Hostname: page-component-7c8c6479df-8mjnm Total loading time: 0 Render date: 2024-03-28T11:25:08.992Z Has data issue: false hasContentIssue false

The Politics of Welfare Retrenchment: Evidence from Mass Medicaid Disenrollment in Two States

Published online by Cambridge University Press:  19 October 2022

Vladimir Kogan*
Affiliation:
Department of Political Science, Ohio State University, Columbus, OH, USA
*
Rights & Permissions [Opens in a new window]

Abstract

In 2005, Missouri and Tennessee tightened eligibility for their public health insurance programs, resulting in widespread coverage losses. Leveraging county-level variation in subsequent disenrollment, I show that voters in Tennessee punished the incumbent governor for the Medicaid cuts. In Missouri, by contrast, disenrollment had no impact on the subsequent gubernatorial election but did increase support for Democrats in 2006 state legislative elections, possibly due to the strategic entry and exit of candidates. In both states, the loss of Medicaid coverage was associated with lower support for Democratic presidential candidates, although these declines appear part of a longer-term trend that preceded the coverage loss. The results speak to the potential political costs of welfare spending cuts and the electoral consequences of reducing income-targeted social programs.

Type
Original Article
Creative Commons
Creative Common License - CCCreative Common License - BY
This is an Open Access article, distributed under the terms of the Creative Commons Attribution licence (http://creativecommons.org/licenses/by/4.0), which permits unrestricted re-use, distribution and reproduction, provided the original article is properly cited.
Copyright
© The Author(s), 2022. Published by Cambridge University Press and State Politics & Policy Quarterly

Introduction

Shortly after Sen. John McCain cast the pivotal vote to defeat the Republican repeal of the Affordable Care Act in July 2017, two Vox reporters summarized the lessons learned: “Repealing Obamacare is hard,” they wrote, “because taking away benefits is hard.”Footnote 1 Their intuition echoed an argument put forward by Pierson (Reference Pierson1996), who noted that the politics of welfare retrenchment is “typically treacherous, because it imposes tangible losses on concentrated groups of voters in return for diffuse and uncertain gains” (145).

Pierson’s thesis helps explain the historical dearth of successful welfare repeal efforts. The rarity of large-scale cutbacks, however, also means that a key implication of the theory—that welfare cuts lead to a loss of electoral support—remains difficult to test empirically. And although Pierson’s argument focuses primarily on universal, contributory social insurance programs—such as Social Security and Medicare in the US and the National Health Service in England—that are broadly popular with voters, it is not obvious that a similar logic applies in the context of redistributive, means-tested welfare programs serving much smaller and disadvantaged constituencies.Footnote 2 The beneficiaries of these programs are economically and politically marginalized (e.g., Schlozman, Verba, and Brady Reference Schlozman, Verba and Brady2012) and, according to some critics, further demobilized through their interactions with complex and paternalistic welfare bureaucracies (e.g., Bruch, Ferree, and Soss Reference Bruch, Ferree and Soss2010; Michener Reference Michener2018).

This article offers new empirical evidence on the political costs of welfare retrenchment by examining the fallout from mass public health insurance disenrollment events in Missouri and Tennessee. In response to growing concerns about the financial sustainability of their Medicaid programs, which during the period I study primarily served low-income parents and their children,Footnote 3 both states dramatically reduced Medicaid eligibility in 2005.Footnote 4 The changes resulted in the loss of insurance for tens of thousands of low-income, able-bodied adults in each state—a cut affecting about 1 in every 10 existing enrollees. The Tennessee cuts were made via executive order by a Democratic governor, while in Missouri they were approved by the legislature at the behest of a newly elected Republican executive.

Although these eligibility changes were implemented statewide in both jurisdictions, the resulting disenrollment was greater in some geographic areas than others because the affected populations were not uniformly distributed across each state. I leverage the resulting county-level variation in the proportion of the population affected by the loss of public health insurance to implement a difference-in-differences analysis allowing me to isolate the effect of disenrollment on electoral outcomes in both states.

In Tennessee, I find that the incumbent Democratic governor’s share of the major-party vote declined between 1.1 and 1.8 percentage points for every percentage point decline in the share of the local population covered by Medicaid when he ran for reelection two years later. But the cutbacks appear to have had no impact on down-ballot legislative races. In Missouri, by contrast, I find that disenrollment did not influence the subsequent gubernatorial election (during which the incumbent responsible for the cuts chose not to stand for reelection) but did increase support for Democrats in the 2006 state legislative elections. This effect appears to have been driven by the strategic entry and exit of candidates rather than intentional changes in voter behavior. In both states, the loss of Medicaid coverage was associated with lower support for Democratic presidential candidates, although these declines were part of a longer-term trend that preceded the coverage loss.

The effects I document are noteworthy because Medicaid provides an especially difficult case for Pierson’s theory. Due to a combination of disadvantages related to income, race, and poor health, Michener (Reference Michener2018) notes that Medicaid enrollees are “among the most politically vulnerable persons in the country” (10) and argues that most “remain disempowered in the face of polarized pivotal public debates that may literally determine whether they live or die” (11). She finds that (on average) Medicaid recipients vote at lower rates than nonrecipients and that these differences are particularly pronounced immediately after coverage cutbacks, which she argues discourage political participation by highlighting the scarcity of government resources and the capriciousness of policy makers.

By contrast, my results suggest that the threat of electoral retribution can, under certain conditions, serve as an important source of political influence for this otherwise disadvantaged constituency and potentially helps insulate poor voters from the loss of public benefits. On the other hand, the threat of electoral retribution may also keep public officials from expanding welfare programs in the first place out of fear that a benefit expansion adopted during times of budgetary surplus will be difficult to undo if fiscal conditions deteriorate. This is one rationalization offered by officials in some Republican-controlled states to justify opting out of the Affordable Care Act’s Medicaid expansion.

