The geographic concentration of disadvantage is a key mechanism of inequity. In the United States, the spatial patterning of disadvantage renders it more than the sum of its individual parts and disproportionately harms economically and racially marginalized Americans. This article focuses specifically on the political effects of Medicaid beneficiaries being concentrated in particular locales. After offering a framework for conceptualizing the community-wide consequences of such policy concentration, I analyze aggregate multiyear data to examine the effect of Medicaid density on county-level voter turnout and local organizational strength. I find that, as the proportion of county residents enrolled in Medicaid increases, the prevalence of civic and political membership associations declines and aggregate rates of voting decrease. These results suggest that, if grassroots political action is to be part of a strategy to achieve health equity, policy makers and local organizations must make efforts to counteract the sometimes demobilizing “place-based” political effects of “people-based” policies such as Medicaid.
“Place matters” in profound, multitudinous ways and it is acutely consequential for those who inhabit the economic and racial margins of American society (Dreier, Mollenkopf, and Swanstrom 2004). The power of place is neither incidental nor innocuous. Instead, the social, economic, and political significance of where a person lives stems from public policies that create, contour, and reinforce systemic inequity. One way that policy does this is by facilitating the geographic concentration of people who are structurally vulnerable. Concentration is a mechanism through which place-based detriments are distributed (Dreier, Mollenkopf, and Swanstrom 2004; Jargowsky 2015; Massey and Denton 1993; Rothstein 2014; Sharkey 2013; Wilson 1987). The density of disadvantage renders it more than the sum of its individual parts. A person who is poor and living in a community disproportionately populated by other people who are poor will suffer the deleterious consequences of poverty more severely than a similarly indigent person residing in an affluent area (Chetty, Hendren, and Katz 2016; Kneebone and Nadeau 2015). An analogous logic applies for race (Massey and Denton 1993; Sharkey 2013; Wilson 1987).
Evidence of this abounds in the domain of health. Racial and economic health disparities in the United States are powerfully linked to geographic context (Acevedo-Garcia and Lochner 2003; Auchincloss and Hadden 2002; Cattell 2001; Do et al. 2008; Grady 2006; LaVeist 1989; Marcus et al. 2016; Yen and Kaplan 1999). Counties, cities, neighborhoods, and states influence access to resources and demarcate exposure to risks. Public policies that concentrate disadvantage in particular ways are central features of the processes that link place to health (Ludwig et al. 2013). Advancing health equity will therefore require crafting, passing, and implementing policies that offset the penalties of concentrated disadvantage. Generating political demand for such policy should involve mobilization within the communities with the most at stake (Bambra, Fox, and Scott-Samuel 2005; LaVeist 1992; Schroeder 2007). However, concentrated disadvantage itself may influence the capacity for such communities to galvanize. Figure 1 illustrates the interrelated sociopolitical processes that connect policy, place, and political power.
Most of the pathways outlined in this conceptual model have been subject to empirical scrutiny (see relevant citations throughout this article). Social scientists of all stripes have authored voluminous explanations of the public policy roots of concentrated disadvantage (A). Sociologists have provided accounts of the relationship between concentrated disadvantage and health disparities (B). Political scientists have produced illuminating research about the democratic repercussions of public policy (E) and the effects of health for (individual-level) political participation (D). What remains underexplored is systematic study of how concentrated disadvantage structures political participation (C).1 This article analyzes one facet of that topic by focusing on disadvantage channeled through policy concentration. After clarifying the importance of policy concentration and its relationship to both health equity and political behavior, I analyze aggregate multiyear data to determine how the concentration of Medicaid beneficiaries affects county-level political outcomes. I find that, as the proportion of county residents enrolled in Medicaid increases, civic and political membership associations decline, as do aggregate rates of voting. To the extent that political participation is imperative for achieving health equity, these findings suggest that we must pay close attention to the ways that “people-based” policies such as Medicaid (i.e., those targeted toward individuals with little regard for where they live) have “place-based” social and political effects.
Concentrated Disadvantage, Health Equity, and Community Participation
The concentration of disadvantage occurs when people who lack significant resources and/or bear substantial burdens cluster together in particular communities.2 Such clustering enables “serial patterns of social contact and exposure that become crucial factors for how people construct interpretations of social reality” (Young 2003: 1073). This has ramifications for health equity. Health disparities are larger and more consequential in places marked by racial and economic segregation, even net of individual-level factors (Acevedo-Garcia and Lochner 2003; Cattell 2001; Do et al. 2008; Grady 2006; Marcus et al. 2016; Yen and Kaplan 1999). Though the full range of mechanisms that account for this has not been definitively established, there is little doubt that the geographic concentration of indigent people aggravates health disparities. From birth weight to diabetes to mortality, concentrated disadvantage drives divergences in health between whites and people of color as well as between affluent and poor Americans (Finch et al. 2010; Gaskin et al. 2014; Grady 2006; Marcus et al. 2016).
The prospects for progress on this front depend on cultivating the political capacity necessary to pursue policies that promote health equity. This is because “the biggest gains in population health will come from attention to the less well off,” but policy change that benefits this group may be less likely to occur, “unless they have a political voice” (Schroeder 2007: 1226). Though it is certainly no panacea, political involvement from the bottom up holds promise for improving health equity. Even if the political engagement of those who bear the brunt of health inequities has only a limited direct influence on policy at the national or state levels, local political participation itself engenders collective efficacy, empowerment, and social capital, all of which are associated with improved health outcomes (Browning and Cagney 2002; Kawachi, Venkata Subramanian, and Kim 2008; LaVeist 1992; Ohmer 2007; Szreter and Woolcock 2004; Wallerstein 1992). Moreover, civically and politically organized communities attract health resources and disseminate health information more effectively than their less engaged counterparts (Viswanath, Randolph Steele, and Finnegan 2006). Overall, political participation is an important tool for improving health.
