Abstract
Context: States have experimented with the income eligibility threshold between Medicaid coverage and access to subsidized Marketplace plans in an effort to increase coverage for low-income adults while meeting other state priorities, particularly a balanced budget. In 2014, Wisconsin opted against adopting an ACA Medicaid expansion, instead setting the Medicaid eligibility threshold at 100% of the poverty level—a state-funded partial expansion. Childless adults gained new eligibility, while parents and caregivers with incomes between 101–200% of poverty lost existing eligibility.
Methods: We used Wisconsin's all-payer claims database to assess health insurance gains, losses, and transitions among low-income adults affected by this partial expansion.
Findings: We found that less than one third of adults who lost Medicaid eligibility definitely took up commercial coverage, and many returned to Medicaid. Among those newly eligible for Medicaid, there was little evidence of crowd-out. Both groups experienced limited continuity of coverage. Overall, new Medicaid enrollment of childless adults was offset by coverage losses among parents and caregivers, rendering Wisconsin's overall coverage gains similar to nonexpansion states.
Conclusions: Wisconsin's experience demonstrates the difficulty in relying on the Marketplace to cover the near poor and suggests that full Medicaid expansion more effectively increases coverage.
Since Medicaid's inception, states have made choices about the program's income eligibility thresholds—choices that reflect considerations about health care access, the role of state government, state fiscal policy, and other state priorities. Before the Affordable Care Act (ACA), Medicaid coverage remained quite limited in most states, leaving many low-income adults uninsured (KFF 2020a). The ACA had aimed to create a uniform eligibility standard up to 138% of the federal poverty level (FPL), using Medicaid expansion as one of the act's principal mechanisms to expand coverage. Several states, however, continued to pursue alternative coverage strategies, particularly focused on where Medicaid interacts with private Marketplace-based coverage. Wisconsin—the only partial Medicaid expansion actually implemented since the ACA—set income eligibility at 100% FPL and directed near-poor adults to obtain subsidized private coverage. In this study, we assess the health insurance gains, losses, and transitions among low-income adults affected by partial Medicaid expansion.
Under the ACA, 38 states and the District of Columbia expanded Medicaid coverage for adults with incomes of up to 138% FPL, incentivized through enhanced federal matching rates (KFF 2020b). Among nonexpansion states, Medicaid eligibility for adults varies and is often limited to well below the federal poverty level (KFF 2020a). ACA expansions have increased and sustained health insurance coverage among low-income adults—i.e., those with incomes of up to 138% FPL—relative to the experience of nonexpansion states (Garfield, Orgera, and Damico 2020; Haley et al. 2018).
ACA implementation has been subject to various judicial, executive, and legislative actions. Accordingly, implementation has evolved over the years and has varied among states, as has its impact (Nikpay, Pungarcher, and Frakt 2020). A wide range of benefits are associated with Medicaid expansion, including improvements in access to care, financial security, and some measures of health status/outcomes (Guth, Garfield, and Rudowitz 2020). Policy makers often focus on potential state-level economic benefits and the effect on providers (Levy et al. 2020). Medicaid expansion also offers potential to strengthen private-sector risk pools by removing the lower-income and likely less healthy population (Sen and DeLeire 2018).
Nonetheless, Medicaid expansion does not align with all states' priorities. From the start, some state leaders expressed concerns about the ongoing growth of program enrollment and the requirements for state match (Sommers and Epstein 2013). Governors expressed skepticism about federal commitment to sustaining the enhanced matching rate and concerns about exposing state budgets to substantial new obligations. For example, the concerns voiced in 2014 by Scott Walker, then governor of Wisconsin, rested on broader principles regarding the proper roles of the public and private sectors: “More people depending on government-run Medicaid should not be our end goal. People who are not living in poverty should be allowed to purchase a plan of their own choosing in the private market and they should remain in the private sector for the health insurance coverage” (WOCI and WDHS 2014).
Policy leaders embraced the longstanding message from the provider sector that Medicaid payments fall short of costs and warned of growth in a “hidden tax” on employers, individuals who have private health care plans (WHA 2017), and state budgets paying hospitals for a disproportionate share of undercompensated care (Wisconsin Health News2020). Augmenting these concerns, commentators opposing Medicaid expansion offered a view of Medicaid as fundamentally inefficient, ineffective, and harmful (Friedsam 2014).
