Federal public policies in the United States are often implemented by state and local governments, which have significant power over the administration of such policies. Prior work has found evidence that when people directly interact with government programs that benefit them, they are more likely to be politically active (Campbell Reference Campbell2003; Mettler Reference Mettler2005; Soss Reference Soss2000). However, applying for and receiving welfare (i.e., means-tested) benefits, such as financial assistance from the Temporary Assistance for Needy Families (TANF) program or food stamps from the Supplemental Nutrition Assistance Program (SNAP), can be a difficult and demoralizing experience, resulting in negative effects on political participation (Plutzer Reference Plutzer2010; Soss Reference Soss1999). Taken together, there are mixed findings of the impact of being a government program recipient on voting and other political behavior and attitudes. However, these studies have one major limitation: they either examine the simple receipt of a program—rather than the intricacies of the way in which it was administered—or they examine the administration of one program in one narrow geographic location or at one snapshot in time.
In recent work, Clinton and Sances (Reference Clinton and Sances2018) take the study of policy feedback one step of abstraction higher, examining the mass-level effects of the expansion of Medicaid benefits through the Patient Protection and Affordable Care Act (ACA) of 2010. They find that high-Medicaid-eligibility counties in states that expanded Medicaid observed an increase in voter registration (in 2014 and 2016) and turnout (in 2014) compared with similar counties in bordering nonexpansion states—that is, they find some positive mass participatory effects of Medicaid expansion. However, they do not consider the way in which Medicaid is administered; rather, they test the effect of whether the state expanded Medicaid eligibility. To better understand the direction of individual- and mass-level effects of government policies, I argue that we must consider the ways in which policies are administered. Opening the black box of policy recipiency, I advance Clinton and Sances’s (Reference Clinton and Sances2018) work by measuring and adding policy administration to the equation of how Medicaid expansion affected mass-level voter turnout.
This article makes both theoretical and empirical contributions. I develop a theory that the administrative burden (i.e., the “experience of policy implementation as onerous” [Burden et al. Reference Burden, Canon, Mayer and Moynihan2012, 741]) of government programs and elections creates interpretive effects on program recipients, their close contacts, and the rest of the mass public by influencing attitudes about the government and political self-efficacy, which in turn affect political participation. I argue that there are two conditions under which administrative burden may shape the likelihood of voting for recipients and nonrecipients alike: (1) the program is visible and the actor administering the program (the government) is traceable, and (2) the program benefits are substantial and important in individuals’ lives (i.e., those programs providing food, housing, healthcare, etc.). Differing from previous studies, I incorporate the extent to which the administration of elections may create administrative burdens that simultaneously influence voter turnout along with the burdens associated with government program administration. I argue that government programs and elections are highly salient sites of contact with the administrative state that have the potential to shape political participation. I bring together scholarship on policy feedback and administrative burden, arguing that administrative burden is a key instrument of policy feedback effects. Finally, I quantitatively measure the administrative burden of Medicaid and elections separately. I demonstrate that Medicaid and election administration burdens vary across at least three dimensions: (1) geographies, due to the nature of federalism; (2) time, as policies or political environments change; and (3) program, depending on policy design. To ensure the validity of the theory and measurement of Medicaid administrative burden, I conducted 10 interviews with means-tested program recipients.Footnote 1
I then test my theory that policy administration is a critical pathway through which policies affect mass-level political participation. Leveraging the implementation of the ACA in 2014, I utilize a generalized differences-in-differences research design, finding support for my hypothesis that there is an inverse relationship between Medicaid administrative burden and voter turnout. The effect of a county having the highest Medicaid burden level versus the lowest burden level was a 1.14 percentage-point decrease in turnout in recent national elections. This effect is net of Medicaid expansion status, the administrative burden of registering to vote and voting, and many other factors known to be associated with voter turnout. I also find that election administration burden is negatively associated with turnout. The implications of these findings are that if we measure the intricacies of the ways in which a policy is administered, scholars and practitioners can predict whether a policy is likely to create positive or negative mass feedback effects.
Citizen–State Interactions and Policy Feedbacks
Interacting with the government—most often via the administrative state—is commonplace in citizens’ lives. Whether they are renewing a driver’s license at the Department of Motor Vehicles or applying for necessities like food stamps and healthcare coverage from welfare programs, these citizen–government interactions are important in shaping individuals’ perceptions as well as providing resources. Critically, these interactions can have many downstream indirect effects. To arrive at a theory of how the administrative burden of government programs influences mass political participation, it is necessary to first understand what scholars have already uncovered about the impacts of citizen–government interactions.
Policy Feedback Effects on Government Program Recipients
Scholars have repeatedly shown that there are positive policy feedback effects for government program recipients. In her study of Social Security, Andrea Campbell (Reference Campbell2003) demonstrates that senior citizens—particularly low-income seniors—became invested in the Social Security program and experienced increased political efficacy in part because their well-being became linked to the program’s, leading to the creation of a constituency. Campbell shows that Social Security inspired political engagement from its recipients to protect the program from future cuts. In other words, the Social Security program created positive policy feedback effects.
Suzanne Mettler finds similar dynamics with the GI Bill, which in part financed education and training for military veterans beginning after World War II. She finds that veterans who used the education benefits, especially those from lower socioeconomic backgrounds, felt a sense of reciprocity toward the government and subsequently wanted to give back to society, which they did by becoming civically and politically engaged (Mettler Reference Mettler2005). Like Social Security, the education benefits of the GI Bill were not means tested, and there were positive participatory political effects on program participants.
Considering nonuniversal programs, and as summarized in Donald Moynihan and Joe Soss’s (Reference Moynihan and Soss2014) survey article, some means-tested programs that recipients perceive to be administered in a fair and participatory fashion can increase feelings of political efficacy and enhance engagement with the political process (Bruch, Ferree, and Soss Reference Bruch, Ferree and Soss2010; Soss Reference Soss2000). In sum, there is strong reason to believe that those who receive government benefits, under certain conditions, will have a greater sense of political efficacy, more trust in the government, and demonstrate increased political activity.
At the same time, applying for and receiving means-tested benefits, such as food stamps from SNAP, can be a difficult and stigmatizing experience. Though the aforementioned studies find positive effects of being a recipient of a government program, other work has found negative impacts of being a recipient on political efficacy and participation (e.g., Plutzer Reference Plutzer2010; Soss Reference Soss1999). Soss (Reference Soss2000) finds that, just like senior citizens and veterans, individuals learn about the power of government when they apply for means-tested programs. However, given the onerous and demeaning interactions that can come with applying for welfare (as opposed to GI benefits), this contact can disincentivize political participation (Soss Reference Soss2000). In their study of TANF, a paternalistic program with work requirements, time-limited aid, and street-level bureaucrats responsible for much of the implementation, Bruch, Ferree, and Soss (Reference Bruch, Ferree and Soss2010) find negative effects on the political participation of recipients. In Jamila Michener’s (Reference Michener2018, 77–78) study of Medicaid, she finds that Medicaid recipients were significantly less likely to vote, register to vote, and politically participate in other ways compared with nonrecipients. Finally, experience with the criminal justice system—a decidedly negative interaction—also causes unfavorable attitudes about the government and decreases civic and political participation (Weaver and Lerman Reference Weaver and Lerman2010).