Still, participation is no easy fix. There are many reasons that low-income Americans are less likely to take political action: personal, institutional, and contextual barriers hinder their ability and motivation to participate, making it less likely that they will “go to” politics (Alex-Assensoh 1997; Cohen and Dawson 1993; Hahn 2009; Pacheco and Fletcher 2015; Rosenstone and Hansen 1993; Schlozman, Verba and Brady 2012; Verba, Schlozman and Brady 1995). At the same time, the distance and disinvestment of mobilizing institutions makes it unlikely that politics will “come to” them (Michener 2016; Rosenstone and Hansen 1993). Public policy plays a notable role in these processes (and is also the result of them; fig. 1, path E) by shaping the dispositions and behavior of those who experience the benefits/burdens of government programs (Campbell 2012; Lerman and Weaver 2014; Mettler 2005; Michener, forthcoming; Soss 1999). One often-overlooked mechanism through which policy can operate is a phenomenon I call policy concentration (Campbell 2012: 340).
Policy concentration is a form of concentrated disadvantage that happens when particular geographic locales have disproportionate numbers of residents affected by a given policy. Of course, the distribution of policy across the population is not arbitrary. Even when policy is people-based, it is constrained and contoured by preexisting structural arrangements. Because of economic and racial segregation, low-income policy targets are often sequestered in particular counties, census tracts, or neighborhoods. Policy benefits/burdens are thus heavily concentrated in these communities. In an uncommon example of research that attends to this, Traci Burch (2013) shows that concentrated patterns of incarceration create communities from which residents (overwhelmingly African American and Latino) are disproportionately removed and imprisoned. Burch focuses on North Carolina and Georgia, where residents of disadvantaged neighborhoods experience imprisonment at ten times and fourteen times the national average, respectively (Burch 2013: 5). She finds that carceral concentration of this sort “diminishes the ability of all neighborhood residents to participate in politics,” and she points to “social dynamics and economic resources” as the most likely mechanisms driving these effects (Burch 2013: 6, 9). In this article, I investigate the political upshot of Medicaid concentration and offer a framework of policy contact to explain why the local density of Medicaid beneficiaries affects the political life of entire communities.
Medicaid Policy Concentration
With more than 70 million enrollees, Medicaid is the largest public source of health coverage in the United States and the leading insurer for low-income Americans. It is the third most costly domestic program in the federal budget (following Social Security and Medicare), and the biggest source of federal revenue in state budgets (Paradise 2015; Rudowitz, Snyder, and Smith 2015). Recognizing Medicaid's immense significance, scholars have studied it closely, uncovering evidence that it has effects on outcomes ranging from mortality to mental health to educational achievement (Baicker 2013; Cohodes et al. 2016; Sommers, Baicker and Epstein 2012). Few studies have investigated whether and how the effects of Medicaid extend beyond individuals to communities. There is good reason to do so.
Geography is a key basis of heterogeneity in the distribution of Medicaid benefits. Due to the powerful institution of American federalism, states have been afforded immense discretion in fashioning the contours of Medicaid policy (Lukens 2014; Michener, forthcoming; Sparer 1996). By deciding the scope of eligibility, states exercise control over how large the program grows and what populations it covers (Andrews 2014; Sparer 1996). For example, in 2014, 54 percent of non-elderly Americans below 100 percent of the Federal Poverty Level (FPL) received health care coverage through Medicaid. But this proportion varied sizably across states: only 29 percent of North Dakotans below the FPL were covered, while 72 percent of West Virginians were.3 Similarly, in Massachusetts Medicaid covered 68 percent of families without any full-time or part-time workers compared to only 30 percent in Virginia.4 Parallel patterns hold along the lines of race and ethnicity: in Virginia, only 17 percent of blacks and 20 percent of Latinos were covered by Medicaid, while Iowa covered 53 and 41 percent, respectively.5 These differences underscore the extent to which state decisions about policy design produce uneven concentrations of Medicaid beneficiaries across the country.
Comparable patterns exist for counties. Take California, for example, where county-level proportions of (adult) beneficiaries range from a low of 6 percent (Placer County) to a high of 31 percent (Modoc County). Some of this variation is straightforwardly explained by spatial differences in poverty and race, but much is not. For example, four very poor California counties include Kings, Fresno, Madera, and Tulare. These counties are situated in close proximity to one another. As shown in table 1, they have high poverty rates (between 21 and 26 percent), similar racial demographics, and comparable rates of English language non-proficiency. However, only 10 percent of adult residents in Kings County are enrolled in Medicaid, versus 20 percent in both Tulare and Fresno and 29 percent in Madera.
There are many reasons for these differences. California devolves responsibility for the administration of Medicaid to counties, so some of the differences in enrollment likely stem from heterogeneous approaches to poverty governance across counties (Sharp 2012; Soss, Fording, and Schram 2011). Since a confluence of policy choices, administrative decisions, and demographic configurations generate county-level variation in Medicaid density, these patterns are not a simple function of race and/or poverty. Evincing this fact: across all US counties, the bivariate correlation between the percent of adult Medicaid beneficiaries and the percent of African Americans is only 0.17, while the bivariate correlation with the percent of people living in poverty is 0.45. A basic regression model (not shown) predicting the percent of Medicaid beneficiaries in a county with controls for the percent of residents who are black, Latino, and below poverty (respectively), explains only 23 percent of the total variation. Insofar as the density of beneficiaries is (related to but) distinct from configurations of racially or economically marginalized populations, then Medicaid policy concentration may have a unique influence on local political behavior. Below, I submit that postulate to empirical scrutiny.