My findings contribute to a growing literature on the political “feedback effects” of Medicaid (e.g., Clinton and Sances Reference Clinton and Sances2018; Haselswerdt Reference Haselswerdt2017; Haselswerdt and Michener Reference Haselswerdt and Michener2019). Much of the published work on Medicaid (and, indeed, other American welfare programs) tends to focus on voter participation, and these recent papers find evidence that gains in Medicaid coverage under the Affordable Care Act (ACA) resulted in higher turnout, likely among beneficiaries.Footnote 5 By contrast, I focus on the consequences of vote choice and, ultimately, election outcomes.Footnote 6 Although Haselswerdt and Michener (Reference Haselswerdt and Michener2019) examine the same Tennessee disenrollment event and focus on vote choice in the gubernatorial election that followed the cuts, they report “ambigious” effects that are sensitive to model specification. I extend their analysis to include more election years and contests for other offices and find much stronger effects in the gubernatorial election. No other research, to my knowledge, has examined the Missouri case.

The results speak to the possible political repercussions of off-and-on-again efforts to roll back portions of the Affordable Care Act. With support from the Trump Administration, a number of Republican-controlled states attempted to impose new work requirements for able-bodied adult Medicaid recipients who benefited from the ACA Medicaid expansion, although these efforts have faced new opposition from President Biden since the 2020 election.Footnote 7 Both the fraction of current beneficiaries that could lose coverage as a result of these requirements and their socioeconomic profiles closely match the number and types of enrollees who were removed from the Medicaid rolls in Tennessee and Missouri, suggesting that these states’ experiences can provide important insights about the electoral consequences that such retrenchment efforts are likely to produce.

Medicaid Disenrollment in Two States

This section provides a brief overview of recent Medicaid history in the two states I examine in my empirical analysis and describes the political context at the time each program was cut. These summaries are based on my review of archives of the major newspapers in each state. In Tennessee, I also rely on a 2003 independent evaluation of the TennCare Medicaid program written by consulting firm McKinsey that was commissioned by a group of health insurers and providers in the state.

Tennessee: Origins and Collapse of TennCare

Tennessee’s decision to dramatically reduce Medicaid eligibility in 2005 was connected to an ill-fated wholesale redesign of the system carried a decade earlier, under Democratic Gov. Ned McWherter. Rising Medicaid costs in the early 1990s led to budget shortfalls that appeared to require either significant tax increases or coverage reductions. However, McWherter instead proposed—and subsequently won legislative and federal approval for—a complete overhaul of the program he thought would reduce costs and consequently free up state resources to expand coverage to more people. The governor’s plan replaced the state’s previous fee-for-service Medicaid program with a new managed-care system called TennCare. Under TennCare, the state would make capitated per-enrollee payments to private health insurers, which would be responsible for negotiating rates with providers and aggressively managing care to reduce costs. These managed-care organizations, rather than the state, would bear the financial risk if state payments fell short of the cost of care provided.

With the new managed-care system projected to significantly decrease costs per enrollee, McWherter’s plan called for the savings to be used to expand eligibility to people who previously did not qualify for Medicaid. The newly eligible included families with incomes up to 200% of the federal poverty level and those who lacked employer-provided insurance but had costly health conditions that made them “uninsurable” in the individual insurance market. Unlike traditional Medicaid beneficiaries, both of these newly insured groups would pay premiums and be responsible for other forms of cost-sharing, levied on a sliding scale linked to income.

By 2003, TennCare had expanded to cover one in four Tennessee residents, but the underlying funding model had unraveled. Early on, efforts by managed-care organizations to reduce costs were impeded by lawsuits from provider and patient advocates that greatly limited their ability to effectively manage care and restricted the scope for cost savings. With state funding insufficient to cover their costs, insurers reduced reimbursement rates for providers, limiting access to specialists. In the early 2000s, several managed-care organizations filed for bankruptcy, forcing the state to step in and accept the full financial risk for the cost of care, effectively dismantling the cost-saving managed-care reforms. Between 1994 and 2005, annual expenditures on the TennCare program increased more than threefold despite flat enrollment.

In 2002, former healthcare executive Phil Bredesen was elected governor as a Democrat in part on a platform of reforming TennCare to stabilize its funding and make the program sustainable. Once in office, Bredesen commissioned outside consultants who revealed that the program was in even worse financial shape than many feared and, absent changes, would bankrupt the state by consuming nearly every new tax revenue dollar over the next half-decade. Initially, Bredesen forged bipartisan consensus behind a reform bill that left eligibility rules unchanged but greatly reduced the generosity of the available benefits for the two expansion populations who did not qualify for traditional Medicaid, essentially creating a two-tiered benefit system. However, these changes were likely to be blocked by the courts and, when new estimates suggested they would not deliver the savings the governor sought anyway, Bredesen announced in December 2004 that he would instead use his executive authority to dissolve TennCare and revert to a traditional Medicaid program, ending coverage for the expansion populations who had gained it under McWherter’s reform.Footnote 8 The initial estimates suggested that nearly half a million beneficiaries would lose their health insurance, but the governor decided to spare most children the cuts would have affected. Ultimately, about 170,000 adults lost their coverage, representing just under 13% of the total TennCare enrollment at the end of 2004.

Although the legislature adopted a replacement program to provide partially subsidized coverage—and budgeted transitional funding for community health centers and other safety net programs to serve the newly uninsured—Bredesen’s decision proved to be controversial and a target for criticism from his Republican opponent, state Sen. Jim Bryson, during the governor’s 2006 reelection campaign. The issue was featured prominently in Bryson’s television advertising and came up repeatedly in the televised debates between the candidates.

Missouri: A New Governor’s Broken Promise

As in Tennessee, Missouri’s 2005 Medicaid cuts followed an expansion implemented during the previous decade. In 1998, the state received federal approval for Democratic Gov. Mel Carnahan’s plan to expand Medicaid coverage to children in families with incomes up to three times the federal poverty limit. Although it received less public attention at the time, the federal waiver also included an “Uninsured Parents Program” that allowed adults transitioning out of welfare to retain their Medicaid coverage, with federal funds paying most of the cost.Footnote 9 In 2000, Carnahan ran for the US Senate and was replaced as governor by fellow Democrat Bob Holden.Footnote 10

Early in his term, Holden faced large budget deficits as the national recession cut state revenues and Medicaid costs kept climbing. Although Holden endorsed limited Medicaid reforms, including trimming optional services such as vision and dental benefits not required by federal law, the Republican-controlled legislature called for greater cost savings. In early 2004, the state House voted to cut the maximum income threshold for Medicaid eligibility for low-income parents from 77% of poverty down to 50%, projected to remove roughly 50,000 low-income adults from the Medicaid rolls. However, the proposal was subsequently defeated in the state Senate.

