More Than Words? How Highlighting Target Populations Affects Public Opinion about the Medicaid Program

Abstract

Context: The Medicaid program provides health insurance coverage to a diverse set of demographics. We know little about how the policy community describes these populations (e.g., on Medicaid-related websites or in public opinion polls and policy writings) or whether and how these descriptions may affect perceptions of the program, its beneficiaries, and potential policy changes.

Methods: To investigate this issue, we developed and fielded a nationally representative survey of 2,680 Americans that included an experiment for priming respondents by highlighting different combinations of target populations of the Medicaid program as found in the Medicaid policy discourse.

Findings: Overall, we find that Americans view Medicaid and its beneficiaries rather favorably. However, there are marked differences based on partisanship and racial animosity. Emphasizing citizenship and residency requirements at times improved these perceptions.

Conclusions: Racial perceptions and partisanship are important correlates in Americans’ views about Medicaid and its beneficiaries. However, perceptions are not immutable. In general, the policy community should shift toward using more comprehensive descriptions of the Medicaid population that go beyond the focus on low income and that include citizenship and residency requirements. Future research should expand this work by studying descriptions in the broader public discourse.

The Medicaid program—established under Title XIX of the Social Security Act in 1965 as a public health insurance program for those eligible for public assistance from the Aid to the Blind, Aid to Families with Dependent Children, and Aid to the Permanently and Totally Disabled programs—has grown substantially in size and complexity since its inception (Engel 2006; Olson 2010; Stevens and Stevens 2003). Although Medicaid was initially only envisioned as an adjunct to welfare for those already receiving public assistance (Herz 2010), repeated expansions in benefits and eligibility quickly added to its enrollment and significance (Brown and Sparer 2003). The program's expansion under the Affordable Care Act (ACA), albeit reined in by the US Supreme Court, further increased Medicaid's importance for health care providers and beneficiaries (Callaghan and Jacobs 2017; Haeder and Weimer 2015b; Jacobs and Callaghan 2013). Moreover, while the program saw reductions in enrollment that exceeded 2.4 million under the Trump administration (Artiga and Pham 2019), mainly as a result of increases in administrative burdens (Haeder, Sylvester, and Callaghan 2021a; Ku and Brantley 2017), the COVID-19 pandemic has highlighted the program's role in the American safety net (Frenier, Nikpay, and Golberstein 2020; Schubel 2020). Indeed, because of the prolonged public health emergency caused by COVID-19, the Medicaid program grew to an unprecedented 90 million beneficiaries by July 2022 (Tolbert and Ammula 2023).

With successive expansions since 1965, the Medicaid program has become increasingly complex, and its reach into the American population continues to evolve. Indeed, the existence of more than 60 distinct Medicaid eligibility groups indicates the program's diversity (CMS 2019). Moreover, unlike Medicare, which is administered at the federal level, Medicaid is jointly administered by state and national governments. This arrangement has led to substantial diversity in program eligibility, benefit design, and even the actual naming of the respective Medicaid programs across states, further adding complexity (Haeder and Weimer 2015a; USDHHS 2021). As a result, Americans, who are already subjected to a highly confusing amalgam of a health care system (Kindig, Panzer, and Nielsen-Bohlman 2004), may be particularly confounded by the complexity of Medicaid (Norton, DiJulio, and Brodie 2015).

This inherent complexity is exemplified by the diverse descriptions of target populations used by those in the Medicaid policy community who are working in state Medicaid agencies, think tanks, and academia, among others who are, by definition, highly knowledgeable about the program. Even a cursory reading of policy documents about the Medicaid program highlights how substantially descriptions differ in the information conveyed to readers. Indeed, some of these descriptions tell very little, such as the one by the Pennsylvania Department of Human Services, which states that Medicaid “pays for health care services for eligible individuals” (PDHS 2022). No additional information about eligibility groups or criteria is provided. Other descriptions go a step further to note the importance of income criteria; but even here, there are differences in the details presented. Whereas one description states that “Medicaid is the nation's main public health insurance program for low-income people” (emphasis added) (KCMU 2013), another one provides more detail beyond the program's income requirement: “Medicaid is a public insurance program that provides health coverage to low-income families and individuals” (emphasis added), but then goes on to describe the specific groups eligible for coverage, such as “children, parents, pregnant women, seniors, and people with disabilities” (CBPP 2020). A few others note the citizenship/permanent residency requirements. These include limited descriptions like “eligibility for all [New Mexico] Medicaid programs requires that individuals meet certain federal guidelines. These include citizenship, residency and income requirements” (emphasis added) (NMHSD 2022) as well as more detailed ones like: “Many groups of people are covered by Medicaid. Even within these groups, different requirements must be met. These may include your age; whether you are pregnant, age 65 or older, blind or disabled; your income and assets; and whether you are a US citizen or a qualified immigrant” (emphasis added) (KDHE 2022).

How the Medicaid program is framed in the discourse within the aforementioned policy community and which target populations are highlighted may lead to significant consequences for how the program, its beneficiaries, and potential Medicaid policy changes are perceived. Specifically, whether a description's highlighted target populations of the program include those groups who are generally considered to be most deserving, such as the aged, disabled, or pregnant women (Haeder, Sylvester, and Callaghan 2021a), or whether a description includes citizenship or residency requirements, may carry substantial implications for perceptions of individuals who come into contact with these program descriptions (Ingram and Schneider 1993; Schneider and Ingram 1997, 2005). Importantly, these descriptions may shape perceptions of the American public if they make their way into the broader public discourse outside the Medicaid policy community. This may hold particularly true for the Medicaid program, which has been shown to be highly confusing for Americans (Call et al. 2013). Equally important, public opinion scholars have long struggled with appropriately introducing Medicaid to potential survey respondents without overly and unduly priming them, which could bias their responses and ultimately the researchers’ findings about Medicaid (AARP 2011; CNN 2011; CR/RI 1981; Haeder, Sylvester, and Callaghan 2021a; HSPH/NYT/CF 2019; Jacobs and Mettler 2011; KFF 2020; Pascale, Roemer, and Resnick 2009; Wu 2021). This carries additional weight and importance for governmental surveys, and most importantly for the US census and the Current Population Survey (Davern et al. 2009). Of course, how the program is described may also impose psychological costs on (potential) beneficiaries, and perceived or real stigma may discourage them from enrollment (Herd and Moynihan 2018). Despite these potentially significant implications, the impact of highlighting different target populations remains largely unexplored empirically.