State decisions about Medicaid income eligibility thresholds occur in the context of the ACA's provisions to expand private-sector coverage. In all states, consumers with family incomes of 100–400% FPL may be eligible for federal subsidies to purchase private coverage through the ACA Marketplace. Despite the overlap in Medicaid expansion and Marketplace subsidy income parameters, consumers may not qualify for both Medicaid and Marketplace subsidies. Consumers in nonexpansion states can access tax credits and cost-sharing assistance at 100% FPL. Consumers in expansion states may only access such assistance at 138% FPL.
Persons living near poverty experience frequent changes in income, family status, and other circumstances that place them at high risk of transitioning between different coverage types, a phenomenon sometimes called “churn” (Sommers et al. 2016). Continuity of health care among individuals at this income threshold depends on take-up of Marketplace coverage and regularly managing eligibility transitions between Medicaid and Marketplace or other forms of private insurance when programmatic eligibility changes occur. Medicaid and Marketplace plans differ substantially in their benefit structure, provider networks, and cost exposure for enrolled members. Some studies have suggested that substitution of Marketplace coverage for Medicaid eligibility could increase uninsurance and out-of-pocket expenses (Allen et al. 2021; Blavin et al. 2018) or result in greater access to care but less affordability (Selden et al. 2017).
States have sought to experiment with and leverage this overlapping eligibility range (Pearson, Carpenter, and Sloan 2017). Working with states, the federal government has used comprehensive waivers to explore alternative approaches to Medicaid and ACA policy (Gusmano and Thompson 2020). Some ACA Medicaid expansions have occurred under waiver authority, enrolling the 100–138% FPL group in “private option” Marketplace-based Medicaid plans, rather than as standard ACA-model Medicaid expansions (Antonisse and Rudowitz 2019; Freedman, Richardson, and Simon 2018). States have also pursued partial Medicaid expansions, proposing to obtain the ACA's enhanced federal match to raise the Medicaid income eligibility level but not to the full 138% FPL. In 2018, Idaho, Georgia, and Utah requested waiver authority for Medicaid expansion up to 100% FPL in lieu of standard ACA Medicaid expansion. Arkansas and Massachusetts, which previously adopted very different forms of an ACA Medicaid expansion, proposed to scale back Medicaid income eligibility from 138% FPL to 100% FPL while retaining the enhanced federal match (Rudowitz and Musumeci 2019). In 2019, CMS rejected these state requests to access ACA enhanced Medicaid matching funds for such partial expansion models, as it had under the Obama administration (CMS 2019). The possibility of partial expansions with enhanced federal match is, for now, unavailable.
Nonetheless, major stakeholders—including Centene, the largest insurer on the ACA exchanges and a major Medicaid managed care provider—continue discussions at the federal level about changes to the ACA income thresholds (Liss 2019). Following the Supreme Court's Texas v. United States decision, which removed a major legal threat to the ACA, future policy depends greatly on the waiver-related and regulatory decisions of the new administration (Keith 2020; Rosenbaum 2021). New routes to coverage may emerge for the near-poor population. Some states show interest in Medicaid buy-in plans as something of a “public option” model, and such plans themselves vary substantially in their reliance on the private sector (Sandoe and Golberstein 2020). The Biden administration, in its campaign, proposed a new premium-free public option (Biden Harris n.d.), and the American Rescue Plan Act of 2021 provides even stronger incentives for states to expand Medicaid. States will continue to consider the relative merits of various coverage mechanisms, and setting Medicaid income eligibility thresholds below 138% FPL remains a pressing policy consideration.
Lower Medicaid income eligibility levels leave the near poor with the option to enroll in partially subsidized commercial coverage. The degree to which this population takes up and sustains insurance coverage has significant fiscal impact. Recent evidence shows expansion states experienced substantially higher growth in Medicaid spending than nonexpansion states between 2013 and 2018—growth dependent on and subsidized almost entirely by federal funding (Gruber and Sommers 2020). If Marketplace plans can attract, enroll, and retain individuals near poverty, they could avert the need for state budget share of full Medicaid expansion. The state would forgo federal revenue, but federal Marketplace subsidies would remit to private health plans for the near-poor population and still bring substantial federal resources to the state. However, if the near poor are more likely to be uninsured when subsidized coverage is available through the Marketplace as opposed to Medicaid, several costs ensue: the near-poor population incurs ongoing uncompensated care costs for providers while the state forgoes available federal Medicaid matching funds, and federal subsidies do not materialize for the private sector.