Holden staked his 2004 reelection bid largely on a platform of protecting Medicaid—calling for a constitutional amendment to make future cuts more difficult. However, he lost the Democratic primary to State Auditor Clare McCaskill, who was subsequently defeated in the fall by Republican Secretary of State Matt Blunt.

During the campaign, Blunt spoke out against the cuts proposed by legislative Republicans. “I am opposed to changing the eligibility requirements until we have done everything we can to ensure we are efficiently spending the Medicaid dollars we have today,” he said at a campaign event several months before the election.Footnote 11 After the November vote, however, he quickly reversed course. In his first State of the State speech delivered the following January, Blunt called for even deeper cuts than legislators had sought the year before—lowering the income threshold for Medicaid eligibility for adult parents to less than 30% of poverty. The governor’s proposal was subsequently adopted by the legislature on largely party-line votes, with the first enrollment cuts taking effect in summer 2005. Approximately 100,000 adult beneficiaries ultimately lost coverage as the cuts were implemented over the next two years, or about 1 in 10 existing Medicaid enrollees.

Polls showed that the Medicaid cuts were unpopular among voters, driving down Blunt’s overall approval. By early 2008, surveys suggested that Blunt would likely lose reelection to his Democratic challenger, Attorney General Jay Nixon, and the governor announced he would not run again. Nixon, who had announced his own candidacy just months after the Medicaid cuts were adopted and framed his campaign as an effort to restore Medicaid coverage, went on to win the 2008 gubernatorial election against Republican Congressman Kenny Hulshof. Hulshof publicly endorsed Blunt’s cuts but proposed restoring insurance for former Medicaid enrollees by providing partial state subsidies to purchase high-deductible private insurance, with much of the fall campaign focused on the issue of healthcare.

Summary

Table 1 provides a brief summary of the two cases, highlighting the similarities as well as the important differences between the two states. In both cases, the enrollment losses represented roughly one-tenth of each state’s existing Medicaid caseload and affected primarily able-bodied, working-age adults. In Tennessee, however, the affected population included relatively higher-income households. There were also important differences both in party control of state government at the time of the enrollment cuts—Democrats in Tennessee, Republicans in Missouri—and whether the legislative or executive branch was the primary policy mover. As I discuss in the conclusion, these key differences may explain the somewhat different political and electoral dynamics that followed.

Table 1. Summary of cases

a The TennCare cuts were announced approximately a week after the 2004 election, when Democrats lost control of the lower house of the legislature but before new members were sworn in.

The table also summarizes the elections covered in the analysis. Although cuts in both states occurred in 2005, the timing of the subsequent elections differed between the two states. Tennessee’s next gubernatorial election took place the following year, while the Missouri gubernatorial contest was held in 2008.Footnote 12 Legislative elections were held in both years in each state, and I include both rounds of contests in the analysis to examine potential over-time dynamics for these races.

Empirical Strategy

Although there are some important differences between the Medicaid eligibility changes implemented in the two states I study, both produced significant loss of coverage for thousands of able-bodied, low-income adults. Because the distribution of the affected populations varied considerably within each states at the time these cuts were implemented, the extent of coverage loss was geographically clustered as well. Figure 1 plots the county-level decrease in Medicaid coverage (measured as a percent of the total population) in each state. In some counties, nearly 1 in 10 residents lost coverage, while other areas saw only minimal impacts on enrollments. I follow Garthwaite, Gross, and Notowidigdo (Reference Garthwaite, Gross and Notowidigdo2014; Reference Garthwaite, Gross and Notowidigdo2018) in taking advantage of these county-level differences as a source of identifying variation to examine how the loss of public health insurance affected elections in each state.

Figure 1. Medicaid enrollment decline by county (as a percent of total population).

Of course, enrollment losses were not randomly allocated to counties. In the Supplementary Material, I present regression estimates that model Medicaid enrollment declines as a function of county demographic characteristics using data from the 2000 Census. In both states, smaller (in terms of population) and economically poorer counties suffered larger coverage cutbacks.Footnote 13 My analytic strategy directly accounts for preexisting differences between counties that likely affected both the magnitude of the enrollment losses and election outcomes.

Intuitively, I examine how the political dynamics in each county evolve over time—comparing voter behavior before the cuts were made to election outcomes afterward—and ask whether these political trends diverged in counties that experienced deeper Medicaid coverage losses. Since the units of analysis are counties, any differences I document need not be driven by the voting behavior of former Medicaid beneficiaries themselves. For example, if hospitals faced larger burdens for uncompensated care in counties harder hit by Medicaid cuts (as Garthwaite, Gross, and Notowidigdo Reference Garthwaite, Gross and Notowidigdo2018 find), any resulting downstream political effects will be captured in my estimates as well. It is also important to emphasize that my approach identifies only local effects—if Blunt’s cuts were uniformly unpopular statewide, regardless of how many people were directly affected locally, and voters everywhere showed their dissatisfaction by voting against Republican candidates, my estimates would not capture this effect.

Specifically, I estimate a difference-in-differences model (Angrist and Pischke Reference Angrist and Pischke2009) using the following OLS specification:

$$ {\mathrm{DemVote}}_{ct}\hskip0.35em =\hskip0.35em {\alpha}_c+\beta \hskip0.35em {\mathrm{EnrollmentDecline}}_c\cdot {\mathrm{Post}}_t+{\delta}_t+{\varepsilon}_{ct}, $$

DemVote $ {}_{ct} $ represents the share of the two-party vote won by Democrats in county $ c $ during election year $ t $ , with separate estimates for each office type. The analysis includes a full set of county ( $ \alpha $ ) and election year ( $ \delta $ ) fixed effects, to account for time-invariant political characteristics of each county and political shocks specific to each election. The coefficient of interest is $ \beta $ , which captures the interaction between Medicaid enrollment decline resulting from the eligibility changes (again, measured as a percent of total county population) and a dummy variable indicating that the election took place after the enrollment cuts had taken effect. The main effects of EnrollmentDecline $ {}_c $ and Post $ {}_t $ are absorbed in the county and election year fixed effects, respectively, so they need not be included separately in the equation. All standard errors are clustered by county to account for serial correlation (Bertrand, Duflo, and Mullainathan Reference Bertrand, Duflo and Mullainathan2004).