Overall, while some prior work has shown positive policy feedback effects after interacting with the government via public policies and programs, other studies have demonstrated negative effects. Most existing work either examines program administration at one snapshot in time—often in one location—or it examines the effects of being a program recipient in a binary way, without considering the way in which administration may vary within a program. In contrast, the empirical work that follows examines the policy feedback effects of the variation in program administration across the US and over time, while holding the program constant.
Mechanisms of Policy Feedback
Scholars have also probed why participation in government programs produces changes in political participation and trust in government. The main mechanisms that create policy feedback effects are the distribution of resources, the generation of interests, and interpretive lessons. Resource effects are anything that increases participatory capacity as a result of being a program recipient (Pierson Reference Pierson1993). They include the results of the benefits themselves that the program provides (e.g., lack of hunger, better health, more money to spend on necessities) as well as free time and personal capacity. “Interpretive” effects on program recipients occur when these individuals learn about how the government operates and treats them (Pierson Reference Pierson1993). Especially for means-tested program administration, interactions between citizens applying for these programs and the bureaucrats who administer them are direct and personal, and the benefits are substantial and important in the lives of recipients (Soss Reference Soss2000); these factors pave the way for the possibility of interpretive effects. As Mettler and Soss (Reference Mettler and Soss2004) argue, policies teach program recipients about important democratic concepts such as the meaning of citizenship and influence who is perceived as included in the political community and society. Therefore, the main mechanisms in my theory linking the administrative burden of means-tested programs to voter turnout for program recipients are interpretive effects, which is consistent with Mettler’s (Reference Mettler2002, 353) general policy feedback framework of “how policy affects civic engagement.”
Individuals eligible for means-tested benefits are not the only members of society who may be affected by means-tested program administration. For nonrecipients, the interpretive effects of administrative burden may spill over for a highly visible and substantial public program. Policies influence the public, even when individuals are themselves not recipients, because the ways in which a program is designed and administered influence perceptions of societal programs and their recipients. The government conveys messages to the public about group characteristics based on policy design and implementation (Mettler and Soss Reference Mettler and Soss2004); for example, capricious administration of means-tested programs, which by definition target a low-income population, may lead the public to view program recipients as undeserving and form negative opinions of the program.
Interpretive effects of government programs may also extend to nonrecipients when someone they know well tells them about their direct experience interacting with the government or when they observe that person’s interactions with the state (e.g., White Reference White2019), or when they live in a community with a high concentration of government program recipients (e.g., Michener Reference Michener2017). In their framework for the analysis of mass feedback processes, Soss and Schram (Reference Soss and Schram2007) emphasize that “participant status” in a policy or program is not a necessary condition for experiencing policy feedback effects, especially for policies that are highly visible to the mass public and have some degree of proximity. Going further to characterize how different subsets of the public may interact with government programs, Michener (Reference Michener2017) establishes a policy contact framework, which categorizes members of society based on their closeness to the policy or program (personal or impersonal) and whether they are direct recipients of the policy or indirectly related to it. Especially in communities with a high concentration of government program recipients, “[p]olicies can teach entire communities about government and politics” (Michener Reference Michener2017, 873). Studying the political effects of proximal contact with the criminal justice system, White (Reference White2019) finds that people with household members who have been jailed or convicted in the weeks leading up to an election experience a demobilizing effect on their propensity to vote in that election. Taken together, these studies provide evidence of effects on the political participation of people with indirect contact with the government.
Scholars have also investigated mass effects of means-tested program administration on political attitudes by using experimental methods, which help to demonstrate that even indirect administrative burden can have effects on political attitudes. Utilizing a survey experiment about TANF, Keiser and Miller (Reference Keiser and Miller2020) find that Republicans, who overwhelmingly hold negative views about the program, view it more positively when presented with a highly burdensome description of its application process. Democrats, who hold more positive views of the program, were not affected by a high- or low-burden treatment compared with the control, a neutral description of the program’s application process. In another survey experiment with multiple types of government program applicants—a veteran, a disaster food-aid recipient (“deserving” applicants), a parolee, and a SNAP recipient (“undeserving” applicants)—Nicholson-Crotty, Miller, and Keiser (Reference Nicholson-Crotty, Miller and Keiser2021) find that high levels of burden decrease the approval of disaster food aid (not means tested) for the veteran applicant but not for the parolee or SNAP recipient. Low burden levels decrease the approval of SNAP (means tested) when the applicant is a parolee but not when they are a veteran or a disaster-aid recipient. These results support the idea that even just knowing about the administrative burden of government programs has effects on political attitudes.
Both studies link administrative burden of government programs to the public’s views of the programs themselves through political ideology and opinions about the applicant and program type. But they do not demonstrate if there is a link between administrative burden or views of government programs (influenced by administrative burden) and political participation. Examining the expansion of (means-tested) Medicaid benefits through the ACA, Clinton and Sances (Reference Clinton and Sances2018) find that high-Medicaid-eligibility counties in states that expanded Medicaid observed an increase in voter registration in 2014 and 2016 and in voter turnout in 2014, compared with neighboring counties in bordering nonexpansion states. This is evidence of the mass positive participatory effects of this means-tested program. While this Medicaid study connects the existence of an expanded Medicaid program to mass-level participation, it does not consider the way in which state Medicaid programs are administered, which I will show varies greatly by state. Relatedly, Michener (Reference Michener2018, 79–80) finds that the negative relationship between Medicaid recipiency and political participation is evident for recipients in some states, but not for those from other states. These findings also suggest that there is something deeper than program recipiency that is affecting political participation. In this article, I empirically establish a mass-level connection between administrative burden and voting, complementing previous studies.
Administrative Burden as an Instrument of Policy Feedback Effects
One of the key policy characteristics that generates policy feedback effects is the lived experience of program administration, or “administrative burden” as it is known in the literature. In the past decade, scholars from political science, sociology, and public administration—led by Pamela Herd and Donald Moynihan—have contributed to this fast-growing field. In this section, I define administrative burden and the ways in which it varies for means-tested programs to lay the foundation for my theory of the political implications of administrative burden.
Administrative burden is defined as an individual’s “experience of policy implementation as onerous” (Burden et al. Reference Burden, Canon, Mayer and Moynihan2012, 741). It can take many forms: time spent learning about program eligibility criteria, strict document verification, the perceived stigma of receiving government aid, or long lines at government offices. Herd and Moynihan (Reference Herd and Moynihan2018, 18) aptly note that administrative burdens are not intrinsically bad, as they might serve a genuine purpose like ensuring only those eligible are able to enroll in a program; these burdens are simply the result of the existence of citizen–government interactions. The costs of these burdens may be put upon individuals, the administrative state, or both.
More specifically, administrative burden is the totality of the costs—learning, psychological, and compliance—that citizens experience when interacting with the government (Herd and Moynihan Reference Herd and Moynihan2018, 15). Learning costs are associated with learning about a program’s existence and its potential benefits, eligibility criteria, and enrollment processes. Psychological costs are associated with program stigma or other emotions that come with applying for or receiving program benefits, such as a sense of a loss of autonomy, frustration, or stress with the process (Moynihan, Herd, and Harvey Reference Moynihan, Herd and Harvey2015, 49). Compliance costs are the burdens of following rules and requirements for accessing programs, such as completing applications or providing documentation confirming eligibility.