To assess whether Medicaid program descriptions commonly used in policy discourse have the potential to affect public opinion, we first reviewed the academic and policy literature about the Medicaid program, federal and state Medicaid websites, and public opinion polls to understand how the Medicaid policy community commonly presents the Medicaid program. We then developed and fielded a nationally representative survey of Americans. The survey included an experiment based on descriptions from the Medicaid policy discourse that primed respondents by highlighting specific Medicaid sets of target populations as they are found in descriptions of the Medicaid program used by the policy community. Because of the large number of potential beneficiary groups (CMS 2019), we restricted our analyses to several commonly presented combinations. Our analysis focused on respondents’ perceptions of the Medicaid program, perceptions of Medicaid beneficiaries, their willingness to enroll in Medicaid, and lastly, perceptions of four proposed policy changes to the Medicaid program, including block-granting the program, implementing new premiums and out-of-pocket costs, implementing wellness incentives, and implementing work requirements. We conclude with a discussion of our findings and the potential limitations of our analysis.

Theoretical Expectations

As noted above, today's state Medicaid programs have evolved into a smorgasbord of eligibility criteria, starting with the inclusion of those eligible for to receive assistance from the Old Age Assistance, Aid to the Blind, Aid to Families with Dependent Children, and Aid to the Permanently and Totally Disabled programs (Engel 2006; Gilman 1998). Over the years, expansions have occurred numerous times, most prominently for such groups as children (Engel 2006; Grogan and Patashnik 2003a; Iglehart and Sommers 2015), pregnant women (Howell 2001; Piper, Mitchel, and Ray 1994), and the aged, blind, and disabled (Grogan and Park 2018; Grogan and Patashnik 2003a; Thompson 2015). However, today the detailed list of eligible groups also includes “Deemed Newborns,” “Individuals Who Are Essential Spouses,” “Disabled Widows and Widowers Ineligible for SSI due to Increase in OASDI,” “Independent Foster Care Adolescents,” “Individuals Receiving Home and Community Based Services under Institutional Rules,” “Individuals with Tuberculosis,” “Certain Women Needing Treatment for Breast or Cervical Cancer,” “ Medically Needy Blind or Disabled Individuals Eligible in 1973,” “Ticket to Work Basic Group,” and “Individuals Participating in a PACE Program under Institutional Rules,” to name but a few (CMS 2019).

Of course, the diversity of the Medicaid program has been apparent from its very inception (Thompson 1981), as has the diversity in how states use Section 1115 waivers (Thompson 2012) and managed care products (Pauly and Grannemann 2010). Many states have even created state-specific names for their Medicaid programs, such as DenaliCare, Healthy Families, HuskyHealth, and Apple Care, potentially causing confusion among beneficiaries or the broader public (Call, Davern, and Blewett 2007; Draper, Hurley, and Short 2004; Eberly, Pohl, and Davis 2005; Grogan and Park 2017a; Kincheloe et al. 2006). There are reasons to believe that the potential for confusion only increased under the Affordable Care Act's efforts to expand and nationalize the Medicaid program (Haeder, Sylvester, and Callaghan 2021a; Norton, DiJulio, and Brodie 2015). Some states’ subsequent resistance to program expansion, and the resulting uneven implementation of the ACA Medicaid expansion, served as one of the potential sources of additional perplexity (Callaghan and Jacobs 2017; Haeder and Weimer 2015b; Jacobs and Callaghan 2013). Moreover, some states, such as Arkansas, expanded their Medicaid via the so-called private option, allowing newly eligible individuals to purchase plans on the ACA Marketplace (Allison 2014). Finally, research has also documented constant bewilderment among Americans about the differences between the Medicare and Medicaid programs (Pascale 2004).

The confusion surrounding Medicaid is further exemplified by the findings of a large number of studies that have established that Medicaid beneficiaries often struggle to identify themselves as such. Studies have found misidentification rates of 11%–43% (Call et al. 2013). A particularly prominent series of studies were conducted under the auspices of the Medicaid Undercount Experiment in Minnesota, which identified a substantial undercount of Medicaid beneficiaries (Austin 2008). Similar undercount rates were found when the experiment was extended to other states, including California, Florida, and Pennsylvania (Austin 2008; Call et al. 2008a, 2008b). The problem also extends to government surveys. For example, a detailed analysis of data from the Current Population Survey indicated that 43% of Medicaid enrollees “answer[ed] the survey as though they were not enrolled and 17 percent report being uninsured” (Davern et al. 2009), while a study in Maryland indicated an undercount in the state's Medicaid enrollment of 34%–38% (Eberly, Pohl, and Davis 2005).