The assumption necessary to interpret $ \beta $ as the causal effect of coverage loss is that counties with similar prior levels of Democratic support, but which experienced different intensity of Medicaid disenrollment, would have followed parallel political trends over time in the absence of the eligibility changes. In my preferred specification, I relax this assumption further by including a commute-zone-by-election-year fixed effect, replacing $ {\delta}_t $ with $ {\delta}_{zt} $ , where $ z $ indexes commute zones. Commute zones are defined by the federal government and represent sets of economically integrated adjacent counties that together make up a single labor market. My preferred specification thus allows for neighboring counties to experience common political shocks during each election cycle. In addition, I directly test for parallel trends using an “event study” specification that estimates a separate effect of EnrollmentDecline $ {}_c $ for each election cycle. As an example, the event study specification for Tennessee gubernatorial elections takes the following formFootnote 14:

$$ {\mathrm{DemVote}}_{ct}\hskip0.35em =\hskip0.35em {\alpha}_c+\sum \limits_{j=1990}^{2006}{\pi}_j\hskip0.35em {\mathrm{EnrollmentDecline}}_c\cdot 1\left({\tau}_{ct}=j\right)+{\delta}_{zt}+{\varepsilon}_{ct}. $$

If the trends are indeed parallel, the coefficients on EnrollmentDecline $ {}_c $ should be null for election years that precede the Medicaid cuts ( $ {\pi}_{j<2005} $ ), providing a useful placebo test.

Given the prominent role that governors of both states played in the enactment of the cuts, I begin by examining gubernatorial elections. I obtained county-level election results for all gubernatorial elections in Tennessee for the years 1990–2006 and gubernatorial elections in Missouri from the years 1992–2008 from CQ Press. I also examine county-level presidential election results during this period in both states to explore potential top-of-the-ballot effects. The final set of analyses looks at state legislative election results, which I collected from official returns in each state. Because legislative districts often cross county lines, I tabulated the votes cast in each legislative race for district segments overlapping each county,Footnote 15 and aggregated across all districts in each county, producing a single measure of Democratic vote share at the county level. My sample of legislative contests includes all state House of Representatives elections held during years 2002–8.Footnote 16 I do not include elections prior to 2002 because these were held under a different redistricting plan, with a different pattern of overlap between individual districts and counties.

The analysis of legislative elections presents unique complications. Because all voters chose between the same candidates in the statewide gubernatorial and presidential races, the election year fixed effects account for any candidate-level attributes that may affect vote choice—such as candidate quality and incumbency. In the legislative races, by contrast, voters in different counties confront different candidate choice sets, including potentially consequential differences in incumbency status and whether a given district was contested by both parties. On the one hand, it is important to account for these race-level factors, since they almost certainly affect voting behavior. On the other hand, these race-level dynamics are potentially post-treatment—if a Republican legislator in Missouri chose to retire in 2006 because she expected a tough race due to voter disaffection with local Medicaid cuts or if a strong Democratic challenger emerged to run against her sensing this vulnerability,Footnote 17 controlling for incumbency status and contestation would absorb some of the electoral consequences of Medicaid disenrollment, producing downward bias. Thus, I present the results for legislative races both with and without controls for incumbency status and whether each district was contested by both parties.Footnote 18

Measuring the level of Medicaid coverage loss in each county was straightforward in Tennessee both because the eligibility changes took effect on a known date in the summer of 2005 and also because the state provides quarterly enrollment counts in its historical TennCare annual reports. Thus, I simply calculate the difference in TennCare enrollment between the third quarters of 2004 and 2005 and divide by total county population.

The calculation is more difficult in Missouri for two reasons. First, although some Medicaid eligibility changes took effect immediately in 2005, most beneficiaries were allowed to extend their coverage up to one additional year. Second, the historical county-level Medicaid enrollment numbers I was able to obtain via public records requests included only annual counts, averaged across all months in each fiscal year. So the first reports in which the enrollment cuts would be fully reflected in the annualized data are for fiscal year 2007. Unfortunately, the Medicaid enrollment records provided by the state for this fiscal year proved to be particularly prone to typographical errors—for example, two counties reported sudden, sharp increases in Medicaid enrollment that year, with total Medicaid caseloads several times greater than the total population of each county. For this reason, I use enrollment numbers from fiscal year 2008 as the post-enrollment period to quantify coverage loss at the county level.Footnote 19

Results

Table 2 reports the estimated effect of Medicaid disenrollment on gubernatorial elections, with Panel A focusing on Tennessee and Panel B presenting results from Missouri. In Tennessee, the standard difference-in-differences specification in the first column indicates that every percentage point decline in Medicaid coverage reduced the Democratic share of the two-party vote by 1.1 percentage points. Since my sample includes one post-disenrollment gubernatorial election—the 2006 race featuring Bredesen—the results indicate that counties experiencing greater coverage loss were less supportive of the governor responsible when he stood for reelection. The second column reports my preferred specification, which includes commute-zone-by-year fixed effects to relax the parallel trends assumption, and shows a somewhat larger effect, a 1.8 percentage point decline for every percentage point loss in Medicaid coverage. I discuss the differences between my results and those reported in Haselswerdt and Michener (Reference Haselswerdt and Michener2019) and examine several possible explanations for these differences in the Supplementary Material.

Table 2. Effect of Medicaid disenrollment on democratic gubernatorial vote share

Note. Robust standard errors clustered by county in parentheses.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

By contrast, I find no evidence that voters responded to local Medicaid coverage losses in Missouri. As in Tennessee, this estimate is identified through one (2008) election, the first held after the eligibility changes had taken effect. As I note above, Blunt did not stand for reelection, and so perhaps his absence from the ballot may explain why the Medicaid cuts seemed to matter less in Missouri.