The Medicaid-to-Politics Link
The literature on “policy feedback” already provides reasons to believe that Medicaid is germane to political behavior. This body of work establishes that policies are not just an output of the political process but are also a critical input, structuring the relations among political institutions, government elites, and mass publics (Campbell 2012; Mettler and Soss 2004; Pierson 1993; Schneider and Ingram 1993; Skocpol 1992). Some of the most seminal work in this vein has shown that cash assistance programs, GI benefits, Social Security, and criminal justice policies influence civic and political participation by channeling resources, generating interests, and shaping interpretive schemas (Bruch, Marx Ferree, and Soss 2010; Campbell 2003; Lerman and Weaver 2014; Mettler 2005; Soss 2000).
Extending this literature, several recent studies have investigated how Medicaid affects political participation. Michener (forthcoming) shows that Medicaid beneficiaries are significantly less likely to register, vote, and take other kinds of political action. Importantly, this work demonstrates that the strength and direction of this individual-level relationship varies geographically: it is most pronounced in states that have recently reduced benefits and it is reversed in states that have recently expanded benefits. Complementing these findings, several other studies have identified an increased likelihood in voting at the district (Haselswerdt 2017) and county levels (Clinton and Sances 2017) in the wake of state Medicaid expansions spurred by the 2010 Patient Protection and Affordable Care Act.
Extending the work described above, I consider the theoretical relevance of Medicaid's concentration. I advance a simple but important hypothesis: Medicaid policy concentration has community-wide political effects. To clarify the logic underlying this claim, I offer a framework for understanding the channels through which Medicaid density might affect the political life of communities. I call this the policy contact framework (PCF) and describe it in detail below.
The Policy Contact Framework (PCF)
As shown by Joe Soss in his work on cash assistance programs, interpretive learning processes are a primary mechanism for policy feedback effects (Soss 1999, 2000). More precisely, public policies convey messages to beneficiaries that “teach” them about their political status and shape their political behavior. In this article, I emphasize that such messages are not limited to actual policy beneficiaries; they also educate those who encounter policy as a result of living alongside beneficiaries. Soss and Schram (2007: 122) make a related assertion, noting that, “participant status” does not define the scope of policy feedback effects. They astutely aver that, “participant status is only a particular form of a more general phenomenon: the experience of public policy as a visible and directly consequential factor in one's life” (Soss and Schram 2007: 122). I extend and build on this claim by focusing specifically on the role of place, and highlighting the mechanism of policy concentration as a means of making policy a “directly consequential” feature of community life. Policies can teach entire communities about government and politics. Quite crucially, however, the reach and content of such lessons hinges on how community members are positioned vis-à-vis two axes of contact with policy: (1) direct/indirect, and (2) personal/impersonal.6Table 2 summarizes the combinations of these forms of policy contact.
Direct and personal contact (box A, table 2) is what policy beneficiaries experience firsthand as they interact with government. This is the kind of contact implicitly theorized in most of the policy feedback literature. As far as Medicaid goes, such contact affects the political behavior of beneficiaries by teaching them contextually varied (but often negative) lessons about the capriciousness and (in)capacity of national and state government (Michener, forthcoming). Notice that the guardians of beneficiaries are also included in box A. This incorporates the parents of low-income or medically needy children and the adult caretakers of elderly or severely disabled persons. Although such folks are not typically a part of the policy feedback story, their experiences with government programs on behalf of the people they love can be transformative (Campbell 2014; Levitsky 2014). For example, sociologist Sandra Levitsky (2014) offers a compelling qualitative description of the distinct attitudinal feedback effects of Medicaid on middle-class caretakers who turn to the program to meet the prohibitively expensive long-term care needs of their elderly dependents. These caretakers tend to judge Medicaid by the expectations set via their experiences with (non-means tested) social insurance policies such as Social Security. In such a light, they usually find the program wanting and develop sharp criticisms, not only of Medicaid but also of related political phenomena (such as immigration and anti-poverty policies). The pertinent insight from Levitsky's work is that for Medicaid, policy feedback extends to non-beneficiary guardians who have direct personal contact with policy. Since such guardians are often part of the same communities as the beneficiaries they assist, they also increase in density as Medicaid beneficiaries become more locally concentrated.
Further expanding the scope of contact, the policy contact framework (PCF) next considers those who have indirect personal contact with Medicaid (box B). This includes friends and family of policy beneficiaries who do not have involvement as guardians (and thus do not directly interface with policies on behalf of beneficiaries), but who nonetheless have opportunities to hear about and observe the benefits and burdens that Medicaid brings. The political relevance of indirect personal contact is underscored by the recent work of Hannah Walker (2014: 811) who finds that “proximal contact” with the carceral state via someone connected to the criminal “justice” system has spillover political effects on broader groups of individuals who exist on that system's periphery. In the carceral realm, this includes folks like the aunt of someone who is incarcerated, the mother of a crime victim or the friend of a person who is unlawfully stopped and searched (Miller 2008). In the domain of Medicaid, analogous examples might include the parents of adult children who rely on Medicaid or the (ineligible) spouse of an actual beneficiary. Such people experience personal (i.e., connected to a loved one) but indirect contact with Medicaid that may shape their views of policy and politics. For example, in the 2015 Kaiser Family Foundation Survey, 27 percent of respondents report having been covered by Medicaid at some point in their lives, and 37 percent report having friends or family who have been covered (Norton, DiJulio, and Brodie 2015). Especially critical is that those with secondhand knowledge of Medicaid have distinct policy attitudes relative to those with no personal connection to the program. As shown in fig. 2, among survey respondents who noted having some connection to Medicaid (via family or friends), 70 percent agreed that the program was very important and 42 percent thought that spending should be increased (compared to 51 percent and 28 percent, respectively, for those with no personal connection). Crucially, vicarious connections are more likely in places with high Medicaid density and are thus one channel through which Medicaid concentration might come to influence local political life.