Additionally, the extensive research on health literacy and health insurance literacy point toward many challenges for Americans more generally related to health care and health insurance. Given the complexity of the US health care system, it is not surprising that overall health literacy is relatively low (Long and Goin 2014). Indeed, only 12% of Americans are fully health literate (Kutner et al. 2006). A 2004 study by the Institute of Medicine estimated that 90 million Americans are not sufficiently proficient to understand and act on health information (Kindig, Panzer, and Nielsen-Bohlman 2004). Undeniably, the US health insurance system is equally confusing, as most Americans lack confidence in correctly identifying even basic insurance concepts (Blumberg et al. 2014; Long and Goin 2014). A 2013 survey of insured Americans found that only 14% could answer four simple multiple-choice questions about their insurance (Loewenstein et al. 2013). Past research has shown convincingly that individuals of lower socioeconomic status, those with lower levels of education, older individuals, the uninsured, and those of minority descent fare disproportionately worse in this regard (Bartholomae et al. 2016; Brown et al. 2004; Kim, Braun, and Williams 2013; Loewenstein et al. 2013; Martin and Parker 2011; McCormack et al. 2009; Tennyson 2011). Ultimately, most individuals in the United States are unfamiliar with the workings of the health care system and have trouble understanding their insurance status and coverage. These issues lead to challenges navigating their health care and greater susceptibility to differences in program framing, altering enrollment and policy attitudes.

More broadly, the confusing nature of the Medicaid program and health care opens the door for potentially influencing individuals through priming and thus through survey experiments. The term priming, at times interchangeably used with “framing” or “heresthetics,” refers to the process where individuals’ attention is focused on a particular aspect of an issue (Iyengar and Kinder 1987; Lenz 2009; Riker 1986). This specific focus then triggers a reaction by the individual, often emotionally (Kennedy-Hendricks, McGinty, and Barry 2016), that initiates a cognitive shift in how individuals perceive an issue (Mummolo and Hopkins 2017). There is evidence that priming efforts may shape public opinion on policy issues (Chong and Druckman 2007; Schattschneider [1960] 1975), including health policy issues like the Affordable Care Act and Medicaid (Barry et al. 2018; Bergan and Risner 2012; Fowler et al. 2017; Haeder 2020; Hopkins 2017). There is also evidence that the effects of priming and framing may elicit policy feedback effects (Jacobs and Mettler 2018), thus adding additional importance to a proper assessment of how Medicaid framing influences public attitudes.

Given this significant diversity, the question emerges of how the Medicaid program's target populations are presented in the policy discourse. We turned to the websites of state Medicaid agencies for all 50 states and of federal entities such as the Centers for Medicare and Medicaid Services (Medicaid.gov) and HealthCare.gov as well as public opinion polls featured on Roper iPoll, to determine how the Medicaid policy community describes the program's target populations.¹

First, we found that several descriptions were relatively bland and offered few indications of which groups may qualify for coverage (see table 1 for examples).² For instance, in Colorado, the website hosted by the Colorado Department of Health Care Policy and Financing refers to Medicaid as “public health insurance for Coloradans who qualify” (CDHCPF 2022). Similarly, in Pennsylvania, the Pennsylvania Department of Human Services notes that “Medical Assistance (MA), also known as Medicaid, pays for health care services for eligible individuals” (PDHS 2022). In contrast, in Virginia, the Virginia Department of Medical Assistance Services notes that the Medicaid “programs pay medical service providers for medical services rendered to eligible individuals” (VDMAS 2020). Lastly, the West Virginia Bureau for Medical Services states that Medicaid provides “coverage of medical services for specific groups of citizens” (WVDHHR 2022).

Second, we found that other descriptions focused on the generally low-income nature of the target population. For example, the Arizona Department of Economic Security notes that “Medicaid is a state medical assistance program for low-income individuals and families” (ADES 2022). In contrast, the Hawaii Department of Human Services states that Medicaid “provides eligible low-income adults and children access to health and medical coverage through managed care plans” (HDHS 2022). Lastly, the South Carolina Department of Health and Human Services described Medicaid as a “program by which the federal and state governments share the cost of providing medical care for needy persons with low income” (SCDHHS 2022).

The focus on the low-income aspect of the Medicaid program is unsurprising because the primary pathway into today's Medicaid program is through low income, and because the Medicaid program was developed as a direct successor to local and state-based medical assistance programs for the poor (Engel 2006; Grogan and Park 2018; Grogan and Patashnik 2003a; Haeder 2019a; Haeder, Sylvester, and Callaghan 2021a; Iglehart and Sommers 2015; Thompson 2015). However, the emphasis on “low income” in descriptions carries with it some potentially negative implications (Gilens 2009; Haeder, Sylvester, and Callaghan 2021a; Stern 1946; Terris 1951). More generally, the prevalent American ideology has long looked suspiciously on individuals living in poverty and has sought to assign responsibility for poverty to poor individuals’ unwillingness to improve their lot in life. Indeed, the credo of personal responsibility (Gilens 2009) and the aspirations of the “American Dream” (Cullen 2004) have played important roles in American political discourse. Notably, much of the negative public sentiment associated with low-income assistance programs are driven by whether Americans think of a specific program as “welfare” or not (Gilens 2009). As respondents make the welfare connection, we expect that priming respondents to think of Medicaid beneficiaries as “low-income individuals and families” in the program description will generally have a negative effect on respondents.

Third, some program descriptions went a step further to provide additional details on eligible groups beyond their low-income status and included several groups commonly considered to be “deserving,” such as children, pregnant women, and the disabled (Engel 2006; Grogan and Park 2018; Grogan and Patashnik 2003a; Haeder, Sylvester, and Callaghan 2021a; Iglehart and Sommers 2015; Thompson 2015). For example, the Kansas Department of Health and Environment listed eligible groups as “children, pregnant women, people with disabilities, the aged, and the elderly” (KDHE 2022), whereas the Kentucky Department for Medicaid Services refers to “eligible low-income residents including children, families, pregnant women, the aged and the disabled” (KDMS 2022). Lastly, the Minnesota Department of Human Services notes that Medicaid in the state “serves children and families, pregnant women, adults without children, seniors and people who are blind or have a disability” (MDHS 2022). Notably, we found that generally, these groups were all prominently listed in combination and virtually always included the modifier “low-income.”