Since my panel includes every election since 1990, I can also estimate separate effects for election years during which the TennCare expansion was in effect (1994–2002). Because historical county-level TennCare enrollment records back to the 1990s are not available, I use the number of people disenrolled in 2005 as a proxy for the size of the expansion population over the entire time period. This specification is presented in Table 3, with separate effects reported for “post-expansion” (1994–2002) and “post-disenrollment” (2006) years. Note that both effects are estimated relative to the 1990 election baseline, the year in my sample that preceded both. Overall, I find suggestive evidence that the TennCare expansion increased support for Democratic gubernatorial candidates, although the relevant coefficients fail to reach statistical significance at conventional levels. However, I continue to find a significant decline in Democratic support in the post-disenrollment election. Since both the “post-expansion” and “post-disenrollment” coefficients are estimated relative to 1990, the quantity of interest is the difference between the two coefficients. Although neither is statistically different from zero, the difference between the two is statistically significant in both models in Table 3, consistent with larger TennCare losses reducing Democratic support in 2006 relative to Bredesen’s performance in each county in 2002.

Table 3. Medicaid disenrollment and democratic gubernatorial vote share in Tennessee

Note. Robust standard errors clustered by county in parentheses.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Figure 2 visually presents the more flexible event-study estimates for each state. Although I find no evidence of immediate impacts after the TennCare expansion went into effect in 1994, the figure shows that counties with larger expansion populations recorded higher Democratic support starting with the 1998 election. The effect remained positive in 2002, but not significant at conventional levels. By contrast, after the TennCare expansion was terminated in 2005, the effect became significantly negative in 2006. Taken at face value, the results suggest that Democrats would have been better off never expanding Medicaid at all, rather than expanding it only temporarily, since the negative effect of ending benefits more than offset the initial positive effect of providing coverage.Footnote 20

Figure 2. Event study results for impact of Medicaid disenrollment on gubernatorial elections.

In Missouri, I find no evidence that either the initial Medicaid expansion in 1998 or the subsequent cuts in 2005 produced any significant change in voting behavior in gubernatorial elections. Recall that no adults actually gained new coverage as a result of the 1998 reforms—some adults were simply allowed to continue receiving Medicaid benefits they would have otherwise lost as they transitioned from welfare to work—so the former result is not surprising, although the lack of any effect in 2008 is perhaps puzzling.

Table 4 reports comparable estimates for the state house elections. Since I observe two post-disenrollment elections in the data, I estimate separate effects for each. The first two columns include no other controls, while the second two columns account for incumbency status and contestation.Footnote 21 Overall, I find mixed evidence that Medicaid losses affected state legislative elections in Tennessee. While the coefficients in the first column show a decrease in Democratic vote share as counties experience higher rates of Medicaid enrollment losses, these estimates are not robust to the inclusion of commute-zone fixed effects. Figure 3 provides some evidence of Democratic over-performance in the 2004 legislative elections in areas that would soon suffer heavier enrollment declines. Since the future of TennCare was a salient consideration during the 2004 election, coming very soon after the release of Bredesen’s last-ditch plan to save the program, one interpretation is that those whose coverage was most at risk hoped a Democratic majority would protect the program from elimination.

Table 4. Medicaid disenrollment and democratic state house vote share

Note. Robust standard errors clustered by county in parentheses.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Figure 3. Event study results for impact of Medicaid disenrollment on state house elections.

The results in Missouri are somewhat different. The first two columns of Panel B do provide evidence that Democratic legislative candidates appeared to win more votes in 2006 (and perhaps 2008) in counties that saw the deepest enrollment cuts. I replicate this result in the event study analysis, reported in Figure 3.

Interestingly, these effects disappear when I add controls for incumbency and whether each race featured candidates from both major parties, presented in the second two columns of Panel B in Table 4. One explanation is that enrollment loss caused Republican incumbents in some districts not to stand for reelection, or caused Democrats to field candidates in some districts that would have otherwise gone uncontested.

Since the enrollment data is measured at the county rather than district level, I cannot model candidate decisions directly as a function of disenrollment. However, Table 5 provides the closest approximation. Rather than examining whether certain candidates run or not, I measure the proportion of votes in a county cast in a district where a Democratic or Republican incumbent was running (Panel A) as well as the proportion of votes cast in a district where a Democratic or Republican candidate faced no major-party opponent (Panel B). These results are consistent with Medicaid enrollment influencing the strategic retirement of incumbents as well as the entry decisions of challengers.

Table 5. Strategic candidate entry and exit in Missouri

Note. Robust standard errors clustered by county in parentheses. Because the unit of analysis is counties, rather than districts, the dependent variable in Panel A is coded as the percent of total major-party votes cast in each county that came from districts where the Democratic (Models 1 and 2) or Republican (Models 3 and 4) incumbents were running. Similarly, for Panel B, the dependent variable is coded as percent of votes cast in districts with an unopposed Democratic (Models 1 and 2) or Republican (Models 3 and 4) candidate on the ballot.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Results in Panel A suggest that Democratic legislative incumbents in Missouri were more likely to stand for reelection both in 2006 and 2008 if they represented counties that experienced heavier Medicaid coverage loss. By contrast, the estimates in the last two columns indicate that Republican incumbents were actually more likely to retire instead of running again in both years if they represented areas harder hit by the Medicaid cuts. Panel B also shows that Republicans were less likely to field candidates in 2006 (but not 2008) in districts that the party had contested in earlier elections if these districts were located in counties that saw greater Medicaid cuts.Footnote 22 However, I find no evidence of strategic contestation among Democratic candidates. Thus, the results suggest that incumbents from both parties and Republican challengers responded strategically to local Medicaid cuts, and these elite decisions may explain the differences in Missouri voting behavior documented in Table 4.

The final set of analyses examines the impact of Medicaid disenrollment on presidential elections. Table 6 presents results when pooling all pre-2005 elections, while Table 7 allows for separate post-expansion and post-disenrollment effects in Tennessee, similar to the gubernatorial results presented above. Interestingly, estimates in both states provide some evidence that heavier Medicaid losses reduced support for Democratic presidential candidates.

Table 6. Effect of Medicaid disenrollment on democratic presidential vote share

Note. Robust standard errors clustered by county in parentheses.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

Table 7. Medicaid disenrollment and democratic presidential vote share in Tennessee

Note. Robust standard errors clustered by county in parentheses.

*** p < 0.01;

** p < 0.05;

* p < 0.1.