Evidence to date about partial Medicaid expansion generally relies on extrapolations from eligibility contractions at higher income levels and on cross-state comparisons (Schubel 2019). The “private option” mechanism adopted in Arkansas and several other states provided Medicaid eligibility up to 138% FPL and a fully subsidized Medicaid package, without commercial premiums and cost sharing (Maylone and Sommers 2017). Arkansas demonstrated minimal differences in access to care compared to another state's standard Medicaid expansion (Sommers, Blendon, and Orav 2016). However, these comparisons do not demonstrate how this near-poor population fares when not eligible for a Medicaid-like product. Comparison between expansion and nonexpansion states suggests the substitution of Marketplace for Medicaid eligibility could lower coverage rates.
We consider this question directly by evaluating the net coverage effects of the state of Wisconsin's partial Medicaid expansion. In 2014, the state of Wisconsin used a section 1115 waiver to expand Medicaid eligibility to nonelderly, nondisabled adults without dependent children (“childless adults”) up to 100% FPL while eliminating income eligibility for adults at 200–100% FPL, the vast majority of whom were parents of dependent children. These changes rendered adults with incomes of up to 100% FPL income-eligible for Medicaid, while those over 100% FPL were directed to subsidized ACA Marketplace coverage. This waiver, funded at the state's standard matching rate, was unique nationally, gaining Wisconsin designation as the only “nonexpansion” state without a coverage gap between Medicaid and Marketplace subsidy eligibility (Garfield, Orgera, and Damico 2020).
Wisconsin's Medicaid model—setting adult eligibility at 100% FPL, along with a rollback of existing Medicaid coverage for adults above 100% FPL—offers a unique empirical opportunity to examine take-up and continuity of coverage for near-poor adults. We use enrollment files from the state all-payer claims database from 2013 to 2016 to show the health insurance gains, losses, and transitions for individuals most likely to have been directly affected. To put these changes in the context of the broader shifts in coverage during this time, we use national survey data to compare the health insurance coverage of Wisconsin residents to coverage in states that did and did not expand Medicaid. The Wisconsin experience can inform state and federal debates about how to best align program eligibility, and it illustrates how low-income families navigate Medicaid and private insurance in the post-ACA context.
Data and Methods
We used two data sources for this study: Wisconsin's all-payer claims database (published by the Wisconsin Health Information Organization, or WHIO) and the American Community Survey (ACS) (Ruggles et al. 2020; WHIO 2020). We followed the coverage experience of Medicaid and commercial insurance members from April 2013 to December 2016 by using the enrollment file from Wisconsin's all-payer claims database, a data source not typically used to study Medicaid enrollment policy (Dworsky 2017; Gordon 2019). These data allowed us to track health insurance enrollment and its duration across insurers and insurance types, regardless of whether members actually incurred claims. The enrollment file contains information on one type of health insurance per person per month, and the individual's age and sex. Type of health insurance is defined by payer, including commercial, Medicaid, Medicare, and dual Medicaid/Medicare.
The study population comprises two mutually exclusive cohorts. The first cohort is Medicaid beneficiaries who experienced the April 1, 2014, eligibility loss. We defined this group as individuals who were enrolled in Medicaid for at least the three months prior to the policy change (January–March 2014) but were not enrolled in Medicaid for at least three months beginning in April 2014. We require at least three months of nonenrollment because renewals and redeterminations can sometimes cause temporary coverage disruptions, and we wanted to ensure we were identifying those who lost eligibility.
Similarly, we defined the group who gained Medicaid eligibility on April 1, 2014, as individuals who were not enrolled in Medicaid for at least the three months prior to the policy change but were enrolled in Medicaid for at least three months beginning April 2014. The cohort definition of three months was intended to balance the risk of over- and underidentifying those most likely to be facing the eligibility changes. In table 3 in the online appendix, we present alternative definitions of the cohorts, and we show that the study's conclusions are unlikely to be particularly sensitive to such alternative cohort definitions.