Next, the PCF further parses types of contact by pointing to a group of people with impersonal direct ties to Medicaid: those whose employment brings them face to face with policy beneficiaries (box C). This includes (but is not limited to) bureaucrats working in local Medicaid offices, nurses staffing community health clinics, and home health aides taking on grueling and underpaid labor to provide medically needy beneficiaries with vital services. Though these “working class” constituents are generally beyond the purview of policy feedback studies, they are squarely within the reach of policy itself. As Celeste Watkins-Hayes (2009) shows, the people who do the work of implementing public policy are pushed to engage the political and social contexts that generate those policies. Moreover, in the course of fulfilling their duties, welfare bureaucrats exchange political ideas and draw on political beliefs that are reflective of their own race, class, and place-based identities (Watkins-Hayes 2009). The density of beneficiaries dictates the extent and nature of direct impersonal ties. As a result, Medicaid concentration may shape the political perceptions of the working classes that undergird the diffuse administrative apparatus of the program.
The final form of policy contact that I highlight is indirect and impersonal (box D). This includes local community members who have heard about Medicaid but have not dealt with the program or its beneficiaries in either a direct or personal way. Though they may seem too far removed to comment upon, people who fall into this category have numerous opportunities for meaningful policy contact, especially in a context of concentration. They may be connected to those who do have direct or personal contact (e.g., my brother's friend has a child that is on Medicaid). They may also be exposed to local messages targeted to beneficiaries (e.g., signs outside of health clinics urging eligible persons to sign up for benefits). More generally, since Medicaid finances many local services, the various institutions that rely on such funding may actively seek to shape the ideas of community members. Children's hospitals, medical providers, and other local organizations are potentially crucial intermediaries. For example, Texas Children's Hospital hosts a blog created specifically for the 12,000 “team members” it employs. This blog features regular pieces about health policy issues. On July 30, 2015, Mark Wallace, the president and CEO of the hospital, penned a post entitled, “Medicaid: Safety Net and Stepping Stone.”7 Wallace began by emphasizing how many people relied on Medicaid in the state of Texas and he attempted to convince his employees of the value of Medicaid to children and families. He said things like, “For me, regardless of my own political beliefs . . . I think about all the children and families who need our help,” and “The simple truth is our federal and state governments save money by investing in health care for our children.” At the end of the post, Wallace encouraged readers to contact their local political officials. As an elite positioned on the supply side of the health economy, Wallace used his platform to reach thousands of employees. Many of these people may not have had direct contact with Medicaid beneficiaries (either personal or impersonal), but they are likely to live in communities within driving distance of the hospital, and this is one example of how indirect and impersonal contact can teach key lessons about Medicaid.
A distinct but comparable process may occur when community members become the targets of local political officials. Budgetary politics in Rockland County, New York, illustrate this possibility especially well. In 2012, members of the Rockland County legislature voted for a bill that mandated a transparency measure requiring the inclusion of a separate line item for the “Medicaid tax” portion of local tax bills (previously, there was a single item called “county tax”). This was intended to raise public awareness (and ostensibly ire) of the significant financial costs of Medicaid. Each year since the enactment of this legislation, tens of thousands of Rockland residents have received tax bills highlighting the “burden” that Medicaid places on them. Politicians leverage such communications to gain support from constituents. For example, a (now former) Rockland County legislative representative, Barry Kantrowitz, emphasized the encumbrance of Medicaid in a salient post on his website. As shown in the screen shot included in fig. 3, after noting that recent tax bills “included a breakout of the Medicaid expense as a line item,” Kantrowitz devoted a section of his site to “tackling Medicaid fraud,” pointing to his “efforts to prevent the abuse of Medicaid and other public assistance,” and initiating provocative language about punishing “criminals.” In this way, Kantrowitz sent a stigmatizing message to his constituents linking Medicaid to criminality. In view of such moves, opponents of the dual line policy now (quite reasonably) argue that it creates “resentment among our heavily burdened taxpayers about Medicaid.”8
Though it is difficult to know how idiosyncratic the politics of Medicaid in Rockland County really are, they do not appear to be singular. A simple Google search turns up an example of another county in New York (Monroe) that devotes a separate line for Medicaid on local tax bills (see fig. 4). Moreover, since counties all across the United States have to contend with Medicaid costs to varying degrees, local officials have an incentive to mobilize constituents in opposition to the program, particularly in the context of policy concentration.
In sum, the policy contact framework helps to explain how Medicaid concentration can affect community political participation. For three of the four types of policy contact (direct personal, indirect personal, and direct impersonal), I draw on existing social science research to build the prima facie case for the expectation that contact will bear on political behavior (references included in table 2). The fourth type of contact (indirect impersonal) is not explored in scholarly research, but it is credible given examples such as that of Texas Children's Hospital and Rockland County, New York.
By delineating this policy contact typology, I provide conceptual support for the general hypothesis that Medicaid policy concentration has community-wide political effects. Recall that the mechanism I propose to explain this effect is an interpretive learning process by which experiences with policy confer lessons about government and politics. The PCF points to learning opportunities that extend beyond policy beneficiaries to the various groups of people living alongside them.
To be clear, the PCF details a wide range of relationships and social dynamics that I cannot fully explore in this article. It is an initial springboard for theorizing concentration effects, but it is presently quite indefinite. Notice that I do not hypothesize about the direction of behavioral effects across contact types. On the one hand, social science research does not yet offer the theoretical tools necessary for understanding the range of ways that policy concentration will affect community members with different degrees of personal/direct contact. On the other hand, such effects are likely conditional and thus not easily pinpointed. For example, whether impersonal direct contact will mobilize, demobilize, or have no effect at all will depend on the nature of such contact (e.g., as a school nurse who sees students perform better when they have health coverage or as an overworked bureaucrat in an agency unsustainably serving more and more people). Similarly, while Hannah Walker (2014) shows that proximate contact (which I classify as personal and indirect) mobilizes, she also finds that the strength of this effect varies across racial groups. Though the same may hold true for Medicaid, it is also possible that indirect contact demobilizes in the domain of health where policy may be injurious, but it is not so punitive as to stoke a deep sense of social (in)justice.