As Figure 4 makes clear, however, caution is warranted when interpreting these estimates as the parallel trends assumption does not appear to hold for presidential elections. Indeed, there is evidence that counties with a larger share of residents who eventually lost Medicaid coverage began trending away from the Democratic Party years earlier, starting with 1996 in Tennessee and 2000 in Missouri. There is evidence of an additional pro-Republican shift following disenrollment in 2008 in Tennesee, but no significant changes that year in Missouri.

Figure 4. Event study results for impact of Medicaid disenrollment on presidential elections.

One reason for why areas with more Medicaid-eligible adults in both states became less Democratic in national elections starting in the late 1990s may be the 1996 welfare reform law signed by President Clinton. Recall that the adults who lost their Medicaid coverage in both states were low-income adults—precisely the population most directly affected by the welfare reforms. Indeed, in Missouri, initial Medicaid eligibility for those who lost coverage was mechanically tied to previous welfare participation.

There are two possible reasons for why these voters may have become less likely to vote for national Democrats in the late 1990s. First, previously welfare-eligible adults may have associated their loss of cash benefits with President Clinton and punished the national Democratic Party accordingly. Second, former welfare recipients who transitioned to work likely saw their incomes increase (see Haskins Reference Haskins2016)—although they remained poor enough to qualify for Medicaid in both states—and this income change may have itself produced a more conservative turn in voting behavior. It is worth emphasizing that the observed shifts in national partisanship did not appear to have affected voting in state gubernatorial elections. Indeed, in Tennessee, I previously showed that Democratic support in gubernatorial elections actually increased in the late 1990s in areas with larger expansion populations, even as the same counties began trending away from the Democrats in national elections.

Robustness Checks

The Supplementary Material presents a number of alternative specifications that illustrate the robustness of the results. First, I relax the linearity assumption used in the main analysis and instead classify counties as experiencing above- or below-median levels of enrollment losses and focus on this dichotomized treatment variable. Second, I replicate the Missouri analysis using the 2007 enrollment records, dropping the eight counties where the data are clearly incorrectly reported for that year. Both of these approaches produce results that are strikingly similar to the estimates featured in this section.Footnote 23 Finally, I also include several additional gubernatorial elections for Tennessee and show that the significant 2006 impacts did not persist in subsequent contests.

Conclusion

The results provide evidence that political parties responsible for Medicaid cuts implemented in both Tennessee and Missouri paid a short-term electoral penalty, although there are also important differences between the two cases. Before concluding, I briefly highlight these differences and offer potential explanations for them as well as broader implications that can be examined in future studies.

First, the difference in the impacts on the gubernatorial elections in these two states suggests that voters hold individual incumbents—not the incumbents’ parties—responsible for unpopular policies. This is consistent with Garz and Martin (Reference Garz and Martin2021), who find much larger electoral penalties for rising unemployment for individual incumbents than for their parties. In Missouri, where the incumbent governor chose not to stand for reelection, voters had no one to hold responsible.

Another possible explanation is the demographic difference between the affected populations in the two states. While beneficiaries who lost coverage in both states were disproportionately low-income, able-bodied adults, the affected group was poorer in Missouri (where Medicaid eligibility was previously limited to below 75% of poverty) than in Tennessee (where eligibility extended up to 200% of poverty). In addition, the TennCare expansion population included medically “uninsurable” adults regardless of income. Given the strong association between socioeconomic status and election turnout, it is perhaps not surprising that eliminating coverage from relatively higher-income groups would produce a larger political backlash.

The second important lesson is about the role of strategic candidate entry and retirement decisions, which appear to drive the aggregate electoral effects in the Missouri legislative elections.Footnote 24 Thus, the decisions made by elected officials in anticipation of voter behavior appear to play an important role in producing electoral accountability—even when elites’ expectations prove to be wrong.

Although the experience of these two states provides suggestive evidence about the underlying political processes, my analysis cannot definitively identify the mechanisms by which Medicaid coverage loss influenced voting behavior given the limitations posed by aggregate-level data. However, I can rule out several possibilities. I show in the Supplementary Material that Medicaid disenrollment produced no broader negative local impacts in terms of employment or wages, suggesting that broader economic disruptions are not driving the electoral dynamics I document.Footnote 25 In addition, although news accounts did bring attention to the plight of those who lost coverage and thus could have influenced the behavior of voters who were not personally affected, this is unlikely to explain my results. Because neighboring counties share media markets, the impact of negative press coverage should be absorbed in the commute zone fixed effects, and the results remain substantively unchanged regardless of whether these effects are included. Finally, it seems likely that the effects were due to changes in vote choice (commonly referred to as “persuasion”) rather than differential mobilization, since the latter mechanism should have produced more consistent effects down the ballot across different types of races.

Because my analysis focuses on a two high-profile disenrollment episodes, it is also important to acknowledge that the results may not generalize to other contexts. Both states’ experiences were somewhat unique because the effects were felt almost immediately and the attribution of political responsibility for the observed loss of insurance was straightforward. As Pierson (Reference Pierson1996) notes, neither is typically true in most successful welfare reform efforts. Nevertheless, Tennessee’s—and to a more limited extent, Missouri’s—experience suggests that electoral accountability can provide an important mechanism through which otherwise economically marginalized citizens can influence public policy. On the other hand, during times of fiscal stress, such electoral pressures can also impede the adoption of painful—but necessary—reforms. And fear that electoral calculations could make it difficult to cut benefits once they start flowing to constituents can even prevent welfare programs from being enacted or expanded in the first place.

However, it is important to put the magnitude of the effects documented in this study in context. Despite the political blow back from the TennCare cuts, Gov. Bredesen won reelection in 2006 by the largest margin in nearly a quarter century. His experience suggests that, although welfare cuts may be “politically treacherous,” they may not prove to be electorally decisive.

