurance rates did not significantly differ by race and ethnicity within census tracts not subject to redlining (appendix exhibit 9).35 Also consistent with the results of the first joint test of significance in our primary analysis, when we conducted these tests in the analysis across redline categories within each racial and ethnic group and included census tracts not subject to redlining, we likewise found that Medicaid expansion’s impact on uninsurance rates varied by redline category in the aggregate (�<0.01) and in the non-Hispanic Black (�<0.05) population (appendix exhibit 8).35
Further demonstrating the robustness of our primary model, our point estimates were not significantly different across all nine alternative linear regression specifications (appendix exhibit 10).35 Also supporting the interpretation of our main results is that for every alternative specification except for specification 7, in which Massachusetts and New York were considered treated, and specification 8, in which these two states were excluded, we consistently found that Medicaid expansion reduced uninsurance rates the most in the census tracts in redline category 4 (appendix exhibit 10).35 The results of the only nonlinear specification, a quantile regression model in alternative specification 10, show that Medicaid expansion had its greatest impact reducing median uninsurance rates in redline category 4 in all racial and ethnic groups (appendix exhibit 10).35
Finally, supporting the internal and external validity of our model, we found that Medicaid expansion had the largest impact increasing public insurance coverage for all adults in redline category 4 and had the largest impact lowering uninsurance rates among adults with incomes below 100 percent of poverty in redline categories 3 and 4 (appendix exhibit 11).35 We also found that Medicaid expansion’s impact on uninsurance rates for the total study population was highest in the census tracts in redline category 4, suggesting that our results extend beyond nonelderly adults, at least in the aggregate (appendix exhibit 11).35 Our placebo tests found that across all redline categories, Medicaid expansion had no significant effect on rates of private insurance coverage, Medicare coverage, or uninsurance among non-Hispanic Black, non-Hispanic White, or Hispanic children (appendix exhibit 11).35 Among adults ages sixty-five and older, we found that Medicaid expansion significantly lowered uninsurance rates only for Hispanic populations in census tracts in the highest redline category, thus suggesting potential spillover effects of Medicaid expansion in this population.
Our analysis examining the effect of Medicaid expansion on uninsurance rates among nonelderly adult populations in metropolitan core communities subject to residential redlining found that the effect of expansion differed across redline categories and was greatest in redline category 4 (the highest level of redlining). This finding provides evidence of a major health policy’s differential impact by exposure to systematic segregation, a form of structural racism. Conversely, we found no differences in Medicaid expansion’s impact on uninsurance rates by race or ethnicity within each redline category.
Although measuring disaggregated health outcomes remains critical for advancing health equity, our findings support the growing evidence that health services and policy research focused solely on individual-level measures of disparities may be missing important contextual or community-level measures related to multiple forms of racism.40 Our findings also demonstrate that incorporating both individual and contextual factors into research can yield deeper insights than research considering individual or contextual factors only. Medicaid expansion had its greatest impact on uninsurance rates in the highest redline category. Although we found no statistically significant differences by race and ethnicity within each redline category, our joint tests of effects found significant differences across redline categories in the aggregate and in the non-Hispanic Black nonelderly population. To inform policies aimed at reducing effects of structural racism, future research should explore underlying reasons for the differential impact of Medicaid expansion across redline categories.41
Despite the impact of the ACA’s Medicaid expansion on insurance coverage, health care use, and self-reported health outcomes, many state policy makers continue to oppose expansion.10–12,30,31,42 By reducing uninsurance rates in the most heavily redlined census tracts, Medicaid expansion may have helped reduce the burden of historical racial segregation. This benefit has not yet been realized in the most heavily redlined communities in nonexpansion states.43–45 The gap is especially detrimental for the pursuit of health equity in these states.46
Among the thirty-five states in our analysis, seven have yet to expand Medicaid: Alabama, Florida, Georgia, Mississippi, South Carolina, Tennessee, and Texas.30 Our study found that in nonexpansion states, after the ACA’s enactment, average uninsurance rates in the highest-redlined tracts were 15.7 percent for non-Hispanic Black, 13.6 percent for non-Hispanic White, and 23.4 percent for Hispanic nonelderly adults. Not expanding Medicaid in these states has prevented millions of low-income adults from accessing health insurance.47 On the basis of our findings, we predict that the bulk of these potential beneficiaries reside in historically redlined communities.
States that expanded Medicaid pursuant to the ACA reduced uninsurance rates and narrowed racial and ethnic disparities in health insurance coverage. We are among the first to use digitized historical HOLC data to estimate the effect of Medicaid expansion on uninsurance rates by exposure to structural racism as measured by historically redlined census-tract categories. We found that Medicaid expansion had its greatest impact on lowering uninsurance rates for non-Hispanic Black, non-Hispanic White, and Hispanic adults in communities that experienced the most severe redlining. Within each redline category, however, we found no statistical evidence that Medicaid expansion’s impact on uninsurance rates significantly differed by race or ethnicity. By reducing uninsurance rates in the most heavily redlined census tracts, Medicaid expansion may have helped reduce some of the negative consequences of structural racism and racial segregation. Our results add to the evidence that structural racism, not race, is the cause of health insurance disparities. When implementing and evaluating health policy reforms, policy makers should address not only individual-level factors but also contextual factors such as structural racism.