These costs, and therefore their administrative burden, can vary in many ways. In particular, policy design and policy implementation contribute to the creation of burden (Herd and Moynihan Reference Herd and Moynihan2018, chap. 1). Means-tested programs are, by design, more likely to be highly burdensome than universal programs due to eligibility determinations.
Federal policies that are administered by states and localities are often intentionally vague in terms of implementation specifics, such as the design of customer service, program information outreach and advertising, and the ways through which one can seek help with applying for a program’s benefits. Many of the programs that provide basic human necessities to vulnerable populations are both means tested and implemented by states and localities rather than the federal government; this includes income and food assistance (TANF and SNAP, respectively), health insurance (Medicaid), and public housing assistance (Section 8 housing). Furthermore, even among means-tested programs, levels and aspects of administrative burden can differ substantially. As Barnes, Michener, and Rains (Reference Barnes, Michener and Rains2023) demonstrate in their qualitative study of the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC), SNAP, and Medicaid, there is significant variation in administrative burden levels, which they attribute in part to differences in policy design and corresponding pressures on the street-level bureaucrats who administer the programs.
There are also several studies that examine the effects that administrative burden has on means-tested program recipients. Fox, Feng, and Reynolds (Reference Fox, Feng and Reynolds2023) quantitatively measure the administrative burden from 2000 to 2016 of three federal–state partnership programs—TANF, SNAP, and Medicaid—and find that reducing the number of burdensome program rules was associated with greater policy take-up rates for each program, even controlling for program generosity and eligibility rules. Likewise, Barnes, Halpern-Meekin, and Hoiting (Reference Barnes, Halpern-Meekin and Hoiting2023) utilize interviews with WIC recipients to show how experiences in accessing and utilizing WIC benefits contribute to their participation in the program. In particular, negative experiences with WIC may cause mothers to opt out of the program even when they are in need of its benefits. Focusing on Medicaid, Michener (Reference Michener2018) finds in her interviews with recipients in multiple states that the variations in the program’s provision affects recipients’ political participation and attitudes. Continuing in this vein, I take a different approach to examine how variation in Medicaid administration may affect mass political participation. I measure Medicaid administrative burden over a decade-long time period in all 50 states and in Washington, DC, examining its effects on the likelihood of voting for recipients and nonrecipients combined.
How Administrative Burden Influences the Likelihood of Voting
Though I am not the first to suggest that the way in which a government program is administered has political implications, I bring together literatures on policy feedback and administrative burden to describe how administrative burden created by the government has interpretive effects on individuals with direct and indirect contact, shaping their likelihood of voting. I focus on voting rather than other acts of political or civic participation because voting is the most widespread and accessible form of participation (Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995).
Throughout this article, I refer to “the government” as one entity despite its complexity and span across three branches of government—and the departments, agencies, and offices within—and federal, state, and several local levels. This is an intentional choice, appropriate for the setting of visible, means-tested program provision. Soss’s (Reference Soss1999, 368) work demonstrates that means-tested program recipients see “welfare institutions as part of ‘one big system’ of government.” Utilizing in-depth interviews with recipients of Aid to Families with Dependent Children, the supplemental income and family assistance program that preceded TANF, Soss finds that recipients associated the local welfare agency with which they interacted with “the government” more broadly. He attributes this to the direct provision of welfare benefits—that is, the high traceability of the program, which, for many recipients, is their only or most important interaction with any governmental entity (Soss Reference Soss1999).
There is also evidence that the public views government agencies as unified entities in Carpenter’s (Reference Carpenter2010; Carpenter and Krause Reference Carpenter and Krause2012) work on organizational reputation. Given that public agencies are made up of “differentiated functionaries or specialists” and play a variety of roles, they are difficult for the public to characterize (Carpenter and Krause Reference Carpenter and Krause2012, 28). In addition, those who are outside of public agencies tend to view these agencies as more unified than they are (Carpenter and Krause Reference Carpenter and Krause2012). This leads the public to rely upon heuristics to simplify a complex organization into one seemingly less complex unit: “the government.”
Finally, I have evidence from the 10 interviews I conducted in March 2022 with individuals eligible for means-tested programs that “the government” is often viewed as one entity. First, as in Soss’s interviews, individuals consistently referenced “the government” in our conversations about their interactions accessing benefits, such as Medicaid and food stamps, from their local welfare office. Second, I asked participants who had applied for or enrolled in Medicaid to whom they attributed that benefit: several individuals were confused by the question, answering that it was “the government.” When they were pushed, I found no consensus as to what “the government” was: some said it was their local government, while others said it was their state or the federal government, or some mix of all these entities. Therefore, when I say “the government” in this article, I am referring to whatever the public and program recipients view as the government—often one vague, unified entity—because my theory concerns individual-level psychological processes.
The Sites of Contact with the Administrative State
Figure 1 presents two highly salient sites of contact with the administrative state that, I argue, influence one’s likelihood of voting: government program administration and election administration. There may be downstream effects of administrative burden experienced at these sites on the decision whether to vote, based on the extent of the individual-level interpretive effects on attitudes about the government, the program, and recipients, as well as on political self-efficacy, all of which may influence the likelihood of voting. Such interpretive effects of direct and indirect government program contact have already been documented by studies discussed in the preceding sections (e.g., Bruch, Ferree, and Soss Reference Bruch, Ferree and Soss2010; Nicholson-Crotty, Miller, and Keiser Reference Nicholson-Crotty, Miller and Keiser2021; Soss Reference Soss2000). This theory is most applicable to those with direct or indirect personal contact with these administrative burdens: recipients and their close friends and family. Additionally, other individuals living in a locality with a high concentration of government program recipients may also have their political attitudes and participation shaped by administrative burden, in line with Michener’s (Reference Michener2017) work demonstrating that counties with higher concentrations of Medicaid recipients had lower rates of political participation.
The Pathway between the Sites of Administrative Burden and Political Participation
In the dotted box in figure 1, individuals have direct or indirect contact with the government and, in doing so, encounter several aspects of administrative burden. Considering those eligible for means-tested programs, they must learn about that program’s existence and its eligibility criteria, evaluate if they are eligible based on the criteria, and then figure out how to apply. If they are able to overcome those learning costs and any psychological costs, they may attempt to enroll in the program, at which point they have some type of interaction with the administrative state. During these interactions with the government via street-level bureaucrats, state-run websites, and the program application itself, the individual faces all types of costs.
Some individuals will succeed in enrolling in the program, having overcome the burden, while others will not be able to do so. In either case, potential recipients’ views of their own political efficacy and opinions about the government and program will be affected—in other words, there will be interpretive effects of this administrative burden. When the next election nears, these individuals will additionally encounter electoral administrative burdens when they interact with the government in the voter registration and voting processes. Such burdens include the ease of registering to vote, how soon before an election they can register, knowing where to vote and how to get there, having the required identification to verify their identity at the polls, how long they must wait to cast their ballot, and more. As with means-tested program administration, citizens learn about the extent to which the government values them through their interactions via election administration. Based on the overall tolerability of the burdens associated with both the administration of means-tested programs and elections, individuals may decide to opt out of the process altogether or to persevere. For nonrecipients, those in contact with someone who is eligible—such as a family member, close friend, or community member—may experience interpretive effects that are similar to those experienced by recipients as they observe or hear about the burdens faced during program administration.