Since these disenrollment episodes took place more than 15 years ago, I conclude by considering the lessons they may hold for present political dynamics. Given the important demographic similarities between the affected populations in these states and the beneficiaries of the Medicaid expansion under the Affordable Care Act (ACA)—both covered low-income, able-bodied adults—the results suggest that Republican efforts to roll back these coverage gains may impose some electoral costs at the ballot box. Unlike the broader-based and generally unpopular cutbacks experienced in Tennessee and Missouri, however, targeted cuts focused on specific groups perceived to be less “deserving”—for example, via Medicaid work requirements (see Haeder, Sylvester, and Callaghan Reference Haeder, Sylvester and Callaghan2021)—may produce more limited backlashes. In addition, the Medicaid cutbacks in these states affected relatively long-standing programs and long-term beneficiaries. It is far less clear whether the expiration of temporary welfare programs, such as the Child Tax Credit created under the American Rescue Plan, would produce similar political repercussions.Footnote 26

Supplementary Materials

To view supplementary material for this article, please visit http://doi.org/10.1017/spq.2022.15.

Acknowledgments

This manuscript was been much improved thanks to feedback from Jacqueline Chattopadhyay, Stéphane Lavertu, Zac Peskowitz, Elizabeth Rigby, Mike Sances, and Miranda Yaver.

Data Availability Statement

Replication materials are available on SPPQ Dataverse at https://doi.org/10.15139/S3/ZSKW5X.

Funding Statement

The author received no financial support for the research, authorship, and/or publication of this article.

Conflict of Interest

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Author Biography

Vladimir Kogan is an associate professor of political science and (by courtesy) public affairs at the Ohio State University. His research focuses on the intersection of politics and public policy in areas including education and social policy.

Footnotes

1 Dylan, Scott, and Sarah Kliff, 2017, “Why Obamacare Repeal Failed,” https://www.vox.com/policy-and-politics/2017/7/31/16055960/why-obamacare-repeal-failed.

2 For example, Schram and Soss (Reference Schram and Soss2001) show that the caseload declines resulting from the 1996 welfare reforms were portrayed positively by the media, and there is little evidence that the law’s architects ended up paying a political price.

3 There are several other covered groups, including some of the disabled and Medicare “dual eligibles.” However, these populations were largely unaffected by the cuts I examine.

4 Although Medicaid is funded jointly by the state and federal governments, states during this period had considerable discretion over eligibility standards. The cutbacks I examine were limited to beneficiaries who were not guaranteed benefits under federal law.

5 However, Haselswerdt and Michener (Reference Haselswerdt and Michener2019) find that the loss of Medicaid coverage can also increase turnout.

6 Most similar to this study, Kogan (Reference Kogan2021) examines the impact of food stamps on vote choice. However, he focuses on the initial rollout of benefits, rather than a loss of eligibility. Theoretically, the political effects of initial welfare provision may not be symmetric to the consequences of subsequent benefit loss. The results presented below provide some empirical evidence of such asymmetries in the context of Medicaid.

7 The work requirements were blocked by federal courts in a number of states, and suspended in several others, after data showed they would produce significant enrollment losses.

8 As part of settlements to lawsuits filed by the Tennessee Justice Center, a patient advocacy group, the state was constrained by consent decrees that blocked some of the reforms Bredesen hoped to adopt. See Matt Pulle, “Your Move, Phil,” Nashville Scene, November 18, 2004. On the shortfall in expected savings, see Ashley M. Heher, “TennCare Strategy Targets Critics, Politics,” Chattanooga Times Free Press, March 6, 2005, p. 1.

9 Note that, unlike the TennCare expansion, this program did not extend coverage to adults who initially lacked it. Rather, it allowed some existing Medicaid recipients to keep their insurance as their incomes rose, which would have previously caused them to lose eligibility. This has important implications for how these dynamics are modeled.

10 Carnahan was posthumously elected, after dying in a plane crash several weeks before the election.

11 “Blunt Opposes Medicaid Cuts Sought by House GOP,” Jefferson City News-Tribune, June 9, 2004.

12 Additional analysis in the Supplementary Material includes several subsequent Tennessee gubernatorial elections as well.

13 In both states, counties with higher poverty rates and lower median incomes saw larger declines. Perhaps counter-intuitively, however, home ownership rates were also positively correlated with enrollment losses. Other demographic variables—including race and ethnicity—did not consistently predict declines across both states. Full results are reported in the Supplementary Material.

14 I present the event study results visually below, but all are estimated using the preferred specification that includes commute-zone-by-year fixed effects.

15 Election returns in each state provided separate vote totals by county for each legislative district.

16 I focus on House instead of state Senate elections because the latter are staggered, so not all districts are observed in every election cycle.

17 Kousser, Lewis, and Masket (Reference Kousser, Lewis and Masket2007)) provide evidence of similar strategic entry in state legislative races following California’s 2003 gubernatorial recall.

18 Since my election outcomes are measured at the county level, and each county may contain multiple districts, I code incumbency as the percent of county votes cast in districts with Democratic incumbents running minus the percent of votes cast in districts with Republican incumbents on the ballot. For contestation, I similarly calculate the percent of county votes cast in districts where the Democratic candidates faced no Republican opponents minus the share of votes cast in districts where Republicans faced no Democratic opposition.

19 For counties without typographical errors in the records, the 2007 and 2008 enrollment levels were very similar. All replication data are available in Kogan (Reference Kogan2022).

20 Of course, the political calculations also depend on the duration of the negative effect.

21 To identify the party of the incumbent (if any), I use incumbency data from Klarner (Reference Klarner2018).

22 It is possible that the results in Panel B are driven by those in Panel A. In other words, Republicans became less likely to contest a district because Democratic incumbents were more likely to run again. However, I find a contestation effect only in 2006, while the strategic retirement effects are observed in both 2006 and 2008. I also find no evidence that Democratic challengers made entry decisions in response to the retirement of Republican incumbents.

23 The magnitudes of the coefficients on the dichotomized treatment variable are substantially larger, but this simply reflects the rescaling of the measure and implies roughly similar effect sizes as the continuous treatment specification.

24 Although not presented above, I found no evidence that Medicaid enrollment losses affected the retirement decisions of incumbents in Tennessee. The analysis did, however, show that Democratic incumbents in counties harder hit by enrollment losses were more likely to face a Republican opponent in both 2006 and 2008.

25 As I show in the Supplementary Material, counties experiencing larger Medicaid coverage losses were also areas that faced localized negative economic shocks in the early- and mid-1990s. It is possible that these shocks also reduced support for national Democrats during this period and explain the negative effects on presidential elections during this period, consistent with local retrospective economic voting (Healy and Lenz Reference Healy and Lenz2017).