Since the policy changes were announced in advance, anticipatory behavior could complicate interpretation of the results for the cohort losing Medicaid if those who were most likely to gain private coverage left Medicaid earlier than April 2014; thus, our analysis is focused on those who remained enrolled through April 2014. Medicaid-eligible individuals are not eligible for Marketplace subsidies, so it would be against the rules for someone to enroll in a subsidized plan until April if their Medicaid eligibility was ongoing through March, although it is unclear to what degree these criteria were checked and enforced during this time. To assess the possibility of anticipatory behavior, we examined patterns of commercial enrollment among cohorts defined by their Medicaid enrollment status at different time points (these analyses are further described in the online appendix and are displayed in appendix figures 1 and 2). Based on these analyses, we concluded that, if anything, those who stayed until April were more likely to gain commercial coverage and that anticipatory behavior did not seem to occur to a large degree.
We then used 2009–2018 ACS data to provide an overview of how coverage gains from Wisconsin's partial expansion compared to states implementing full expansions and to those not expanding. The ACS is an annual cross-sectional national household survey published by the US Census Bureau that collects detailed housing, health insurance, and population characteristics. We chose study states based on geographic proximity to Wisconsin and the timing of their Medicaid expansions, i.e., their expansions occurred in early 2014 or not at all during this time. Expansion states are Minnesota, Michigan, and Illinois, and nonexpansion states are Missouri, Kansas, Nebraska, South Dakota, North Carolina, and Tennessee. In the online appendix, we include a similar analysis with all states. The study population for this analysis includes noninstitutionalized adults ages 19–64.
Analyses
From the all-payer claims enrollment file, we created a balanced panel identifying monthly insurance coverage by payer for individuals ages 19–64. We constructed monthly binary measures of insurance status: Medicaid, commercial coverage, uninsured/unknown, or other (including Medicare). We used the panel to compare the size and timing of the overall expansion of commercial coverage to the expansion of Medicaid coverage during this period.
For the two study cohorts that lost and gained Medicaid eligibility in April 2014, we benchmarked the number of individuals to numbers published by the state to assess the plausibility of our cohort definitions. We described the changes in their enrollment in health insurance throughout the study period. We also showed the cohorts' continuity of coverage during this time (commercial for the Medicaid disenrollees; Medicaid for the new Medicaid enrollees). We examined coverage continuity directly, since disruptions to health care can occur across coverage transitions. Maintained enrollment in commercial coverage requires ongoing payment of premiums, and maintaining enrollment in Medicaid requires keeping up with eligibility paperwork (and reflects ongoing eligibility).
We estimated regressions to describe the differences in mean insurance rates before and after 2014 for selected expansion and nonexpansion states relative to Wisconsin, using the ACS data. We performed the analyses for the entire study population and separately for parents and childless adults; standard errors are clustered by state. We also estimated a version of these regressions that controlled for demographics (indicators for sex; age groups defined as 18–34, 35–54, and 55–64; marital status; race; Hispanic ethnicity; and continuous family size) and state and year fixed effects. All ACS estimates were produced using person weights.
Limitations
The use of all-payer claims data is appealing for studying health insurance coverage transitions, since Medicaid administrative data cannot generally be linked to private insurance data. The data have some limitations, which shaped our choices about cohorts and variable definitions. We discuss these in detail in the online appendix. The Wisconsin all-payer claims database relies on voluntary carrier participation, which results in incomplete capture of the insured population in Wisconsin. The WHIO does not include Medicare fee-for-service data; the data vendor claims to represent two thirds of insured lives in Wisconsin. Representation of working-age, full-benefit Medicaid beneficiaries was likely to be nearly complete during the study period. Throughout the study period, the state of Wisconsin contributed the Medicaid fee-for-service data, and all of the Medicaid managed care plans serving nondisabled adults and children contributed their data to the WHIO.
