Context: The United States is the only high-income country that relies on employer-sponsored health coverage to insure a majority of its population. Millions of Americans lost employer-sponsored health insurance during the COVID-19–induced economic downturn. We examine public opinion toward universal health coverage policies in this context.
Methods: Through a survey of 1,211 Americans in June 2020, we examine the influence of health insurance loss on support for Medicare for All (M4A) and the Affordable Care Act (ACA) in two ways. First, we examine associations between pandemic-related health insurance loss and M4A support. Second, we experimentally prime some respondents with a vignette of a sympathetic person who lost employer-sponsored coverage during COVID-19.
Findings: We find that directly experiencing recent health insurance loss is strongly associated (10 pp, p < 0.01) with greater M4A support and with more favorable views of extending the ACA (19.3 pp, p < 0.01). Experimental exposure to the vignette increases M4A support by 6 pp (p = 0.05).
Conclusions: In the context of the COVID-19 pandemic, situational framings can induce modest change in support for M4A. However, real-world health insurance losses are associated with larger differences in support for M4A and with greater support for existing safety net policies such as the ACA.
Sixty percent of working-age Americans received health insurance through an employer-sponsored plan in 2019 (KFF 2019). Consequently, the massive job losses associated with the COVID-19–induced economic downturn led to an estimated 3–27 million Americans losing their employer-sponsored health insurance in the first months of the pandemic in 2020 (Banthin and Holohan 2020; Fronstin and Woodbury 2021; Garfield et al. 2020). Given that alternative insurance options are often unaffordable, many of these working-age Americans remained uninsured in the midst of a pandemic (Garfield and Tolbert 2020).
The pandemic highlights the risks of relying on employer-sponsored health coverage in two ways: (1) millions of Americans have lost their jobs and often their health coverage (for any illness), and (2) the pandemic itself brings increased risk of illness and associated costs, as a potential COVID-19–related hospital stay could cost tens of thousands of dollars (FAIR Health 2020; Rae et al. 2020). The increased salience of these risks may affect Americans’ views about health insurance in general and about the risks of linking insurance to employment in particular. If so, it offers an opportunity for advocates of expanded health insurance coverage to highlight the limitations of employer-sponsored coverage and make the case for delinking insurance from employment. Moreover, it could enable new political coalitions in favor of universal health coverage, if the millions of Americans who unexpectedly lost employer-sponsored coverage could be persuaded to support this alternative.
Prior to the emergence of COVID-19, policies to achieve universal health coverage (UHC), including through Medicare for All or expansions of the Affordable Care Act, were already on the policy agenda in the United States, most notably during the 2019–20 Democratic presidential primary campaign. Various Democratic candidates proposed plans to increase coverage, ranging from wrap-around policies to fill gaps in the existing system (“Medicare for All who want it” or “Medicare buy-ins”) to more expansive visions of “Medicare for All” (hereafter, M4A), which has become shorthand for single-payer insurance with universal coverage in the United States. Popularized by Senator Bernie Sanders, M4A would fully delink health coverage from employment and provide universal, tax-financed health insurance coverage (Uhrmacher et al. 2020). Yet the candidates most associated with M4A, Senator Bernie Sanders and Senator Elizabeth Warren, lost the primary to Joe Biden; Biden supported a plan to expand health insurance coverage, including with a public insurance option, but did not support M4A. Just as this intra-Democratic primary election was concluding in spring 2020, the COVID-19 pandemic, and the sudden job and health insurance loss that it entailed, became a feature of American life.
Amid this context of increasing health risks, large-scale job loss, and health insurance disruption, we explore public opinion about policies to expand health insurance coverage among an online sample of 1,211 Americans. We examine whether elements of this pandemic, notably the widespread experience of health insurance and job loss, increase support for government's role in the health system; we focus particularly on plans, such as M4A, that delink health insurance from employment. We also include questions about support for the Affordable Care Act as well as more general support for universal health coverage as a goal.