It is worth reemphasizing that these groups—children and pregnant women (Engel 2006; Grogan and Patashnik 2003a; Iglehart and Sommers 2015) and the aged, blind, and disabled (Grogan and Park 2018; Grogan and Patashnik 2003a; Thompson 2015)—have been positively constructed (Grogan and Patashnik 2003b). We thus expect that the inclusion of groups historically considered to be “deserving” of policy benefits—that is, children, pregnant women, the aged, blind people, and disabled people—in Medicaid program descriptions should moderate the potential negative attitudes surrounding benefits given to “low-income individuals and families” (Orentlicher 2015; Schneider and Ingram 1997).

Fourth, we found that a small subset of descriptions highlighted that eligibility generally required US citizenship or permanent residency in addition to the previously described target groups. For example, Medicaid.gov noted that “Medicaid . . . provides health coverage to over 72.5 million Americans, including children, pregnant women, parents, seniors, and individuals with disabilities. . . . [Applicants] must be either citizens of the United States or certain qualified non-citizens, such as lawful permanent residents” (CMS 2022). In Illinois, the Department of Human Services website notes that Medicaid programs “are designed to provide Illinois’ residents access to quality health care” and that “to qualify for medical assistance a person must meet financial eligibility criteria, residency requirements and in most cases must be citizens” (Illinois DHS 2022). Lastly, Iowa's Department of Human Services states that “Medicaid is a health insurance program based on income. . . . The following are some of these general requirements: a child under the age of 21, a parent living with a child under the age of 18, a woman who is pregnant, a woman in need of treatment for breast or cervical cancer, a person who is elderly (age 65 or older), a person who is disabled according to Social Security standards, an adult between the ages of 19 and 64 and whose income is at or below 133 percent of the Federal Poverty Level (FPL), a person who is a resident of Iowa and a US citizen” (Iowa DHS 2022).

The rare nature of these modifiers is of particular scholarly interest because the politics of Medicaid and other public assistance programs has often focused on questions of citizenship as a determining factor of deservingness, particularly since the 1990s (Ku and Pervez 2010; Stone 2002; Wong 2010). Importantly, American public policy has historically drawn strong lines of demarcation between individuals inside and outside their communities, with implications for policy benefits (Stone 2002; Wong 2010). Within many government programs, there have long been restrictions preventing immigrants, particularly those who entered the country illegally, from benefiting from public programs (Kullgren 2003). Indeed, policy developments under the Clinton and George W. Bush administrations have made it harder for immigrants, including those in the country legally, to qualify for benefits (Fragomen Jr. 1996; Shaw and Shapiro 2002). California's particularly harsh Proposition 187, which virtually excluded undocumented immigrants from public services, including education and health care, exemplifies this sentiment most vividly (Alvarez and Butterfield 2000). However, several states have imposed strict limitations on immigrants’ social safety net access (Butz and Kehrberg 2019; Ku and Pervez 2010). These developments accelerated under the Trump administration and its infamous public charge regulations (Parrott, Gonzales, and Schott 2018). Given this history, we expect that explicitly noting that benefits are for citizens and green card holders—and thus not for undocumented immigrants—would lead to higher program support.

Fifth, it seems likely that highlighting different eligibility groups may have differential effects on Americans based on their preconceived notions about the Medicaid program driven by their partisanship, because health policy in the United States has long had strong partisan undertones (Blumenthal and Morone 2010). The Clinton health care reform efforts were highly partisan (Hacker 1997; Skocpol 1997), and partisan controversy and polarization over health issues only intensified in the wake of President Obama's election and the subsequent passage and implementation of the Affordable Care Act marketplaces (Altman and Shactman 2011; Haeder 2013; Haeder and Weimer 2013, 2015b; Jacobs and Skocpol 2010, 2011; McDonough 2011; Noh and Krane 2016; Oberlander 2016; Rigby and Haselswerdt 2013; Starr 2011; SWP 2010), the expansion of Medicaid (Barrilleaux and Rainey 2014; Callaghan and Jacobs 2017; Oberlander 2016; Olson 2015; Shor 2018), and the establishment of Medicaid work requirements under the Trump administration (Haeder 2019b; Haeder, Sylvester, and Callaghan 2021a). At the same time, Republicans have long held more unfavorable opinions toward individuals living in poverty (Hopkins 2009) and immigrants (Hawley 2011; Newman et al. 2012; Wallace 2014; Wroe 2008) than Democrats. As a result, we expect Republicans to respond positively to those program descriptions, which include the aforementioned groups generally considered to be more deserving; we also expect a positive effect when emphasizing citizenship or residency.