The largest source of missing private insurance data in the WHIO is self-insured employer-sponsored insurance. We provide some comparisons of the WHIO data to the ACS in the online appendix to characterize how this exclusion may affect our study sample. Within the non-self-insured private market, the set of carriers that participated in WHIO was constant throughout the study period. These carriers consistently accounted for at least 95% of the market share as a percentage of net premiums earned (WOCI 2014, 2015, 2016). Within the individual private market more specifically, carriers that did not participate in the WHIO accounted for 23% of enrollment in the individual market in 2014, including individual plans offered on and off the federal Marketplace (WOCI 2021). Because of these two sources of unobserved enrollment in the private market—self-insured plans and individual market plans that did not participate in the WHIO—our results for private insurance should be regarded as a lower bound, and our results for uninsurance should be regarded as an upper bound. After presenting the results below, we estimate and discuss a point estimate and upper bound for commercial coverage that accounts for the lack of self-insured coverage and carriers in the individual market. Finally, we are not able to separately identify Marketplace plans from general commercial coverage in the data during this time period.
Results
Wisconsin experienced substantial expansion of insurance coverage during this time, based on monthly all-payer claims data (figure 1).
Total commercially insured lives ages 19–64 in the data were approximately 1.23 million in April 2013 and increased to a high of 1.35 million two years later, leveling off at just under 1.34 million by the end of the study period. There were substantial increases in enrollment in October 2013 and January 2014, likely representing open enrollment in private coverage, as well as in April 2014, when the Medicaid eligibility change occurred. Total Medicaid enrollment in April 2013 was approximately 351,000 individuals and declined throughout 2013 and early 2014 to a low of 320,000, before increasing to a high of 426,000 in March 2015 and ending at around 387,000. The net increase in Medicaid enrollment from its lowest to highest point during this period was 85% of the total increase in private insurance during this time, although private coverage increases were more sustained.
Our method of identifying those who lost eligibility in the all-payer claims data yielded 44,157 unique individuals who satisfied the inclusion criteria. The net change in parental enrollment based on aggregate state administrative data was 44,089, which is very similar but slightly smaller, as might be expected since some childless adults also lost eligibility (WDHS 2020). Our method of identifying those who gained Medicaid eligibility yielded 82,788 individuals, similar to the net change in childless adult enrollment based on aggregate state administrative data (82,217), although it was slightly larger since there was some new parental enrollment. Table 1 summarizes the available demographic data (age and sex) of the study population. The two cohorts had a similar fraction of individuals ages 19–34 (44% in the cohort that lost Medicaid; 43% in the cohort that gained Medicaid), but the cohort that lost Medicaid was significantly younger on average, with just 7% ages 55–64, compared to 17% of the cohort that gained Medicaid. The cohort that lost Medicaid was also more likely to be female, at 62%, compared to 45% of the cohort that gained Medicaid. This is consistent with expectations about the demographics of people eligible for Medicaid because of parent/caretaker status versus childless adult status.
Table 1 also summarizes average health insurance enrollment in the data before and after the April 2014 eligibility change. This information is provided in more detail in figure 2, which shows the trends in health insurance coverage over time for the two cohorts during the study period.
Figure 2a shows that 10,528 of the cohort of former Medicaid enrollees had definitely obtained commercial coverage immediately following the April 2014 disenrollment, increasing to a peak of 13,861 (31%) within two months, while 67% were uninsured or unknown; the remaining fraction was classified as either Medicare or other. At six months, only 60% were uninsured or unknown, slightly less than the initial proportion, as 9% had returned to Medicaid. The fraction who returned to Medicaid continued to increase steadily, reaching 19% by the end of the study period. Meanwhile, 31% remained definitely enrolled in private coverage at six months, with private coverage eroding across the study period to 25% by the end of 24 months.
For those who newly gained Medicaid coverage in April 2014 (figure 2b), Medicaid enrollment remained high throughout 2014, with more than 85% of the cohort still covered in December 2014. Very few (4,600 or 5.6%) definitely had private insurance coverage 12 months prior, which is fewer than the number who previously had Medicaid coverage. There is a slight downward trend in observed commercial coverage from April 2013 to March 2014, when commercial coverage reaches a low of 2.4%. Enrollment in Medicaid was not particularly sustained overall for this cohort, with less than half remaining insured by Medicaid two years later, and the difference was not made up by observed enrollment in private coverage.