The Conditions under Which to Expect Effects of Administrative Burden on the Mass Public
I argue that there are two conditions that must be met for a policy to have mass-level feedback effects—that is, effects that extend to nonrecipients as well as recipients. First, the program must be visible—and therefore traceable to “the government”—and second, it must provide substantial services. Programs that are visible and traceable to the government, rather than those that are “submerged” or imperceptible (Mettler Reference Mettler2011; Reference Mettler2018), are typically applied for and/or given to recipients directly, such as Social Security Disability Insurance, Medicaid, or Pell Grants. They can be means tested or universal. Conversely, “submerged” policies are often doled out through the tax code, such as the Earned Income Tax Credit (Mettler Reference Mettler2018, 59). Applying Arnold’s (Reference Arnold1990, 47) “traceability” framework to this context, programs will be traceable to the government if citizens can trace some observed policy effect—such as the administration of a government program—back to some governmental or policy action and then back to an individual politician’s visible contribution to the policy. For program recipients, they must attribute their experience accessing a policy in a local welfare office—or online—to the government for there to be a traceable link from the administrative burden of the program to its interpretive effects and ultimately to the decision or capacity to vote. For nonrecipients, it is also necessary that they attribute the burden associated with a given program to the government for there to be interpretive effects on and downstream consequences for the likelihood of voting.
Programs may become highly visible and traceable to members of the mass public—not just recipients—if they are highlighted in the news cycle or advertised in public places like on billboards or public transportation. While there are many factors that contribute to what is covered in the news, programs that are new, are highly politically polarized, or affect a large proportion of the population may be more likely to receive significant news coverage or have widespread advertising in public places. Areas with high concentrations of program-eligible individuals are especially ripe for spillover effects on nonrecipients (Clinton and Sances Reference Clinton and Sances2018; Michener Reference Michener2017).
The second condition upon which participatory effects on the mass public rest is that the program benefits must provide substantial and basic human resources such as food, supplementary income, healthcare, housing, or education. The benefits provided will therefore be salient and important in recipients’ lives and will likely activate resource effects, as recipients have more capacity to participate in politics. Most typically, the policies that will meet both conditions will be means-tested programs, since programs for low-income individuals and families provide services to fulfill basic needs that are not otherwise met. However, not all means-tested benefits will meet the first condition of visibility and traceability—for example, the Earned Income Tax Credit.
Interpretive effects are likely to be not the only pathway through which administrative burden shapes the likelihood of voting. For program recipients there may also be resource effects, as they gain resources provided by the government. The inverse relationship between administrative burden and policy take-up demonstrated by Fox, Feng, and Reynolds (Reference Fox, Feng and Reynolds2023) implies that if administrative burden is very high and fewer people therefore enroll in the program and have less access to its resources, then it may be the lack of resources that decreases the likelihood of voting. For program recipients, there may therefore be multiple mechanisms through which administrative burden affects voter turnout: resources and interpretive lessons. However, while this article tests for the mass-level turnout effects of administrative burden, it is outside the scope of this article to test the mechanisms; the rich, in-depth qualitative studies by Michener (Reference Michener2018) and Soss (Reference Soss2000) previously discussed establish the relationship between program administration and political learning or other attitudinal effects among program recipients at a minimum. Given that potential recipients’ lives are likely to be much more significantly impacted by the administrative burden of government programs than the lives of those not eligible, I expect that any observed mass-level effects of burden are driven by a change in political participation by those with a direct or indirect connection to the government program, which may be a significant proportion of the public, depending on the program.
A Focus on Medicaid and Elections
In this section, I explain the decision to focus on Medicaid and elections as critical sites of interaction with the administrative state when examining administrative burden’s effects on mass-level voter turnout. Though administrative burden has been shown to differ by program (Barnes, Michener, and Rains Reference Barnes, Michener and Rains2023; Fox, Feng, and Reynolds Reference Fox, Feng and Reynolds2023), and many recipients of Medicaid are also eligible for and recipients of other government programs, I take the approach of focusing on only one government program, Medicaid, as a first test of the theory of how the administration of certain government programs can shape the likelihood of voting. Additionally, as I discuss in this section, Medicaid is the most prevalent means-tested program in the US, and it experienced a shock to its administrative burden due to the passage and implementation of the ACA, which came with a new policy design and elevated the public visibility of the program. No new laws concerning other means-tested programs were enacted during the 2010s, the period of study. Finally, I highlight the extensive literature on the impact that election laws may have on voter turnout, demonstrating that it is necessary to take election administration burden into account when studying voter turnout.
Background on Medicaid in the 2010s
Medicaid is a joint federal–state program providing health insurance for low-income individuals in the US, significantly lowering the costs of accessing healthcare. The ACA, otherwise known as “Obamacare,” was enacted in March 2010 under the Obama administration to reform Medicaid and, more broadly, healthcare in the US. One of the key provisions of the ACA required states to expand Medicaid coverage to include all adults with household incomes at or below 138% of the federal poverty level, though a June 2012 US Supreme Court ruling made the expansion of state Medicaid programs optional. States began implementing Medicaid expansion in January 2014.Footnote 2 The program directly affects a significant proportion of the American public: as of December 2020, 18.2 million newly eligible adults had enrolled in the Medicaid expansion, with 81.9 million people enrolled overall (Kaiser Family Foundation 2025). In other words, about one in five Americans are currently enrolled in Medicaid. Kaiser Family Foundation (2023) polling from 2017 and 2023 demonstrated that two-thirds of Americans had received Medicaid or had family members or close friends who had. Medicaid’s far reach makes it the ideal program for the study of the mass-level participatory effects of administrative burden.
The ACA and state Medicaid expansion decisions were highly visible in recent election debates, media reports, and Supreme Court decisions, and disputes over work requirements tied to Medicaid eligibility were highly publicized in national and local news, which I argue allowed the administrative burden of the ACA to be visible and relevant, even to those not eligible for Medicaid. Given its high visibility, its traceability to the government, and the substantial and critical benefits it provides, Medicaid fulfills the conditions under which there may be mass feedback effects and is an ideal case to study the effect of program administration on the likelihood of voting among the mass public, including both recipients and nonrecipients. In addition, there is the potential for great variation in how different states administer the program due to its decentralized implementation, as well as in how administration within states changes over time.
The ACA–Administrative Burden Link
Along with the option for states to expand Medicaid coverage, the ACA required all states to streamline their enrollment and renewal processes, delivering a shock to the administrative burden of the Medicaid program. For example, the ACA specified that applicants should have the option to apply online or by telephone (in addition to being able to apply in person), that a face-to-face interview would not be required at enrollment or renewal, and that there would be a minimum 12-month eligibility period. Many states also revised other aspects of their enrollment and renewal processes, such as expanding their online Medicaid account capabilities, making websites and applications mobile friendly, and initiating and processing renewals automatically. The ACA prompted these changes, with many having taken place when the law was first implemented in 2014; this timing also coincides with the first state Medicaid expansion implementations. Therefore, the administrative burden of Medicaid decreased after the implementation of the ACA in 2014 for both expansion and nonexpansion states; however, the extent to which it decreased was determined by each state’s administrative decisions, resulting in great variation as I will demonstrate. Consequently, my national election analyses include data from pre- and post-ACA implementation (from 2010 to 2020) to gain causal leverage over this sudden change in administrative burden. Following from my theory and the literature, I expect that communities exposed to relatively high-burden Medicaid administration will subsequently have lower rates of voting than communities with lower Medicaid administrative burden, controlling for the administrative burden associated with elections and many other factors that may be related to turnout.