Moreover, much of American public opinion toward social and health policy is influenced by perceptions of race (Banks 2013; Gilens 1996, 2009; Haeder, Sylvester, and Callaghan 2021a; Snowden and Graaf 2019; Sylvester, Haeder, and Callaghan 2022). This particularly applies to the Medicaid program (Barrilleaux and Bernick 2003; Haeder, Sylvester, and Callaghan 2021a; Kousser 2002; Leitner, Hehman, and Snowden 2018; Olson 2010). Racially resentful Americans are significantly less supportive of the program, driven by the fact that most Medicaid beneficiaries are not white (Haeder, Sylvester, and Callaghan 2021a; KFF 2021). The racial backlash triggered by the election of President Obama as well as the Affordable Care Act has further exacerbated these tensions, particularly around the politics of the Medicaid program (Banks 2013; Fording and Patton 2019; Grogan and Park 2017b; Henderson and Hillygus 2011; Lanford and Quadagno 2016; Pasek et al. 2009; Segura and Valenzuela 2010; Snowden and Graaf 2019; Tesler 2012). Thus, we expect that individuals exposed to the treatments, especially those focused on “low-income individuals” who are high in racial resentment, may be less supportive of Medicaid and its beneficiaries. However, we expect the opposite effect for these types of individuals when pointing out that beneficiaries are US citizens and legal permanent residents. Importantly, while none of the sources for our treatments specifically emphasize race in their descriptions of the Medicaid program, we expect that some respondents could derive racialized imagery from otherwise race-neutral descriptions like “low income.”

In addition, while the complexities of the US health care system are ubiquitous, there are reasons to believe that they disproportionally affect individuals with lower levels of education. Lower levels of health literacy and health insurance literacy have been repeatedly found among less-educated individuals (Bartholomae et al. 2016; Brown et al. 2004; Kim, Braun, and Williams 2013; Loewenstein et al. 2013; Martin and Parker 2011; McCormack et al. 2009; Tennyson 2011). Moreover, the confusion might particularly affect the Medicaid program, as exemplified by the fact that individuals with lower levels of education are particularly likely to fail to identify themselves as Medicaid recipients (Bartholomae et al. 2016; Brown et al. 2004; Call et al. 2008a, 2008b; Card, Hildreth, and Shore-Sheppard 2004; Davern et al. 2009; Eberly, Pohl, and Davis 2005; Kim, Braun, and Williams 2013; Loewenstein et al. 2013; Martin and Parker 2011; McCormack et al. 2009; Tennyson 2011). Because low-education individuals are less likely to be informed about Medicaid, we expect that we are more likely to detect differences for low-education individuals.

Finally, personal connection to someone benefiting from the Medicaid program may also affect the response to our treatments. There is some evidence that either having been enrolled in the program or having a family member or friend who has been enrolled may positively affect support for and perceptions of Medicaid (Grogan and Park 2017a). However, other work on Medicaid work requirements did not find any such connection (Haeder, Sylvester, and Callaghan 2021a). While the findings are mixed, at the very least, knowing someone on Medicaid should mute the priming effect because these individuals are more familiar with the program and have personal experience with the program beneficiaries. Put differently, those individuals without a personal connection should be subject to larger effects from the experiment. We thus expect that individuals who do not know anyone who has been on Medicaid will be more affected by the various treatments.

Data and Methods

Based on our readings of Medicaid websites and the Medicaid policy literature, we developed and fielded a nationally representative survey of Americans. The survey included an experiment that primed respondents by highlighting various Medicaid target populations derived from the discourse of the Medicaid policy community. Specifically, we exposed respondents to one of six different Medicaid description frames (one control and five treatments) derived from our analysis of the Medicaid policy discourse. These frames alternatively described Medicaid beneficiaries as “low-income individuals and families” and “children, pregnant women,” and described Medicaid as a program for “eligible aged, blind or disabled people whose income is insufficient to meet the cost of necessary medical service” or “US citizens or legal, long-term permanent residents (green card holders) after a five-year waiting period,” or combinations thereof. We note that we sought to replicate the presentations of the Medicaid program as they occur in the discourse of the policy community; that is, we replicated the descriptions of target populations as we found them on Medicaid websites, public opinion surveys, and policy documents.

In this experiment, described below, we highlighted one or more of these target populations to respondents, which served as their introductions before answering questions about their attitudes toward Medicaid (table 1). We developed a purposefully vague, yet nonetheless realistic, description concerning target populations to serve as a control.³ Specifically, the control description noted the following:

Next, we developed a description that built on the control condition but emphasized that beneficiaries were primarily low-income individuals, as several policy descriptions listed above also do:

Our third treatment added several groups: children, pregnant women, the aged, the blind, and the disabled. Here, we specifically relied on what we found in the policy literature on the Medicaid program, which generally combined the various groups and thus asked:

We also developed a treatment with a focus on citizenship or permanent residency:

Finally, we developed multiple complex conditions that combined aspects of other treatments into a more comprehensive description. Treatment 5 combines low income (treatment 2) with citizenship and permanent residency (treatment 4).⁴ In contrast, treatment 6 combines low-income children, pregnant women, and eligible aged, blind, or disabled people (treatment 3) with citizenship and permanent residency (treatment 4).⁵ Specifically, we wrote:

and

To test the causal effects of the various descriptions, we then asked respondents several questions about their perceptions of the Medicaid program and Medicaid beneficiaries. Importantly, we also asked respondents whether they would enroll in Medicaid if they were uninsured and whether they supported several policy changes to Medicaid currently being debated by political elites.

To assess whether and how Medicaid program descriptions affect US public opinion related to perceptions of the program and its recipients, potential enrollment, and policy changes, we developed an original survey administered through Qualtrics. Respondents were obtained from the survey-research firm Lucid via its large, online opt-in panel based on quota sampling from June 8 through June 19, 2020. Lucid samples closely reflect population benchmarks regarding demographic characteristics such as age, race, gender, education, and income. Research to this point has found consistent evidence that Lucid sample quality is higher than convenience samples and approximates representative samples (Baker et al. 2013; Coppock and McClellan 2019; Kennedy et al. 2016; Levay, Freese, and Druckman 2016). Lucid-based data has been used extensively in health-related research (e.g., Haeder 2021; Motta et al. 2021). Overall, 2,086 respondents participated in the survey, and all were randomly assigned to one of the six conditions (one control and five treatments) mentioned in table 1. While the raw data are close to national benchmarks, we weighted all models to reflect national population benchmarks on gender, race, income, and education based on the US Census Bureau Current Population Survey (see appendix table 1).