All of this is to say that while the PCF serves as a basis for more precise hypothesizing in future research, this article cannot closely trace all of the paths that it illuminates. In fact, the aggregate data that I draw on does not allow me to distinguish types of policy contact. So, while the PCF establishes a basis for understanding how policy concentration might affect communities (i.e., mechanisms), the main task of this article is to garner initial evidence that it does so at all (i.e., effects).
A Focus on Counties
To decipher the effect of Medicaid density on local political participation, I examine county-level patterns. Aggregate county data provide ample geographic variation as well as some variation across time. Such data are sufficiently granular that they approximate (albeit imperfectly) community processes. At the same time, because counties are large and bear a significant responsibility for administering social programs, the government collects data on county-level Medicaid enrollment (this is not the case for neighborhoods or census tracts). For these reasons, county data are the best practical choice.
Counties also have more substantive significance. They represent “loci of government” capacity with measurable consequences for poverty, inequality, and governance (Benton 2002; Lobao et al. 2012; Sharp 2012). Counties are often the most proximate sites at which Medicaid benefits are administered. Though there is variation in the capacity of counties to make choices about the social programs they administer, the local caseworkers that they employ, train, and manage retain a tremendous amount of discretion (Sharp 2012: 31; Soss, Fording, and Schram 2011; Watkins-Hayes 2009). In addition, county Medicaid offices are sometimes located in visible places, where beneficiaries and non-beneficiaries alike can detect upsurges in enrollment and can develop ideas about which populations Medicaid serves. Finally, counties hold large enough numbers of people that they can facilitate not only the strong ties that characterize relationships with those most proximate to us, but also the weak ties that are useful for generating social and political capital (Granovetter 1973; Small 2009). Evidence suggests that counties are an important source of networks and social capital (Dillion 2011; Rupasingha, Goetz, and Freshwater 2006). This makes them a suitable context given the wide range of social relationships implicated in the PCF.
Data and Variables
The ensuing analyses are based on two key dependent variables. The first measures aggregate rates of voting in US counties during the 2000, 2004, and 2008 presidential elections. These data are drawn from CQ Press Voting and Elections Collection. Voting is a political outcome of central importance and it is frequently studied by social scientists. It is a basic act of democratic citizenship and is thus essential for the purpose of evaluating the political wherewithal of individuals and communities.
Still, voting is only one among a number of forms of participation. The second dependent variable measures another: the number of civic and political organizations in a given county each year between 2005 and 2010. These data come from the Census Bureau's County Business Patterns (CBP) Survey.9 The associational life of communities is a vital marker of residents' capacity to mobilize politically (Kawachi, Venkata Subramanian, and Kim 2008; Putnam 2000).
Note that due to the time periods covered by the dependent variables, this research does not speak to the effect of the recent Medicaid expansion that has happened in the wake of the ACA. So, in contrast to Haselswerdt (2017) and Clinton and Sances (2017), I investigate the effect of Medicaid concentration, not expansion. These are quite distinct, as the former emphasizes the relevance of the distribution of Medicaid beneficiaries while the latter attends more generally to broadened access to the program. Naturally, the two are connected. Medicaid enrollment grows after an expansion, which affects patterns of concentration. Nevertheless, Medicaid concentration is a separate phenomenon that stems from a contextually contingent amalgam of race, class, and poverty governance. Admittedly, by investigating a period that does not include the most recent expansion of Medicaid, I lose the ability to examine how the political effects of concentration vary across different expansion/access regimes. I acknowledge that scope condition. One consolation is that the post-expansion period has been idiosyncratically marked by a large influx of new beneficiaries, ever-changing policy contours (with proliferating Section 1115 waivers), and a bitterly contentious national political scene. Interpreting patterns observed during this period would be challenging. Focusing on a pre-expansion time frame allows us to understand what happens during more stable times, when beneficiaries, bureaucrats, and policy makers are settled into comparatively routine and predictable patterns. Still, when data are available, scholars should also study the post-expansion period to assess whether and how the findings from the analyses below are altered by expansion.
The primary independent variable in the models gauges Medicaid policy concentration by measuring the percent of county residents younger than 18 years old enrolled in Medicaid in a given year. These data come from the American Community Survey (ACS) and the Medicaid Statistical Information System (MSIS).10 Since the percent of Medicaid beneficiaries under 18 years old is higher than the percent of adult beneficiaries, this variable represents the high bar for capturing how many people in a county have direct contact (even if on behalf of their children) with Medicaid. Moreover, in counties where high proportions of children are enrolled, community members who may never enroll in Medicaid (e.g., school administrators, nurses, etc.) can still encounter the program through exposure to its most common subpopulation: children. As suggested by the PCF, these direct impersonal contacts may be one mechanism through which program density has the potential to influence non-beneficiaries.
In addition to the variables described above, the analyses include controls for important time-varying demographic factors including population size, race (percent black and percent Hispanic), poverty, and income.
Analyses and Findings
The data used here contain information on over 2,600 counties. The voting models span three presidential elections (2000, 2004, and 2008), and the associational models track the presence of membership associations between 2005 and 2010. The unit of analysis is the county-year. Given relatively short time frames across many counties (small T, large N), I estimated panel regression models with county and year fixed effects (Cameron and Trivedi 2009). The first model predicts voting rates using an OLS estimator. The second model predicts counts of associations using a Poisson estimator.11 All models are based on cluster robust standard errors to account for heteroscedasticity and serial correlation (Cameron and Miller 2011).