Election Administration and Voter Turnout
When considering how interactions with the state may shape voter turnout, a very salient site of contact is through the administration of elections themselves. In fact, several scholars have attempted to isolate the effects on voter turnout of voter identification (ID) laws, registration laws and requirements, and other factors; I highlight only a handful of studies. One of the takeaways from this literature is that it is not straightforward to isolate the effects of individual election laws on turnout for many methodological reasons (see discussion of the challenges in Highton [Reference Highton2017]). For example, having reexamined data used by Hajnal, Lajevardi, and Nielson (Reference Hajnal, Lajevardi and Nielson2017), who conclude that strict voter ID laws decrease minority voter turnout, Grimmer and colleagues (Reference Grimmer, Hersh, Meredith, Mummolo and Nall2018) find mixed results depending on the specification of their model: voter ID laws sometimes have positive effects on turnout, but at other times have negative or no effects.
In Highton’s (Reference Highton2004) article on how registration laws and requirements influence turnout, he discusses how registration “closing dates” or deadlines affect turnout; in particular, people who move around and young people are less likely to vote when registration deadlines are early—presumably because they are not able to update their voter registration in time. This is the only registration-related measure Highton reviews that has an effect on turnout. Therefore, a second takeaway from this literature is that some aspects of election administration and laws affect voter turnout, which supports my classification of elections as an influential site of contact with the administrative state.
Measuring across Dimensions of Administrative Burden
The costs of administrative burden must be measured across as many of the following four dimensions as is feasible for data collection: (1) across geographies, due to the nature of federalism; (2) over time, as programs or political environments change; (3) by program, depending on policy design; and (4) across individuals, since burdens are distributive (e.g., Herd and Moynihan Reference Herd and Moynihan2018). This implies that precise measures of burden will be local, time varying, policy specific, and individualized when possible. In the data analysis that follows, my measurement of administrative burden considers the costs across the first three dimensions for both Medicaid and election administration, in line with the approach of Fox and colleagues (Reference Fox, Feng, Zeitlin and Howell2020). While individual-level variation in administrative burden is not incorporated into my measures, my approach of measuring administrative burden at the level of program administration decisions allows measurement across all states and over time, which would not be possible if I focused on individual-level variations in burden.
For Medicaid, I measure the enrollment and renewal processes at the state level, because administration decisions are made by the states. I collected data on many measures of Medicaid administration based on available information from the Kaiser Family Foundation’s (2025) comprehensive annual reports on state-level policies for Medicaid eligibility and enrollment and snapshots of state Medicaid websites obtained via the Wayback Machine, a webpage archive. Fox and colleagues (Reference Fox, Feng, Zeitlin and Howell2020) utilize the same Kaiser Family Foundation reports to construct indices of the dimensions of Medicaid coverage from 2000 to 2016, one of which is related to administrative burden. Out of the 34 indicators used in the Medicaid burden index by Fox and colleagues, I include 27.Footnote 3 I add four indicators that are not captured in their index and reflect important aspects of Medicaid administration, as identified by my interviews with means-tested program recipients. First, I was repeatedly told that some of the biggest hurdles to accessing benefits was providing verification documents (such as a copy of one’s Social Security card, a recent pay stub, a utility bill with current address, etc.) and renewing coverage. I therefore incorporated data on the ability to upload verification documents online and whether it is possible to renew coverage online.
I also collected data on whether the state had implemented work and reporting requirements. In the context of Medicaid, work requirements limit eligibility to those individuals who work a certain number of hours per week or month and can document and report their hours to their state. In 2018, Arkansas became the first state to implement Medicaid work requirements, and a comprehensive study in 2019 of the impact of these requirements found that they were associated with significant rises in the uninsured rate but no change in the employment rate itself (Sommers et al. Reference Sommers, Goldman, Blendon, Orav and Epstein2019), indicating an overall negative impact. Between 2018 and 2020, 11 other states implemented work requirements as a barrier to accessing Medicaid, or attempted to do so. I therefore include an indicator for whether the state had attempted to implement work requirements in the Medicaid burden index.Footnote 4
Finally, the ACA provided grants to states, counties, and local organizations that sought to help people apply for Medicaid benefits. This took the form of navigators, assisters, or insurance agents and brokers. A key way that these groups could provide direct help with citizens’ Medicaid applications was to help with the application process. I therefore add an indicator to the index for whether it was possible for citizens to authorize third-party access to their online Medicaid applications.
Each of these indicators represents a type of cost associated with administrative burden. Many are compliance costs, which often have psychological costs associated with them. Applicants compile documents for verification, complete long applications, and must remember to renew their coverage, while facing stress, anxiety, or frustration as they attempt to comply with these requirements. The indicator related to a lack of Medicaid navigators and assisters represents both learning and psychological costs, as these third parties are experts in navigating the Medicaid application process, sharing their knowledge about these processes with applicants and giving them practical support to help them meet application requirements.
I assign each indicator of Medicaid administrative burden a value of one if it is present and zero if it is not, so that a higher value indicates higher administrative burden. For example, if a state does not allow applicants to upload verification documents online, the state receives a value of one for that indicator. I add the indicators together because in practice, it is the sum of all aspects of administrative burden that influence one’s perception of program administration, rather than each indicator considered in isolation. Summing the 31 indicators in table 1 together, I divide the total in each state-year by 31 to scale the variable from zero to one for ease of interpretation.Footnote 5
Indicators in the Medicaid Administrative Burden Index, by Theme
Note: † Indicators not included in the Fox et al. (Reference Fox, Feng, Zeitlin and Howell2020) measure. NP = not parents; CH = children.
Figure 2 presents snapshots of the Medicaid burden measure over time. As the ACA provisions go into effect, there is more variation across the country in administrative burden levels, and nearly all states lower Medicaid burden to some extent, as shown by the shades lightening over time. Of note, only seven of the 31 high-burden indicators included in the index were eliminated by 2016 in all states; therefore, the levels of administrative burden across the states vary primarily due to individual state decisions on how to administer their Medicaid program rather than due to ACA mandates.
Evolution of the Medicaid Administrative Burden Index
Note: Index ranges from zero to one and includes the 31 indicators in table 1.
Next, figure 3 presents the trend in the average Medicaid burden level across the states in each year. The vertical line at year 2014 indicates when the ACA went into effect. There is a sharp decline in the average administrative burden level after 2014 when the ACA mandated several burden reductions in all states, which serves as a validity check of the measure itself.
Average Medicaid Administrative Burden Index over Time
I expect the Medicaid administrative burden index to have a negative effect on county-level voting rates, as higher burden should result in lower turnout. Figure A1 in the online appendix visually demonstrates that there is a negative bivariate relationship between Medicaid administrative burden and voter turnout in national elections between 2010 and 2020. Utilizing this singular Medicaid burden index permits causal identification and flexible model specification, as detailed in the methods section that follows. Because the indicators that make up the index do not necessarily vary independently of each other, the causal effect of each indicator individually is not identifiable. In fact, my dataset of burdensome measures shows that states rarely changed just one measure at a time (Dost Reference Dost2025).