For all comparisons, we estimated standard ordinary least squares (OLS) models (results omitted for brevity). OLS is appropriate here because we are not interested in the effect of any specific explanatory variable but in whether each experimental treatment affected mean outcomes. For our baseline comparison, we only included indicator variables for each treatment. To assess the potential disparate effect on the various subgroups included in our expectations (such as partisanship or level of racial resentment), we interacted the respective variables with the treatment indicators. We then estimated predictive means and compared differences using the commands mlincom and mchange in Stata (Long and Freese 2014).

Dependent Variables

In our analysis, we were interested in the potential effects of our experimental treatments on respondents’ (1) perceptions of the Medicaid program, (2) perceptions of Medicaid beneficiaries, (3) willingness to enroll in Medicaid, and (4) perceptions of four proposed policy changes to the Medicaid program.

First, to gauge Americans’ opinions on Medicaid, we asked them, “In general, do you have a favorable or an unfavorable opinion of Medicaid?” and provided them with options from “extremely favorable” to “extremely unfavorable” on a 5-point scale. Second, we sought to capture perceptions of beneficiaries with the question, “Do you think that individuals on Medicaid are generally deserving or undeserving of this benefit?” with response options ranging on a 5-point scale from “extremely deserving” to “extremely undeserving.” Third, to assess willingness to enroll in Medicaid, we asked, “If you personally were uninsured, how likely is it that you would turn to the Medicaid program to get health insurance coverage?” with options from “extremely unlikely” to “extremely likely” on a 5-point scale. Lastly, to assess respondents’ perceptions of proposed Medicaid policies, we asked them about their support for block-granting Medicaid, requiring Medicaid beneficiaries to pay premiums and out-of-pocket costs, requiring beneficiaries to participate in wellness incentives, and the establishment of work requirements. Each question included information for respondents, briefly explaining each policy and providing options ranging from “strongly oppose” to “strongly support” on a 5-point scale. Additional information about the wording of these questions, adapted from previously used questions from the Kaiser Health Tracking Poll, is available in appendix table A2.

Explanatory Measures

In our models, we explore four measures commonly utilized in research on public attitudes surrounding Medicaid and other health programs to capture critical moderating factors. First, we utilized Lucid's 10-point partisanship scale to assess the effect of partisanship. The scale ranged from “strong Democrat” to “strong Republican” and contained “other” and “independent” options. We collapsed the scale into a 3-point scale for Democrats, Republicans, and all others.⁶ Second, we utilized Lucid's 8-point education scale ranging from “some high school or less” to “doctorate degree.” For our analyses, we created a three-level variable ranging from “high school or less” to “college graduate or more.” To determine racially based opposition, we used four commonly used questions to assess the level of racial resentment among respondents (Kinder and Sanders 1996). We then split respondents into tertiles based on their levels of resentment.⁷ Finally, we provided respondents with a list of personal connections ranging from themselves to acquaintances and asked whom they knew who was enrolled in Medicaid. We then created a binary variable for individuals who had a personal connection to Medicaid or not.⁸

Results

Medicaid Favorability

Our first analysis focuses on Americans’ favorability toward the Medicaid program. The results are presented in figure 1. Our findings indicate that the mean level of favorability (i.e., the predicted mean) exceeded 3.5 out of 5.0 for all six conditions and ranged from 3.611 to 3.887. The only statistically significant differences for predicted means between control and treatment groups emerged for treatment 4 (“citizens,” p = 0.023), which primed respondents for US citizens and green card holders as Medicaid beneficiaries. Priming respondents with information about citizenship requirements increased support from 3.640 to 3.887. The only other treatment that came close to conventional levels of significance is treatment 6, which, in addition to mentioning US citizens and green card holders, also primed respondents for low-income individuals and children, pregnant women, and those who are aged, blind, or disabled. The treatment increased favorability.

Furthermore, our subanalyses focused on partisanship identified statistically significant differences among Republicans comparing treatment 4 (“citizens,” p = 0.013) and treatment 6 (“low income plus and citizens,” p = 0.007) to the control group, with higher support in the treatments than the control condition. We also found statistically significant differences among individuals with high levels of racial resentment. Exposing these individuals with priming for individuals in poverty (treatment 2, p = 0.015), even when supplemented with deserving groups like children and the aged (treatment 3, p = 0.027), reduced Medicaid favorability among these individuals compared to the control group. Finally, we found no differences among Democrats, those without a personal connection to Medicaid, and individuals with low levels of education when comparing the various treatments to the control group.

Beneficiary Deservingness

Our next analysis addressed perceptions of deservingness for Medicaid beneficiaries (figure 2). Overall, mean levels of deservingness ranged from 3.675 to 3.856. Our primary analysis found no discernible differences between the control group and any treatments. We also found no statistically significant differences across our subgroup analyses based on partisanship, lack of personal connection, or education. However, once more, we found negative effects among individuals with high levels of racial resentment comparing treatment 2 (“low income,” p = 0.048) and treatment 3 (“low income plus,” p = 0.023) to the control group. The negative effects are also substantial. Comparing treatment 2 to the control group, we saw a reduction of 0.419, from 3.739 to 3.320; for treatment 3, the drop was even more significant, at 0.548 (3.739 vs. 3.191).