Table 3 contains the main results. Model 1 provides evidence that child Medicaid density has a negative relationship to aggregate voting rates, and model 2 demonstrates a similarly negative correlation with the presence of civic and political membership associations in a county. Substantively, the significant negative coefficient on the Medicaid density variable in model 1 indicates that, for a given county, a 1 percent increase in child Medicaid enrollment (over time) is associated with a 7 percent decrease in county voting rates and an 11 percent decrease in the mean count of civic and political associations ([e−.119] −1 = −.11). These substantive effects are graphically depicted in figs. 5 and 6. The results support the hypothesis that Medicaid policy concentration is consequential for the political life of communities. Between 2000 and 2010, increasing rates of Medicaid enrollment had a dampening effect on local political participation.
Robustness of Findings
These findings warrant further empirical investigation. Four potential inferential challenges are especially obvious: (1) omitted variable bias, (2) conditional effects, (3) ecological fallacy, and (4) reverse causation. I'll discuss each in turn.
Though the use of fixed effects regression attenuates some concerns about omitted variable bias, the possibility of unaccounted for time-varying confounders remains. I cannot resolve this using observational data. However, I find reassuring indications in the results of models that I estimated including controls for unemployment, health (the incidence of premature death), and county-level political culture (percent voting Republican). Controlling for such factors does not alter the core findings.
One especially worrisome confounder is poverty. If Medicaid concentration is largely a stand-in for poverty (the bivariate correlation between these variables is 0.62), then its significance simply reflects the relationship between socioeconomic status and political participation. I take several steps to account for this possibility. First, I control for the percent of people who are below the poverty line in each county (see table 3). Doing so does not eliminate the statistical or substantive significance of the relationship between Medicaid concentration and either of the political outcomes. In fact, the substantive results are similar with or without controlling for poverty.
Making further efforts to address the entanglement of poverty and Medicaid concentration, I subset the data to include only low-poverty counties. The average poverty rate across all of the counties in the data set was 14 percent and the maximum was 54 percent. There were roughly 990 low-poverty counties (where fewer than 10 percent of residents were below the poverty line). Despite the comparatively minimal prevalence of poverty in these counties, there was still variation in child Medicaid enrollment. If confounded processes unique to poor counties are driving my results, then Medicaid concentration should not affect political participation in low-poverty counties. The subsetted models suggest the opposite (table 4, columns 1–2). The relationship between Medicaid density and political outcomes is significant and substantial in magnitude when the analyses are limited to places that are economically healthy relative to the rest of the country.
Also relevant is whether the effect of Medicaid concentration is conditional on other factors. Most prominent among potential interactions are race, rural status, and the foreign-born composition of the county. Levels of Medicaid enrollment signal different things in different places, and the lens through which community members interpret the policy landscape may depend on the characteristics of the beneficiaries and the community members. In small rural towns where people know and trust one another, Medicaid may not carry the same stigma and Medicaid density may not exert the same dampening effect on participation (Michener forthcoming). Contrastingly, in places where a significant percent of the population is black or foreign-born, Medicaid density may prove especially demobilizing.
Also worth considering is the level of access that county residents have to doctors. Perhaps in counties where there is a shortage of doctors, Medicaid beneficiaries have negative (and potentially demobilizing) experiences stemming from their inability to access care. Further still, is the possibility that in these places non-beneficiaries also lack access, making it more likely that they will view Medicaid beneficiaries negatively or blame them for overburdening the health care system. Either way, doctor shortages could be an additional factor conditioning the effect of Medicaid concentration.
An analysis of interactions suggests that some of these factors are important, but not all. The (negative) correlation between Medicaid concentration and county-level voting rates increases (i.e., there is a positive and significant interaction) as the black share of the population increases (see table 4, column 3). The opposite holds for rural status: the relationship between voting and Medicaid concentration is attenuated in counties that are designated as rural (a negative and significant interaction). However, the effects of Medicaid concentration are not conditional on the proportion of county residents who are foreign-born or on whether a county is designated as a primary care health professional shortage area (insignificant interactions not shown). Furthermore, none of the conditional effects hold for the associational models.
The findings presented above are based on aggregate analyses of county-level data and thus do not give us much leverage for making inferences about the political behavior of individuals. Yet, the mechanisms by which county voting rates and associational activity decline in response to Medicaid density necessarily involve individual political action. Prior studies referenced earlier consider the influence of Medicaid enrollment on the political behavior of individuals (Michener, forthcoming), but not the influence of Medicaid density (a contextual factor) on such behavior. Evidence that Medicaid density has an influence on individual-level political outcomes (net of the influence of individual-level Medicaid enrollment) would support the argument that the effects of policy concentration are distinct.
To assess individual-level patterns, I turn to the third wave of the Fragile Families and Child Well-Being Study (FFS). This survey follows a cohort of nearly 5,000 children born in US cities, interviewing both mothers and fathers around the time of their child's birth and at various intervals subsequently. Since the sample was composed to reflect non-marital births in large US cities, it is not nationally representative in the traditional sense. However, the emphasis on “fragile families” means that FFS contains an unusually large number of poor persons. These data are uniquely well suited for studying low-income populations (Bruch, Marx Ferree, and Soss 2010). Questions about political activity were only asked in the third wave (2001/2004), so that cross-section is employed here. There were 7,529 respondents interviewed in wave three (with an average response rate of 77 percent). Among this number, 53 percent (just under 4,000 persons) reported that either they or their children were enrolled in Medicaid, providing uncommon empirical leverage for understanding patterns of political behavior.12
To generate a Medicaid density variable, I aggregated at the census tract level (the lowest level for which FFS has a geographic identifier). This is not ideal, because there are 2,100 census tracts and the number of respondents in many tracts is quite small. This limitation notwithstanding, I estimate a set of models aimed at gauging whether tract-level Medicaid density is significantly correlated with political and civic outcomes, net of numerous other factors.