Next, I measure the administrative burden of elections at the state-year level, since many election administration decisions are made by the states and vary over time as political environments and policies change. While there is a well-established Cost of Voting Index (Li, Pomante, and Schraufnagel Reference Li, Pomante and Schraufnagel2018; Pomante Reference Pomante2025) that captures many aspects of election administration law in the 50 states going back to 1996, the measure exists only for presidential elections and does not cover midterm years prior to 2022. My period of analysis is all national elections between 2010 and 2020, including midterm years, so I created my own measures of election administration burden.Footnote 6 I collected several relevant indicators, most from the National Conference of State Legislatures website that displays current information on the state of election administration. I utilized archived webpages via the Wayback Machine to collect data on prior years. In all federal election years beginning with 2010 and ending with 2020, I compiled data on whether it is possible to register online to vote, be automatically registered to vote by the state, vote early, obtain permanent absentee voter status, and register to vote in person on the day of the election; on the timing of the voter registration deadline relative to election day; and on the relative stringency of voter identification laws.
The absence of all these measures—except for the registration deadline and voter ID law strictness—represents a raising of the electoral burden (i.e., their absence makes it harder for citizens to register to vote and cast their ballot). My data show that, over time, overall electoral burden has fallen as these measures have been adopted in more states. By 2020, most states had some form of early in-person voting and online voter registration. However, still less than half of the states in 2020 had same-day voter registration or state-sponsored automatic voter registration (e.g., being registered to vote when one renews a driver’s license unless one opts out), or give registered voters the option to have permanent absentee voter status (which allows voters to always vote by mail if desired).
The number of days between the registration deadline and election day also varies over time and by state.Footnote 7 The maximum number of days is 30: this is because Section 8 of the National Voter Registration Act of 1993 (also called the “Motor Voter Act”) requires that states accept voter registration applications if they are submitted at least 30 days before federal elections.
For the measure of voter ID law strictness, I adapted measurements from the National Conference of State Legislatures and the Brennan Center for Justice. States that do not require documentation to vote receive a value of zero. States with the least stringent voter ID laws, which request an identifying document that does not need to be a photo ID, have a value of 0.25. States have a value of 0.5 if a photo ID is requested but not required to cast a ballot, or if a provisional ballot may be cast without a photo ID and be counted without the voter having to take any additional steps. States with values of 0.25 and 0.5 are considered “nonstrict” by the National Conference of State Legislatures, compared with those that have values of 0.75 and one, which are labelled as “strict.” To receive a value of 0.75 or one, a state will allow voters without an acceptable ID (no photo required for 0.75 and a photo required for one) to cast a provisional ballot, but will require them to take steps after election day for their vote to count. Looking over the decade-long period, there are many more states with strict or photo-ID-requiring laws in 2020 than there were in 2010, and some states have not made any legislative changes while others have made many. The trend toward stricter voter identification laws is recent, and my dataset contains nearly the entire history of voter ID laws passed in the US (Weiser and Norden Reference Weiser and Norden2011).
Using these measures, I create two additive indices that measure different aspects of electoral burden, again utilizing composite measures rather than individual indicators because it is the administrative burden imposed by a given government function as a whole that may have impacts on citizens. Both indices are then scaled to range from zero to one. The first index is the burden associated with registering to vote, and includes four measures: online registration, automatic registration, same-day registration, and the time between the registration deadline and election day.Footnote 8 I present election-year snapshots of the index of registration burden in figure 4. As shown, the index varies over time and by state, with voter registration generally becoming less burdensome over time.
Voter Registration Burden Index over Time
Note: Index ranges from zero to one and includes four indicators as described in the text.
Next, I compute an additive index of the burden associated with voting itself, which includes three measures: permanent absentee status, early in-person voting, and voter ID law strictness. I present the index of voter turnout burden in figure 5. Like registration burden, the index varies over time and by state, though there is less variation. Overall, I expect that counties in states with higher electoral burdens will, on average, be associated with lower rates of voter turnout compared with counties with less burden.
Voter Turnout Burden Index over Time
Note: Index ranges from zero to one and includes three indicators as described in the text.
Methodological Approach
Though my theory of how administrative burden shapes political participation operates at the individual level, I aggregate to the county level to accurately measure voter turnout and other community characteristics. The ideal dataset would be individual-level national data on voting behavior, demographics, and Medicaid recipiency, but no such administrative dataset exists. Voter file data do not include validated, detailed demographics or Medicaid recipiency. Nationally representative survey data would potentially be a good option, but there is known to be significant overreporting of voter turnout (e.g., Bernstein, Chadha, and Montjoy Reference Bernstein, Chadha and Montjoy2001; McDonald Reference McDonald2003). Aggregating to the county level permits precise measurement of the dependent variable—voter turnout—because administrative records, rather than self-reported data, are utilized. In addition, using counties allows for substantial geographic variation across the US. As I expect the administrative burden of Medicaid to shape voting likelihood for those who have direct contact with the program as well as those whose contact is indirect, it is appropriate to aggregate to the county level to examine mass-level effects.
I use a generalized difference-in-differences research design centered on the time period in which the ACA was implemented: 2010 to 2020. I employ a two-way fixed-effects estimation, which leverages county and state variation over time while controlling for county and year fixed effects and conditioning on time-varying covariates. Beginning with the universe of all counties in the US, I narrow the sample to counties within 100 miles of the state border to gain causal leverage, since the variation in burden that I measure is at the state level. By focusing on changes in voter turnout at state borders, my design can rule out many alternative explanations for the relationship between administrative burden and political participation.Footnote 9 I borrow from the methodological approach outlined by Clinton and Sances (Reference Clinton and Sances2018), who utilize a variation of the difference-in-differences design, comparing counties within 100 miles of a neighboring state of a different treatment (Medicaid expansion) status.
The main identifying assumption for this approach is that changes in voter turnout in counties with small changes in Medicaid administrative burden are a good counterfactual for changes in turnout in counties with large changes in administrative burden, conditional on pretreatment covariates.Footnote 10 I rely on the assumption that states did not choose how to administer Medicaid programs based on how it would affect voter turnout. I also assume that there was no anticipation: in other words, voter turnout did not change based on anticipation of the change in Medicaid administrative burden. Given that the outcome variable of voter turnout rate can take on any value between zero and one hundred, it is quasi-continuous and I employ a linear model estimation.
I utilize two different two-way fixed-effects estimators. The first is the classic difference-in-differences estimator extended to more than two time periods. There is a burgeoning literature on the limitations and potential biases of the classic estimator’s use in two-way fixed-effects models, and some contributors offer new estimators that may be applied in settings with staggered treatment timing, heterogeneous treatment effects, continuous treatments, dynamic effects, and other situations (e.g., Callaway and Sant’Anna Reference Callaway and Sant’Anna2021; de Chaisemartin and D’Haultfœuille Reference de Chaisemartin and D’Haultfœuille2020; Reference de Chaisemartin and D’Haultfœuille2024; Goodman-Bacon Reference Goodman-Bacon2021; Imai and Kim Reference Imai and Kim2021; Sun and Abraham Reference Sun and Abraham2021). Among the assumptions already stated above, the classic estimator assumes treatment-effect homogeneity, which means the effect of the treatment is the same across units and time periods. Table A6 in the online appendix provides evidence that the treatment effects are likely homogeneous by testing the relationship between the residualized outcome and the residualized treatment. I additionally utilize the de Chaisemartin and D’Haultfœuille (Reference de Chaisemartin and D’Haultfœuille2024) estimator that provides an unbiased estimate of the treatment effect even in cases where the treatment is nonbinary, where it may increase or decrease multiple times, and when there are heterogeneous treatment effects. In the results that follow, I present both the classic and the de Chaisemartin and D’Haultfœuille (Reference de Chaisemartin and D’Haultfœuille2024) estimates.