To help account for potentially missing commercial coverage at self-insured firms or through individual market carriers that did not participate in the WHIO, we performed a back-of-the-envelope calculation that made several adjustments to the observed estimate. Broadly, we adjusted for potential self-insured offers and estimated the fraction of the Medicaid-losing cohort that may have had individual coverage through a carrier that did not participate in the WHIO. Together these adjustments yielded plausible upper bounds and alternative point estimates for private coverage enrollment in the sample. A full description of the methods is presented in the online appendix.
In total, we calculated that a maximum of about 16% of the cohorts could have been eligible for an offer at a self-insured firm that we did not observe. After making assumptions about take-up to generate point estimates, accounting for potentially missing self-insured and Marketplace-insured people suggests that about 42% of those losing Medicaid may have been privately insured in June 2014. For the cohort gaining Medicaid coverage, the point estimate suggests that a total of 8% may have had private coverage at the highest point prior to the Medicaid eligibility expansion. Alternate assumptions about take-up rates can be made; if one instead assumes that all those who newly gained Medicaid coverage who are estimated to have an offer were taking it up and switched to Medicaid when eligible, the upper bound on private coverage would be 22%. For those losing Medicaid coverage, if there were no change in the underlying likelihood of an offer, it would mean that an upper bound of 54% at the highest point (June 2014) could have had commercial coverage if every one of those estimated to have an offer suddenly took it up upon losing Medicaid eligibility.
Figures 3a and 3b show months of continuous coverage for the two cohorts. Figure 3a compares continuity in any commercial plan observed in the data relative to the initial plan and any insurance type for the cohort of Medicaid disenrollees, while Figure 3b compares continuity for the new Medicaid enrollee cohort across Medicaid and any insurance type. The “continuous insurance” category combines any type of full-benefit coverage (commercial, Medicaid, and other), so those who left commercial insurance and did not immediately enroll in Medicaid are presumed to be at least temporarily uninsured/unknown. The “continuous commercial” category reflects enrollment in any type of full-benefit commercial plan, and “continuous plan” reflects continuous enrollment in the same commercial plan type.
Of those who enrolled in commercial coverage after losing Medicaid eligibility, 77% were still insured the following January, 67% in commercial coverage and 63% in the same commercial plan. The following year, 60% had been continuously insured since April 2014, with 46% remaining commercially insured and 41% in their initial plan. Sharp declines in coverage occurred in January of each year, when Marketplace plan renewal was effective. Figure 3b shows that in January 2015, 77% of the newly covered had maintained continuous coverage, 75% through Medicaid; and one year later, only 45% had remained continuously covered since April 2014, 42% by Medicaid. Coverage dropped by almost 20 percentage points in the newly covered group during March–June 2015, when they would have undergone Medicaid eligibility renewal.
Table 2 summarizes how health insurance coverage changed in Wisconsin relative to the groups of expansion and nonexpansion states based on the ACS data. The table shows the average insurance rate by type in the pre-2014 and post-2014 period, as well as the simple relative difference, defined as the difference in the average rate after 2014 relative to before 2014 in Wisconsin subtracted from the difference before vs. after in the state type; and the adjusted relative difference, which is the same parameter estimated in a regression that controls for demographics and state and year fixed effects. The relative differences are presented in the table as regression coefficients along with their corresponding standard errors and can be translated into percentage points; for example, −0.031 becomes −3.1 percentage points.
Wisconsin's overall uninsured rate was lower than both comparison expansion states and nonexpansion states throughout our study period. Overall, the probability of being uninsured declined in Wisconsin by 4.9 percentage points, similar to the decline in comparison nonexpansion states (5.7 percentage points), but significantly different from the decline in the comparison expansion states, which was 8.0 percentage points. However, these patterns are different when considered by family status. There was a small decline in uninsurance for parents in Wisconsin (1.1 percentage points), while larger declines in uninsurance occurred for parents in both expansion (5.0 percentage points) and nonexpansion (4.9 percentage points) comparison states. In contrast, childless adults in Wisconsin saw a decline in uninsurance that was nearly seven times larger than the decline in uninsurance among parents (7.0 percentage points), though not as large as the decline in expansion states (9.6 percentage points).