To confront some of the challenges that accompany cross-sectional observational analyses, I use robust standard errors clustered at the census tract level. I also use a quasi-experimental matching technique to ensure that the effect of individual-level Medicaid enrollment is accounted for as best as possible (Iacus et al. 2012).13 Matching addresses the confounding influence of factors causally prior to the primary independent variable by better aligning the distributions of observed covariates in the treatment group (Medicaid beneficiaries) with those in the control group (non-beneficiaries). This technique (imperfectly) approximates the counterfactual condition where those being studied are similar in every way except for Medicaid participation. Estimates produced by using matched data are less sensitive to model specification, less biased, and more efficient (Ho et al. 2007).14
Using data pre-processed via the matching procedure, I estimate voter registration and civic attitudes.15 Drawing on insights from the political participation literature, I control for key individual characteristics including age, education, employment, income, sex, race, nativity, cash welfare receipt, and church attendance (Rosenstone and Hansen 1993; Verba et al. 1995). I also include a control for self-rated health and controls at the census tract level for race (percent black and Hispanic, respectively) and poverty (percent below the poverty line). The models predict each political outcome as a function of individual-level Medicaid enrollment, census tract level Medicaid density and all of the controls. The voter turnout and registration models were estimated via probit and the broader participation model (which includes being part of a political group and attending a rally) via ordinary least squares.
As shown in table 5, Medicaid density is significantly and negatively correlated with voting and political participation. These results should be interpreted cautiously. They are based on self-reported voting and cross-sectional analysis, which comes with a host of inferential challenges that matching and clustered errors do not solve. Medicaid enrollment at the individual level and Medicaid density are highly correlated (α = .71), so the non-significance of the individual-level Medicaid variable in this model should not be taken to mean that the individual experiences do not matter, especially given the low numbers of Medicaid beneficiaries in some tracts. The larger point here is to provide prima facie evidence that individual-level patterns correspond to aggregate outcomes observed in the county-level data. To that end, it is especially instructive to observe that the relationships are negative, and that they apply to outcomes that map on to those measured in the aggregate analyses.
Another important consideration is reverse or reciprocal causation. Research suggests that civically and politically active communities attract greater resources and do a better job of disseminating health information (Viswanath, Randolph Steele, and Finnegan 2006). If such communities facilitate better health, then residents may have lower demand for Medicaid. In essence, politically engaged counties may have a smaller proportion of residents enrolled in Medicaid because the benefits of living in such places dampen the need for Medicaid.
There are several reasons why this causal interpretation is not convincing. Even in communities that attract health resources (clinics, hospitals), health insurance remains indispensable and such resources should enable residents to have greater access to health coverage (health-related institutions often assist with enrollment in Medicaid). In addition, communities with the means for disseminating health information are also best equipped to spread the word about Medicaid to uninsured or needy residents (Viswanath, Randolph Steele, and Finnegan 2006). By this logic, the benefits of political participation should increase Medicaid enrollment in civic-minded communities (ceteris paribus). If politically engaged communities influence Medicaid enrollment rates, they should do so in a positive direction. The finding of a negative relationship between Medicaid density and political participation suggests that different processes are at work.
The 1970s and 1980s ushered in a shift of national anti-poverty policy away from place-based strategies targeting disadvantaged locales and toward people-based policies that provide resources to individuals without regard to where they live (Kincaid 1999, 2012; O'Connor 1999; Partridge and Rickman 2006; Rich 1993). Despite retrenchment on the front of traditional cash assistance (Aid to Families with Dependent Children/Temporary Assistance for Needy Families), in-kind, people-based policies such as the Women, Infants and Children Nutrition Program (WIC) and the Supplemental Nutrition Assistance Program (SNAP) have enrolled growing numbers of individuals since the 1970s (Moffitt 2015). Moreover, tax-based policies such as the Earned Income Tax Credit (EITC) have also expanded substantially during this time. Finally, Medicaid has grown exponentially, emerging (both fiscally and in terms of enrollment) as chief among people-based, means-tested policies. Most crucially, during the very same time period that Medicaid and the other programs cited above have been swelling, streams of funds dedicated to the needs of local communities have dried up. For instance, in 1978 approximately 15 percent of city revenues came from federal aid; today that number ranges from 1 to 3 percent (Kincaid 1999). Between 1980 and 1990, the Reagan and Bush administrations slashed grants to cities by 46 percent (Katz 1995). Community Development Block Grants were cut by 25 percent, urban jobs programs disappeared and General Revenue Sharing, which had previously provided localities with some funding, was eliminated (Katz 1995; Kincaid 2011). Altogether, federal grants-in-aid to state and local governments for resources distributed directly to people have increased markedly since the mid-1970s, while outlays for programs aimed at places have plummeted (Kincaid 2012).
By offering evidence that Medicaid density is associated with county-level political demobilization, this research underscores the risks of such an exclusively people-based approach to reducing health disparities and other inequities. If politically empowering communities with vulnerable health care constituents facilitates achieving health equity (as I argue it does), then policies that offer people access to health resources must not simultaneously weaken the political strength of their communities. Instead, the distinct geography of disadvantage in the United States necessitates that policies are responsive to both people and places.
This is possible if we complement Medicaid's people-based design with additional resources directed to the most needy communities. Specific efforts might include increasing the funding for and capacity of Community Health Centers, subsidizing community-based health initiatives, and offering greater support for civic and political organizations in distressed communities.16 This also provides a reason to consider the kinds of direct action that government can take to empower the beneficiaries of people-based policies. Though the feasibility of this recommendation is highly contingent on political climate, one promising route would be for the federal government and states to ensure full cooperation with the National Voter Registration Act, particularly Section 7. This statute mandates that public assistance agencies offer every applicant and enrollee an opportunity to register to vote. Recent research indicates that compliance with Section 7 is lax in many states, especially those with growing numbers of black Americans (Michener 2016). Shoring up this law and applying it more equitably is a step toward offsetting the potentially demobilizing community-level repercussions of people-based policies such as Medicaid. Similarly, any local political activity that reinforces the political incorporation of social policy beneficiaries may be useful in this regard (e.g., organizing political meetings or voter registration drives at local health facilities).