Potential Medicaid Enrollment

Next, we focused on whether different conditions would affect individuals’ willingness to enroll in the Medicaid program if they were uninsured (figure 3). Overall, the predicted means ranged from 3.509 to 3.823. We only found substantial differences when comparing the control group to treatment 4 (“citizens,” p = 0.041), increasing the willingness to enroll in Medicaid from 3.552 to 3.823. Our subgroup analyses found differences for the same treatment compared to the control group among Republicans (p = 0.047). Here, we saw an increase from 3.268 to 3.688. The effect became more robust in both cases for the analyses referencing enrolling into Medicaid in the case of need for major surgery (results omitted for brevity). We do not detect any differences among Democrats, those who lack personal connections, those with low education levels, or those with high levels of racial resentment.

Medicaid Policy Changes

Lastly, we questioned respondents about their support and opposition to several policy changes considered in discourse at the time of the survey, including block-granting the Medicaid program (figure 4), expanding the use of out-pocket payments and premiums (figure 5), required participation in wellness incentives (figure 6), and establishing work requirements for Medicaid beneficiaries (figure 7). Predicted means ranged from 2.543 to 2.704 for block-granting, 2.761 to 2.926 for out-of-pocket payments, 3.292 to 3.454 for wellness incentives, and 3.268 to 3.480 for Medicaid work requirements, respectively. We did not detect any substantial differences comparing the control group to the various treatments for wellness incentives or block grants overall or across the subgroup analyses. At the subgroup level, we found statistically significant differences regarding out-of-pocket payments and premiums among Republicans (p = 0.010) and those with lower educational attainment (p = 0.043), comparing the control group to treatment 6 (“low income plus and citizens”). In both cases, the treatment reduced support for the administrative burden. Surprisingly, we found reduced support for Medicaid work requirements for Republicans comparing the control group to treatment 2 (“low income,” p = 0.035); however, the confidence bounds around the estimate are relatively large. This finding differs from previous research and warrants further inquiry.

Discussion

In this study, we sought to understand how changing the way Medicaid is commonly framed among the policy community in terms of its target populations may influence overall favorability toward the program, the perceived deservingness of beneficiaries to receive benefits, how it impacts willingness to enroll, and finally, how it influences policy preferences for four prominent policy changes imposing administrative burdens on beneficiaries. We specifically selected target populations commonly mentioned in the discourse of the Medicaid policy community, including individuals and families with low income, children, pregnant women, the aged, blind, or disabled people, and US citizens and green card holders. We specifically sought to replicate descriptions commonly used by the policy community as found on websites, public opinion surveys, and policy documents to increase the policy relevance of our findings. In addition, we purposefully presented respondents with commonly used descriptions that are deliberately bland, solely focus on the low-income aspect of the program, provide an expansive list of eligible groups, or, although rarely present in policy descriptions, include the citizenship/permanent residency requirements of the program.

Overall, our analysis finds that favorability for the Medicaid program, no matter which description is presented to respondents, is relatively high, even among those demographics commonly known to be less favorable toward the program, such as Republicans and those high in racial resentment, despite a marked decrease for the latter. Unsurprisingly, Democrats are highly favorable toward the program, as are those with lower educational attainment and those with a personal connection. This bodes well for the Medicaid program overall. Moreover, this pattern also holds for assessments of the deservingness of beneficiaries as well as their willingness to enroll in the program. Proposed conservative policy changes such as block-granting the program and implementing premiums and out-of-pocket payments were perceived less favorably by respondents. With that said, support for these changes was not negligible and may indicate the potential for public support of conservatives making adjustments to the program in the future. Lastly, there are substantially higher levels of support for the implementation of wellness incentives as well as work requirements. In all cases, support for these changes is highest among those high in racial resentment.

We also found only marginal evidence that descriptions focusing on various target populations commonly found on Medicaid websites, public opinion surveys, and policy documents influenced the overall favorability of the program when compared to the control group. A notable exception occurred when respondents were primed with the knowledge that Medicaid was only provided to US residents and green card holders, with positive effects for Medicaid favorability and willingness to enroll in the program, particularly for Republicans. We found negative effects by emphasizing the low-income nature of the program among those high in racial resentment, indicating that racial, anti-immigrant, and nativist sentiments continue to play an essential role in shaping attitudes toward the program (Ku and Pervez 2010). However, highlighting citizenship and residency requirements once again appeared to mitigate this animosity to some degree.

Our subgroup analysis added further nuance to our findings. While results are mixed, we often found that the citizenship effect is particularly apparent for Republicans and those with high levels of racial resentment. This is unsurprising given the growing alignment of nativism and xenophobia in the Republican Party (Abramowitz and McCoy 2019; Wetts and Willer 2019). At the same time, we also saw a marked drop in support at times for the treatments that primed individuals to think about how Medicaid provides support for those with low incomes. This is in line with the previous trajectory of the program as well as public assistance more generally (Haeder 2019a). The lack of effects on low-education individuals is noteworthy. Finally, the general results for individuals high in racial resentment stand out. Once more, these findings are in line with previous explorations of the Medicaid program (Haeder, Sylvester, and Callaghan 2021a).

Limitations

Our research, of course, is not without limitations. First, our approach to testing highlighting different Medicaid target populations is based on our reading of the Medicaid policy literature, including Medicaid websites, public opinion surveys, and policy reports produced by entities like the Kaiser Family Foundation. Our readings of these documents indicated that program descriptions generally fall into various categories, including those low on information, those focusing on the low-income nature of the program, and those providing broader descriptions while maintaining modifiers highlighting the low-income nature of the program, while some mention citizenship or residency requirements. Importantly, our approach focused on whether these policy-based descriptions shift public perceptions. While our approach is policy-relevant, we are aware that the combination of eligibility groups creates an amalgam of groups with highly diverse levels of perceived deservingness that may pull respondents in different directions.