The source of the differences in insurance coverage gains is apparent by coverage type. Prior to 2014, parents in Wisconsin were more likely to be covered by Medicaid than parents in the average expansion state, and childless adults were covered by Medicaid at similar rates. After the policy change, parents in Wisconsin saw a decline in Medicaid coverage of nearly 4 percentage points, while childless adults in Wisconsin saw gains in Medicaid coverage (3.2 percentage points). The coverage losses for Wisconsin parents meant that both expansion and nonexpansion states had greater relative gains in parental Medicaid coverage, and for childless adults in Wisconsin the gains in Medicaid coverage were greater than for childless adults in nonexpansion states (1.5 percentage points) but less than in expansion states (5.9 percentage points). As a result, Wisconsin saw a small (0.6 percentage points) overall increase in Medicaid coverage, which was not statistically different from nonexpansion states but was more than 4 percentage points smaller than the gain in expansion states.
In private coverage, Wisconsin initially had lower rates of privately purchased insurance and higher rates of employer-sponsored coverage than both expansion and nonexpansion states. The overall increase in privately purchased insurance in Wisconsin was similar to nonexpansion states for both parents and childless adults and was relatively larger than in expansion states for parents. The differences in coverage gains for employer-sponsored private coverage are similar. The results in table 2 are robust to the full set of expansion and nonexpansion states, as shown in appendix table 3, as well as to the inclusion of group-level time trends from before the study period (results available on request).
In summary, coverage patterns for childless adults show a mix of gains in Medicaid coverage that is between that of expansion and nonexpansion states as well as private-purchase and employer-coverage gains most similar to nonexpansion states. In addition, coverage patterns for parents are quite different from both expansion and nonexpansion states.
Discussion
This study evaluated the effects of a partial Medicaid expansion on take-up and retention of Medicaid and private coverage among near-poor adults, and it compared insurance coverage changes associated with a partial Medicaid expansion relative to full-expansion and nonexpansion states. There are four key findings. First, only one third of those who lost Medicaid coverage coincident with the partial expansion definitely successfully obtained private coverage during the two-year follow up period. Second, the degree to which people retained newly acquired coverage was limited, with noteworthy interruptions in coverage continuity occurring around the time of renewals, whether for Medicaid or for Marketplace. Third, there was little evidence of crowd-out of private coverage among adults who newly enrolled in Medicaid following the partial expansion. Finally, the magnitude of the decline in the overall rate of uninsurance in Wisconsin was most similar to the decline observed in comparison nonexpansion states; however, coverage changes were differently distributed across population subgroups and payer types.
The degree to which near-poor adults move successfully from Medicaid to private coverage has significant policy implications. In Wisconsin, out of the 44,000 adults who lost eligibility in April 2014, approximately 14,000 definitely gained commercial coverage within two months, and more than 10,000 were covered by some form of private insurance two years later. However, the availability of subsidized Marketplace coverage did not fully substitute for the lost Medicaid coverage. Former beneficiaries reenrolled in Medicaid such that, after two years, 19% were enrolled in Medicaid while 55% could not be shown to be enrolled in public or private coverage. One explanation for the imperfect take-up of subsidized Marketplace insurance within the population that lost Medicaid is cost sharing (Hill 2015). Individuals with incomes of 100–138% FPL in states that did not expand Medicaid up to 138% FPL face higher cost-sharing in Marketplace plans than what Medicaid expansion states require of persons in this income group. Out-of-pocket costs can have a substantial effect on take-up of coverage by lower-income individuals in Medicaid and Marketplace plans (Artiga, Ubri, and Zur 2017; Cliff et al. 2021; Dague 2014). Even with generous subsidies available, premiums can deter low-income potential enrollees.
Medicaid expansions, intended to bring uninsured people into coverage, may also crowd out enrollment in private coverage—a long-standing concern (Gruber and Simon 2008). In this study's population of near-poor adults, we found little evidence of crowd-out. Of the 83,000 adults who newly gained Medicaid coverage in April 2014, only about 2,000 came directly from having definite prior commercial coverage, and fewer than 5,000 had definite private coverage one year prior. Since the vast majority of the newly Medicaid-eligible were largely uninsured, expansions in this income range are likely well-targeted toward decreasing the uninsured rate. Although these results are significantly smaller than those found by some of the prior cross-sectional comparisons of pre-ACA State Children's Health Insurance Plan expansions studied in Gruber and Simon (2008), they are consistent with other evidence from the ACA era showing stable employer offers (Abraham et al. 2016; Frean et al. 2017), and with prior work specific to Wisconsin (Dague et al. 2014) showing lower rates of private coverage at the time of Medicaid enrollment for lower-income families. Our findings suggest that crowd-out of private coverage is unlikely to be a major concern for expansions of Medicaid up to the poverty level.