Additionally, foundations with resources to fund grassroots organizations (and those organizations themselves) should be mindful of these patterns when deciding which communities to serve, and should consider cultivating political capital in places where community-wide experiences with policy might otherwise erode it. Policy makers, community planners, and other key players must be cognizant of how places shape the political impact of social policy. Attentiveness to collective political capital should be part of policy discourse and should inform strategies to alleviate health (and other) disparities.
Finally, local political elites who sometimes use Medicaid and other public assistance programs as fodder for their own political ends should closely consider the potential democratic consequences of their actions. Stigmatizing Medicaid (especially in places where many people rely on it) has downstream consequences for how beneficiaries, their family members, and their neighbors think about and engage in politics. While we still have much to learn about the processes through which this happens, this work suggests that the local-level understandings of such programs may have significant stakes. Those who set the terms of political discourse must attend to this possibility.
I would like to thank the participants of the Robert Wood Johnson Investigators Awards Annual Meeting (September 2016), as well as participants in the City and Regional Planning graduate seminar (November 2016) for their invaluable feedback. I am particularly indebted to Suzanne Mettler, Julia Lynch, Leslie Hinkson, Keith Wailoo, and the editors of this issue.
2. Sociologists have overwhelmingly focused on the concentration of poverty (Jargowsky 2015; Kneebone and Nadeau 2015; Wilson 1987). Yet, concentrated disadvantage is a broader concept because it attends to the clustering of aspects of economic vulnerability beyond poverty such as joblessness and “female-headed” households. Although such factors are correlated with poverty, they are distributed across geographies in different ways (Krivo et al. 1998; Massey and Shibuya 1995).
3. Kaiser Family Foundation State Health Facts: Medicaid Coverage Rates for the Non-Elderly by Federal Poverty Level: http://kff.org/medicaid/state-indicator/rate-by-fpl-3/.
4. Kaiser Family Foundation State Health Facts: Medicaid Coverage Rates for the Non-Elderly by Family Work Status: http://kff.org/medicaid/state-indicator/rate-by-employment-status-3/.
5. Kaiser Family Foundation State Health Facts: Medicaid Coverage Rates for the Non-Elderly by Race and Ethnicity: http://kff.org/medicaid/state-indicator/rate-by-raceethnicity-3/.
6. After conceiving of this framework, I came across a distinct but related framework by Soss and Schram (2007). Soss and Schram focus on the dimensions of visibility and proximity, which do not overlap with the dimensions that I present here, though there are logical links between the two. The key difference is that I developed the policy contact framework described in this article with an eye toward understanding how policy affects geographically defined communities, while Soss and Schram focus much more broadly on how policy affects mass publics.
8. See “Sticker Shock with 2016 Rockland County Tax Bill Wallop.” Available online at: www.lohud.com/story/money/personal-finance/taxes/david-mckay-wilson/2016/01/21/sticker-shock-2016-rockland-tax-bill-wallop/79073846/.
9. Counts of civic and political associations were generated via the NAICS codes 813410 and 813940. These codes reflect the CBP data counts of the number of civic and political establishments (i.e., single physical locations) per county.
10. The actual Medicaid data used in the article were calculated by Sarah Miller (an economist at the University of Michigan), who generously shared her county-level Medicaid calculations with me. Miller mapped PUMA-level Medicaid enrollment rates from the ACS to the county level. For years prior to 2008, when ACS data were not available, Miller used state-level Medicaid enrollment data from MSIS to extrapolate county-level rates.
11. There are two relevant estimation techniques that I did not pursue. First, I considered using a negative binomial estimator for model 2 because the variance of outcome (count of associations) was large relative to its mean. However, the negative binomial model would not converge despite numerous attempts. Moreover, though Cameron and Trivedi (2009) suggest that Negative Binomial models can lead to improved efficiency, the Poisson panel estimators have the benefit of relying on weaker distributional assumptions, and may therefore be more robust given the use of cluster-robust standard errors (p. 641). A second issue is that over 400 counties were dropped from the Poisson model because they did not have any civic or political associations over the time period in question. To address this, I considered estimating a zero-inflated Poisson model. However, such models are inappropriate because they require that certain cases never be at risk of an event occurring.
12. Notably, Medicaid beneficiaries in the FFS sample are not reflective of the Medicaid population nationally; FFS beneficiaries are younger, healthier, more likely to be male, and less likely to live below the poverty line.
13. Coarsened exact matching is a multi-step process: (1) temporarily coarsen X (i.e., recode it to assume fewer values), (2) perform exact matching on the coarsened X by sorting observations into strata with unique values of C(X), (3) eliminate any stratum missing treatment or control variables, (4) pass on original uncoarsened variables, except those omitted as per step #3, and (5) analyze original data using stratum derived from step #2 as weights in the analysis.
14. The variables used in the matching procedure were: age, sex, TANF participation, education, race (black), marital status, and household income. I selected covariates likely to influence both the treatment and outcomes, while omitting variables that could be affected by the treatment (e.g., health), and thus induce post-treatment bias.
15. Civic attitudes were measured using a scale combining respondents' ratings of the importance of five activities: voting, serving in the military, jury duty, volunteering, and reporting a crime. Relatively few respondents were part of a community group or political group, so I opted to use an attitudinal measure of civic capacity, which is a likely mechanism for the associational outcome observed in the aggregate data.
16. For an example of what the CDC under the Obama administration has done on this: www.cdc.gov/nccdphp/dch/programs/partnershipstoimprovecommunityhealth/index.html.