Below is the two-way fixed-effects equation:
$$ Turnou{t}_{cs t}={\displaystyle \begin{array}{l}\tau AdministrativeBurde{n}_{st}+{\gamma}_{cs}\\ {}+\hskip2px {\theta}_t+\alpha {\boldsymbol{X}}_{cs t}+\beta {\boldsymbol{X}}_{st}+{\epsilon}_{cs t}\end{array}} $$
where
$ Turnou{t}_{cst} $
is voter turnout in county
$ c $
within state
$ s $
in year t;Footnote
11
$ AdministrativeBurde{n}_{st} $
is the administrative burden index of Medicaid in a given state
$ s $
and year
$ t $
, and takes on a value between zero and one;
$ {\gamma}_{cs} $
are county fixed effects and
$ {\theta}_t $
are year fixed effects; and
$ {\boldsymbol{X}}_{cst} $
are time-varying county-level controls for demographics and politics. County-level demographic covariates include the percentage of non-Hispanic white individuals, the percentage with a high-school degree or less, the percentage aged 65+, and log median income.Footnote
12 County-level political covariates include log voting-age population, Democratic vote share in the previous presidential election,Footnote
13 and whether the county had a high-potential-eligibility population for Medicaid and its interaction with state expansion status.Footnote
14
$ {\boldsymbol{X}}_{st} $
are time-varying state-level covariates, including Medicaid expansion status;Footnote
15 the two indices of electoral administrative burden (registration and turnout); and, to capture election competitiveness, swing-state status in previous presidential elections and, in midterm years, dummies for the Senate race in the state, the gubernatorial race, and their interaction. These midterm-year competitiveness indicators are set at a value of zero for presidential years regardless of the presence of a Senate or gubernatorial race in a state, so these variables also control for any unmeasured differences between midterm and presidential elections more broadly. With the exception of the two indices of electoral burden, the covariates chosen are the same as those in Clinton and Sances (Reference Clinton and Sances2018) and account for many demographic, ideological, and electoral factors that are likely related to turnout rate. I cluster standard errors at the county level to account for correlation between counties within the same state.
Results
Mass-Level Effects
I estimate the effect of Medicaid administrative burden on voter turnout in national elections using the classic two-way fixed-effects estimator. In table 2, I first examine the effect of Medicaid administrative burden on county-level voter turnout without controlling for election administration burden, shown in columns 1 and 3. Columns 1 and 2 include only the covariates shown plus county and year fixed effects; columns 3 and 4 include all covariates discussed in the previous section.Footnote 16 As demonstrated, results are very consistent across model specifications, regardless of whether measures of election administration burden or other covariates are included.
Impact of Medicaid Administrative Burden on Voter Turnout in National Elections (2010–20) among Border Counties
Note: Classic two-way linear fixed-effects estimation with fixed effects for county and year. Ninety-five percent confidence intervals clustered by county. Analyses are among counties within 100 miles of a state border. Models 1 and 2 include only fixed effects. Models 3 and 4 also control for the following: county-level covariates: percentage non-Hispanic white individuals, percentage with a high-school degree or less, percentage aged 65+, log median income, log voting-age population, Democratic vote share in previous election, and whether the county had a population with high potential eligibility for Medicaid and its interaction with state expansion status. State-level covariates: swing-state status in previous presidential election, and, in midterm years, dummies for the Senate race in the state, gubernatorial race, and their interaction. ** p < 0.01; *** p < 0.001.
I find support for my hypothesis that higher administrative burden has a negative influence on voter turnout. All else equal, these results (column 4) indicate that a county subject to a Medicaid administrative burden level at the 75th percentile (an index value of 0.63) observed a turnout rate that was 0.54 percentage points lower than turnout in a county at the 25th percentile (index value of 0.33). This 0.3-point difference in the index is equal to reducing burden by eliminating nine of the 31 burdensome measures included in the index. Likewise, the effect of a county having the highest Medicaid burden level (index value of 0.77) versus the lowest burden level (index value of 0.13) was a decline of 1.14 percentage points in voter turnout. This effect is net of Medicaid expansion status and both indices of voter registration and turnout burden. These results are meaningful given how close recent national elections have been, and given that many strong demographic and ideological predictors are included in these models in addition to county and year fixed effects.
Turning to the relationships between the administrative burden imposed by voter registration and turnout and the county-level turnout rate, I observe negative relationships as expected:Footnote 17 higher voter registration and turnout burden was strongly negatively associated with turnout rate. However, the turnout burden index was a stronger predictor of turnout rate than the registration burden index: about twice the magnitude. Consistent with Clinton and Sances (Reference Clinton and Sances2018), I find that Medicaid expansion is positively associated with turnout. Counties in states that expanded Medicaid had a higher turnout rate of 0.70 percentage points on average compared with similar nonexpansion counties. This is comparable to the Medicaid administrative burden effect size of 0.54 percentage points for a 0.3-point difference in the burden index. I interpret the comparable Medicaid expansion and burden effect sizes to mean that the potential benefits a community may gain from having wider access to Medicaid (through expansion) are effectively cancelled out if their Medicaid program is administered in a burdensome way.
Finally, to examine the robustness of my results, I utilize the de Chaisemartin and D’Haultfœuille (Reference de Chaisemartin and D’Haultfœuille2024) differences-in-differences estimator that is robust to heterogeneous treatment effects and handles nonbinary treatments that can increase or decrease multiple times, such as Medicaid administrative burden. In figure 6, I present these coefficient estimates compared with the classic two-way fixed-effects estimator. All models are based on counties within 100 miles of their state’s border. The de Chaisemartin and D’Haultfœuille (Reference de Chaisemartin and D’Haultfœuille2024) estimates are slightly smaller than the classic two-way fixed-effects estimates. Taken together, these estimates demonstrate great consistency in the direction and significance of the effect of Medicaid administrative burden on voter turnout in recent national elections. Though she conducts a descriptive examination of Medicaid recipiency rather than administration, Michener (Reference Michener2018, 77–78) finds that being a Medicaid recipient is associated with an about five percentage-point decrease in the likelihood of voting; comparatively, my effect sizes are smaller, though I am examining county-level voting rates rather than the voter turnout of Medicaid recipients.
Effect of Medicaid Administrative Burden on Voter Turnout across Two-Way Fixed-Effects Estimators
Note: 95% confidence intervals clustered by county shown. Analyses are among counties within 100 miles of the state border. “Burden + expansion only” models include the Medicaid burden index, Medicaid expansion status, registration burden index, and turnout burden index, plus county and year fixed effects. “With covariates” includes all covariates listed in table 2’s notes.