Coverage continuity may be a problem even if eligibility is available. We found major drops in continuity at renewal dates. For example, the fraction of Medicaid disenrollees maintaining continuous private coverage dropped 14 percentage points from December 2014 to January 2015, and by 11 percentage points for maintaining any coverage. For new Medicaid enrollees, the fraction maintaining continuous Medicaid coverage dropped during the renewal period by more than 20 percentage points, and the drop in the fraction maintaining any coverage was nearly as large. Coverage continuity has long been a concern for maintaining prescription adherence and other types of chronic care (Goldman and Sommers 2020). Although there is evidence that coverage continuity decreased in expansion states, the lack of continuity we found shows that the problem remains even with seamless eligibility thresholds.
Wisconsin experienced a net decline in uninsurance comparable to nonexpansion states following the implementation of the ACA in 2014, including the introduction of the Marketplace and the state's partial Medicaid expansion. However, the distribution of gains in coverage differed from both comparison expansion and nonexpansion states. In aggregate, new Medicaid coverage for childless adults came at the expense of coverage for previously eligible adults with incomes of 100–200% FPL, mainly parents of dependent children, many of whom did not successfully gain commercial coverage following disenrollment from Medicaid. Adults without dependent children or disabilities have long been largely ineligible for a variety of safety net programs, so this change represents a significant policy shift.
Our findings have some limitations. We did not observe income, eligibility categories, or number of dependent children in the all-payer claims data, so we identified our cohorts by inferring exposure based on coverage patterns. We cannot differentiate among commercial coverage types and so are unable to specify whether commercial take-up is through Marketplace or other sources, such as increased take-up of employer-sponsored insurance. Finally, since carrier participation is voluntary in the all-payer claims, we are likely missing data from self-insured firms and from some carriers that participate in the individual market. We adjusted for this based on estimates from other sources to generate an upper bound on commercial coverage, and we found that at most 22% of the cohort that gained Medicaid likely could have been commercially covered beforehand, and at most 54% of the cohort that lost Medicaid likely could have gained commercial coverage afterward, with point estimates at 8% and 42%, respectively.
Wisconsin's experience offers insights about the relative gains and losses that may occur if federal policy changes to allow a decrease in the ACA Medicaid expansion level to 100% FPL. This study's results indicate that among near-poor adults, the associated disenrollment would not be fully offset by increases in take-up of other coverage, so uninsured rates could increase. Even without a state-level coverage gap in eligibility for fully and partially subsidized coverage, many near-poor adults do not successfully attain or maintain coverage. Policies such as the expansion of Marketplace subsidy eligibility to those below the poverty line are not likely to close the gap between Medicaid expansion states and nonexpansion states. Overall, our findings indicate that fully subsidized public coverage like Medicaid is a key component in strategies to decrease uninsurance rates in low-income populations.
Conclusion
The degree of Marketplace take-up among low-income adults will vary among states based on variations in premiums, plan composition, and other state-specific conditions that promote or mitigate ACA marketplace enrollment, including the structure of state Medicaid programs (Burton et al. 2018). Nonetheless, in each state, consumers face minimal cost exposure with Medicaid enrollment relative to Marketplace and other commercial coverage (Beutel, Gunja, and Collins 2016). Wisconsin successfully decreased its uninsurance rate with a partial Medicaid expansion. However, low take-up of commercial coverage by the near poor suggests full Medicaid expansion achieves higher coverage for this low-income group than is achieved by relying on the Marketplace.
Acknowledgments
We gratefully acknowledge the excellent research assistance provided by Nicolas Badaracco and funding for this study from the Robert Wood Johnson Foundation. We also acknowledge the ongoing support for our team's research and evaluation work provided by our partnership with the Wisconsin Department of Health Services Medicaid agency.