Moreover, the descriptions used in Medicaid websites, public opinion surveys, and policy writings may not necessarily reflect those used in the mass media or public conversations. This particularly applies to potentially racial imagery used in these depictions (Gilens 1996). Further research is needed to explore the extent to which descriptions vary between the policy community and public conversations and the impact of Medicaid frames in public discourse on public attitudes.

In addition, our policy-based approach lumps together various rather different target populations, preventing us from testing public perceptions of the individual groups. Although individually testing each eligible demographic group and the various combinations adds complexity, we hope that future research can disentangle some of the different subgroups and further extend our findings here and add additional nuance with regard to more discrete eligibility groups. Nevertheless, we reemphasize that our descriptions are representative of the policy discourse on Medicaid and thus meaningful from a policy perspective.

Our analysis is also limited by its focus on attitudes at a single moment in time. The cross-sectional nature of our data prevents us from capturing whether specific description characteristics vary over time in importance based on events happening at the state or national levels. Simultaneously, if states choose to change how they describe their Medicaid programs in the future, our analysis cannot assess the impact of those changes.

Another limitation of our analysis stems from the choice of our control treatment. By using a control that provides a generic description of Medicaid instead of no description, we cannot assess how specific descriptions compare to no description of the program. With uncertainty surrounding the public's basic understanding of Medicaid, we thought it essential to provide some sort of introduction. At the same time, we also cannot account for preconceived notions respondents have of Medicaid recipients with regard to important demographics such as race and ethnicity or income status. It is possible that these perceptions make respondents impervious to change based on the information we provide them within our experiment. We also did not use state-specific names of the Medicaid programs, which might have further increased program recognition.

Our analysis is also limited by our sample, provided by Lucid, which closely approximates population benchmarks. Concerning the policy changes we queried respondents about, it is possible that, despite our brief introduction, respondents may not have been fully aware of the content of the proposed changes. We also note that the policy changes presented to respondents focused on adding administrative burdens. Results may be different for other policy changes. Finally, there is a possibility that the survey period (early June 2020) may have been a time when respondents felt particularly sympathetic toward individuals needing health insurance coverage because of the global COVID-19 pandemic.

Despite these limitations, our analysis still provides critical new information about the impact of Medicaid program descriptions in the policy discourse on public attitudes surrounding the program, as used by reputable polling organizations (AARP 2011; CNN 2011; CR/RI 1981; HSPH/NYT/CF 2019; KFF 2020; Latino Coalition 2006; RWJF/HSPH 1997), Medicaid-related websites, and policy writers (CBPP 2020; MACPAC 2022; NCA 2022; Rudowitz, Garfield, and Hinton 2019; USGAO 2020).

Conclusion

Overall, our findings bode well for the Medicaid program. Across all demographic groups of interest in our analysis, favorability and perceptions of beneficiary deservingness are relatively high, despite marked differences for groups traditionally more suspicious of the program, such as Republicans and those high in racial resentment. However, findings on policy changes adding further administrative burdens to the program indicate that specific reforms may plausibly make it harder for beneficiaries to benefit or even stay enrolled. This particularly holds true for the implementation of wellness incentives as well as work requirements.

As highlighted above, our analyses focus on program descriptions commonly found in Medicaid-related websites, public opinion surveys, and policy writing. Within this framework, we generally only found limited effects across the different real-world treatments. That is, within the confines of our treatments, in most cases, individuals are generally unmoved by how the program is described. However, several conclusions emerge from our nuanced findings. First, survey researchers and scholars should be aware that the way they commonly describe the Medicaid program, either as a low-income program for individuals and families or with the addition of target populations such as pregnant women or the disabled, has small but nuanced effects. Thus, these organizations should likely focus on the latter.

Moreover, they should consider adding the important modifier that, in general, beneficiaries must be either citizens or permanent residents. It provides a more well-rounded description of the program and a better approximation of public opinion about the program, its beneficiaries, and potential policy changes. As we highlighted above, these differences particularly affect demographics generally less supportive of the program.

More generally, our findings further add to our understanding of the critical role of partisanship and racial resentment in American politics. It is worth noting that our findings regarding the role of racial resentment add to a growing body of work highlighting the importance of race in US health policy in general and Medicaid in particular. We emphasize again that neither descriptions by the policy community nor our introduction to survey respondents contained any racial indicators. This means that survey respondents derived racialized imagery from otherwise race-neutral descriptions like “low income.” Moreover, public opinion toward Medicaid also falls along partisan lines. Both findings raise important questions about future possibilities for bipartisan changes to the Medicaid program and broader US health care system adjustments.

Our findings presented here also provide guidance about the path forward for proponents of the Medicaid program and those who would like to increase enrollment and eligibility as well as the broader efforts to educate the public on what Medicaid is, its purpose, and whom it impacts. Our findings thus point to the need to clarify that eligibility for Medicaid goes well beyond low-income individuals. This would help potential beneficiaries determine whether they fall within eligibility categories and reduce opposition to the program among individuals who are traditionally critical of the program. At the same time, citizenship and permanent resident requirements should be highlighted, as they increase program support and enrollment. Lastly, however, it must be said that our findings indicate that a substantial part of the US population remains highly suspicious of both individuals in poverty as well as immigrants and their use of the American social safety net.

Finally, these findings potentially have important implications for other government programs beyond Medicaid. It is conceivable that similar findings may occur with programs like the Supplemental Nutrition Assistance Program (SNAP), Temporary Assistance to Needy Families (TANF), or the Children's Health Insurance Program, which also see differences across states. Future research should examine how framing these government programs influences public opinion and policy preferences. This particularly applies to SNAP and TANF, which are even more controversial than Medicaid. It also seems likely that emphasizing citizenship and residency requirements may positively contribute to perceptions about these programs.

Notes

References