26 Internal polling from progressive groups found a slight increase in President Biden’s approval among parents after the benefits began flowing, but did not examine what happened when the program ended in December 2021. See Colin McAuliffe, Ahmad Ali, and Ethan Winter, “Political Effects of the expanded Child Tax Credit,” March 2022, https://www.dataforprogress.org/memos/2022/3/23/political-effects-of-the-expanded-child-tax-credit.

References

Angrist, Joshua D., and Pischke, Jörn-Steffen. 2009. Mostly Harmless Econometrics: An Empiricist’s Companion. Princeton, NJ: Princeton University Press.CrossRefGoogle Scholar
Bertrand, Marianne, Duflo, Esther, and Mullainathan, Sendhil. 2004. “How Much Should We Trust Differences-in-Differences Estimates.” Quarterly Journal of Economics 119 (1): 249–75.CrossRefGoogle Scholar
Bruch, Sarah K., Ferree, Myra Marx, and Soss, Joe. 2010. “From Policy to Polity: Democracy, Paternalism, and the Incorporation of Disadvantaged Citizens.” American Sociological Review 75 (2): 205–26.CrossRefGoogle Scholar
Clinton, Joshua D., and Sances, Michael W.. 2018. “The Politics of Policy: The Initial Mass Political Effects of Medicaid Expansion in the States.” American Political Science Review 112 (1): 167–85.CrossRefGoogle Scholar
Garthwaite, Craig, Gross, Tal, and Notowidigdo, Matthew J.. 2014. “Public Health Insurance, Labor Supply, and Employment Lock.” Quarterly Journal of Economics 129 (2): 653–96.CrossRefGoogle Scholar
Garthwaite, Craig, Gross, Tal, and Notowidigdo, Matthew J.. 2018. “Hospitals as Insurers of Last Resort.” American Economic Journal: Applied Economics 10 (1): 139.Google Scholar
Garz, Marcel, and Martin, Gregory J.. 2021. “Media Influence on Vote Choices: Unemployment News and Incumbents’ Electoral Prospects.” American Journal of Political Science 65 (2): 278–93.CrossRefGoogle Scholar
Haeder, Simon F., Sylvester, Steven M., and Callaghan, Timothy. 2021. “Lingering Legacies: Public Attitudes about Medicaid Beneficiaries and Work Requirements.” Journal of Health Politics, Policy and Law 46 (2): 305–55.CrossRefGoogle ScholarPubMed
Haselswerdt, Jake. 2017. “Expanding Medicaid, Expanding the Electorate: The Affordable Care Act’s Short-Term Impact on Political Participation.” Journal of Health Politics, Policy and Law 42 (2): 667–95.Google ScholarPubMed
Haselswerdt, Jake, and Michener, Jamila. 2019. “Disenrolled: Retrenchment and Voting in Health Policy.” Journal of Health Politics, Policy and Law 44 (3): 167–85.CrossRefGoogle Scholar
Haskins, Ron. 2016. “TANF at Age 20: Work Still Works.” Journal of Policy Analysis and Management 35 (1): 224–31.CrossRefGoogle Scholar
Healy, Andrew, and Lenz, Gabriel S.. 2017. “Presidential Voting and the Local Economy: Evidence from Two Population-Based Data Sets.” Journal of Politics 79 (4): 1419–32.CrossRefGoogle Scholar
Klarner, Carl. 2018. “State Legislative Election Returns, 1967–2016: Restructured For Use.”. Harvard Dataverse. http://10.0.30.230/DVN/DRSACA.Google Scholar
Kogan, Vladimir. 2021. “Do Welfare Benefits Pay Electoral Dividends? Evidence from the Food Stamp Program Rollout.” Journal of Politics 83 (1):5870.CrossRefGoogle Scholar
Kogan, Vladimir. 2022. “Replication Data for: The Politics of Welfare Retrenchment: Evidence from Mass Medicaid Disenrollment in Two States.” UNC Dataverse. V1. http://10.0.59.35/S3/ZSKW5X.Google Scholar
Kousser, Thad, Lewis, Jeffrey B., and Masket, Seth E.. 2007. “Ideological Adaptation? The Survival Instinct of Threatened Legislators.” Journal of Politics 69 (3): 828–43.CrossRefGoogle Scholar
Michener, Jamila. 2018. Fragmented Democracy: Medicaid, Federalism, and Unequal Politics. New York, NY: Cambridge University Press.CrossRefGoogle Scholar
Pierson, Paul. 1996. “The New Politics of the Welfare State.” World Politics 48 (2): 143–79.CrossRefGoogle Scholar
Schlozman, Kay Lehman, Verba, Sidney, and Brady, Henry E.. 2012. Unheavenly Chorus: Unequal Political Voice and the Broken Promise of American Democracy. Princeton, NJ: Princeton University Press.Google Scholar
Schram, Sanford F., and Soss, Joe. 2001. “Welfare Reform, Policy Discourse, and the Politics of Research.” Annals of the American Academy of Political and Social Science 577: 4965.CrossRefGoogle Scholar
Figure 0

Table 1. Summary of cases

Figure 1

Figure 1. Medicaid enrollment decline by county (as a percent of total population).

Figure 2

Table 2. Effect of Medicaid disenrollment on democratic gubernatorial vote share

Figure 3

Table 3. Medicaid disenrollment and democratic gubernatorial vote share in Tennessee

Figure 4

Figure 2. Event study results for impact of Medicaid disenrollment on gubernatorial elections.

Figure 5

Table 4. Medicaid disenrollment and democratic state house vote share

Figure 6

Figure 3. Event study results for impact of Medicaid disenrollment on state house elections.

Figure 7

Table 5. Strategic candidate entry and exit in Missouri

Figure 8

Table 6. Effect of Medicaid disenrollment on democratic presidential vote share

Figure 9

Table 7. Medicaid disenrollment and democratic presidential vote share in Tennessee

Figure 10

Figure 4. Event study results for impact of Medicaid disenrollment on presidential elections.

Supplementary material: Link
Link
Supplementary material: File

Kogan supplementary material

Kogan supplementary material 1

Download Kogan supplementary material(File)
File 10.8 KB
Supplementary material: File

Kogan supplementary material

Kogan supplementary material 2

Download Kogan supplementary material(File)
File 138.8 KB