Individual-Level Patterns
In the preceding analysis of mass-level political effects, I cannot separate out the effect on county subgroups, such as those eligible for Medicaid and its recipients, former recipients, recipients’ close contacts, or nonrecipients. This is due to significant data challenges. Using survey data could potentially categorize respondents as members of one of these categories, but there is significant underreporting and undercounting of means-tested program receipt on surveys (Klerman, Ringel, and Roth Reference Klerman, Ringel and Roth2005).Footnote 18 Additionally, there is significant overreporting of voter turnout on surveys (e.g., Bernstein, Chadha, and Montjoy Reference Bernstein, Chadha and Montjoy2001; McDonald Reference McDonald2003). Finally, Medicaid eligibility varies by state and may be based on income level, assets, household size, parental status, and other factors; this makes it hard to know or model eligibility based on survey questions.
Fortunately, there are some existing surveys that verify self-reported voting history (such as the Cooperative Congressional Election Study [CCES] and the American National Election Studies), but they do not typically include questions about Medicaid recipiency, a respondent’s relationship to Medicaid recipients, or the details needed to accurately estimate their Medicaid eligibility. However, because the mechanisms of how administrative burden shapes voter turnout operate primarily at the individual level, I present correlational analyses that utilize the 2016 cross-section of the CCES as a complement to the preceding mass-level analyses. Though imperfect, the 2016 CCES contains questions on individual demographics, a voting record for 2016 that has been validated by state administrative records, and an indicator of if respondents report receiving health insurance “through a government program, such as Medicare or Medicaid” (Ansolabehere and Schaffner Reference Ansolabehere and Schaffner2017).
I subset the CCES sample into two groups. The first group contains respondents who say they have government health insurance and are less than 65 years old, because nearly all adults aged 65 and older receive Medicare, which provides health insurance for senior citizens and some disabled adults, and it is not possible to know if adults in this age group are receiving Medicare alone or a combination of Medicare and Medicaid. I call the resulting group “Medicaid recipients.” The second group consists of all other adults aged 18 to 64 (i.e., those who say they do not receive government health insurance, or “nonrecipients”). Controlling for demographic characteristics known to be related to the likelihood of voting, table 3 presents the relationship between administrative burden and turnout among these groups as well as among the full sample (to mirror the preceding mass-level analyses). The models employ logistic regression predicting individual-level turnout, presenting log odds for easier comparison with the sign of coefficients in table 2.
Medicaid Administrative Burden and Individual-Level Voter Turnout in 2016
Source: 2016 CCES for individual-level variables.
Note: Logistic regression with fixed effects for US Census Bureau divisions, weighted with postelection validated voter survey weight. Ninety-five percent confidence intervals clustered by state. High Medicaid enrollment is equal to one if greater than or equal to median estimated population share on Medicaid from American Community Survey self-reports and zero if not. Medicaid burden index is the 2016 level relative to 2012, the prior presidential election year. Analyses among adults aged 18–64. White non-Hispanic is omitted as a race/ethnicity category. Results among the total sample of adults (aged 18+) are similar and presented in table A10 in the online appendix. a p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
I find that there is a negative relationship between a state’s Medicaid administrative burden level and an individual’s likelihood of voting in the 2016 election in states with a higher than median estimated Medicaid recipient population share, in the total sample (column 1), among self-reported Medicaid recipients (column 2), and among nonrecipients (column 3). This relationship is stronger among Medicaid recipients compared with nonrecipients. These results are suggestive evidence that there is not a purely ecological relationship between administrative burden and voter turnout.
Discussion and Conclusion
To what extent do the intricacies of policy administration shape political participation? Given the variation in policy administration that may exist across states, over time, by program or policy, and across individuals, this is a difficult question to answer. I argue that government programs with high visibility and traceability to the government and which meet basic needs may create participatory feedback effects on both recipients and nonrecipients. Using my original measures of Medicaid administrative burden and election administration burden, I examine their effect on voter turnout. I leverage the repercussions of the ACA, whose provisions required all states to lower enrollment and renewal burdens and, in practice, gave states the option to expand their eligible Medicaid population. I find that, net of Medicaid expansion status and electoral administrative burden, greater Medicaid burden induced a small but significant decline in voter turnout. In the US, national elections are often very close; therefore, this difference could change electoral outcomes. Given that those most likely to be significantly touched by Medicaid administrative burden are the same low-income citizens who vote at low rates, it is not surprising to find small but significant county-level turnout effects.
Though the mass-level analyses are unable to separate out the effect on potential Medicaid recipients versus all others, I leverage the voter-validated CCES to examine if the burden–turnout relationship holds among self-reported Medicaid recipients and nonrecipients, considered separately. I find that there is a negative relationship between a state’s Medicaid administrative burden level and the individual-level likelihood of voting in states with a large estimated Medicaid population share among all groups examined, which suggests that there is more than an ecological relationship between burden and voting. Future research could design a survey that includes questions about Medicaid utilization by self and family (such as those in Kaiser Family Foundation surveys), political self-efficacy, and attitudes about the government; validates voter records; and collects detailed demographic data to validate and model Medicaid eligibility. Combined with my state-year measures of Medicaid and electoral burden, such data would permit examination of the burden–participation relationship and its mechanisms among subgroups of the population.
Finally, another direction for future work could consider how administrative burden affects other forms of political participation. While studying voter turnout is an important first step in the study of political participation due to its pervasiveness and relative accessibility (e.g., Verba, Schlozman, and Brady Reference Verba, Schlozman and Brady1995) as well as its fundamental role in a representative democracy, future studies of administrative burden could examine a variety of forms of participation. This research would likely necessitate conducting surveys or in-depth interviews to measure an array of participatory forms.
The relationship between the administrative state and the public is an important and growing area of scholarship. Many individuals interact frequently and intensely with the administrative state; therefore, government administration matters. Especially for people relying on means-tested programs, “the experience of the state is the experience of burdens” (Herd et al. Reference Herd, Hoynes, Michener and Moynihan2023, 11). I found evidence that the administrative burden of Medicaid shapes mass-level political participation, even while controlling for many factors known to influence the likelihood of voting. Election administration burden is also negatively associated with voter turnout. By accounting for the impact of the administrative burden of elections on political participation as this article has done, policy feedback scholars will be able to more precisely isolate the participatory effects of the burden associated with other government programs. Even more critically, these findings imply that if we measure the intricacies of the ways in which a policy or program is administered, scholars and practitioners can anticipate whether a program’s administration is likely to create positive or negative feedback effects on political participation.
Supplementary material
To view supplementary material for this article, please visit http://doi.org/10.1017/S1537592725103277.
Data Replication
Data replication sets are available in Harvard Dataverse at: https://doi.org/10.7910/DVN/TWWMXE.
Acknowledgments
For feedback on earlier iterations of this article, the author is grateful to the editors, the four anonymous reviewers, Daniel Carpenter, Jennifer Hochschild, Sarah James, Elizabeth Thom, participants in the APSA 2024 Annual Meeting, and National Poverty Fellowship colleagues at the Institute for Research on Poverty at the University of Wisconsin–Madison. The author also acknowledges Joshua Clinton and Michael Sances, and, separately, Ashley Fox, who provided replication datasets at critical moments in the project.
Open access