We present five main findings. First, we find that respondents who experienced recent health insurance loss have 10–15 percentage points (pp) higher support for M4A (p < 0.01) than those who have not, including when controlling for a wide range of demographic and socioeconomic factors. Following Lawrence R. Jacobs and Suzanne Mettler (2011), we consider this experience a “structural” factor. Second, we find that this effect is moderated by political party affiliation; most movement toward M4A associated with insurance loss is among self-identified Republicans, who have much lower levels of support for M4A overall. Third, we show that priming respondents about the relationship between involuntary job loss and insurance loss shifts their views about M4A. Experimentally priming respondents with emotive vignettes about no-fault job and insurance loss during COVID-19 results in a 5.5 pp increase in support for M4A (p < 0.050). Following Jacobs and Mettler, we consider this vicarious (via vignette) experience of insurance loss a “situational frame.” However, the results of the survey experiment are relatively modest in magnitude compared to the real world, “structural” determinants of opinion, such as the impact of losing one's insurance. Fourth, political party affiliation does not moderate the effect of this situational frame. Fifth, in a secondary battery of questions in which multiple policies for coverage expansions were presented as options, we find that the survey vignette treatment increases support for M4A, but that personal health insurance loss is associated with increased support for the Affordable Care Act and strong opposition to ACA repeal, and less support for M4A.
Changing Support for Medicare for All
Many health policy experts view the barriers to a “Medicare for All” system in the United States as primarily political, rather than technical (Berwick, Nolan, and Whittington 2008). While there are many barriers to comprehensive reform, including the multiple veto points that characterize American political institutions (Steinmo and Watts 1995) and widespread opposition from industry stakeholders, an important element of the political feasibility of M4A is public opinion. In public opinion polls, majorities of the American public have consistently, over the past two decades, favored a greater role for government in health care; 50%–60% have been found to be supportive of greater government involvement in health care in general or universal coverage policies in particular (Gallup 2021a; KFF 2020; Steinmo and Watts 1995). Likewise, a majority of US physicians, a group once overwhelmingly opposed to a national health plan, now support a single-payer system (Bluth 2017). Moreover, M4A plans, in name if not in substance, are generally popular with the public, with majorities expressing support (Karra and Sandoe 2020; KFF 2020).
However, general popularity does not translate into unconditional support for M4A. Previous studies have found support to be sensitive to question wording and framing of the issue, with specific framings either increasing or reducing support (Karra and Sandoe 2020; KFF 2020; Oberlander 2019). In addition, while “Medicare for All” has a clear meaning in the health policy world, it is less clear how it is understood by voters. Many may consider it shorthand for a general expansion of health coverage or may believe it also refers to more incremental Medicare buy-in plans (Oberlander 2019). Furthermore, when given more head-to-head comparisons of different potential health reform options, including keeping and expanding the Affordable Care Act or giving states more flexibility to design public health insurance options for their residents, recent polling has found that the public splits nearly evenly among the three options (30% favoring each option) (McIntyre et al. 2020), including with Democrats somewhat more favorable toward building on the ACA (KFF 2020).
In this study, we focus on M4A approval as our main outcome, as it is the health plan that most directly captures the delinking of employment and insurance. We do, however, recognize that M4A opinions may be a proxy in many voters’ minds for general government support for health insurance coverage. There is ambiguity about whether support for M4A has increased since the onset of COVID-19 in the United States in March 2020: for example, polling by Morning Consult showed a nine percentage point increase in support for M4A between February and March 2020 (Murad 2020), while other polls indicate that support for M4A has remained constant (Hill2020). It therefore remains unclear if COVID-19 is sufficiently disruptive to cause a long-lasting (“structural”) change in public opinion toward M4A as well as whether ongoing experience with the ACA increases public approval of the law. These ambiguities motivate the remainder of this article.
Our theoretical motivation on opinion toward health insurance coverage builds on Jacobs and Mettler's (2011) framework of “situational” versus “structural” framing of public opinion about health care. Jacobs and Mettler (2011) argue that public opinion about the US health system is primarily rooted in structural factors, which reflect citizens’ long-standing, institutionalized interactions with health insurance and the health care system in the course of their lives. This suggests that the salience of one's own lived experience or other relatively fixed characteristics of individuals largely shape views toward health care and health insurance policy questions.
However, in the short run, opinions can also vary depending on situational framing, that is, the way the message is conveyed and the moment or context in which it is conveyed. Such frames may temporarily boost the salience of issues outside one's lived experience. Frames are used by individuals and groups to highlight specific aspects of the problem and to emphasize certain causal links (accurate or not) that temporarily increase or dampen support (Entman 1993). Health issues frequently typify a competitive framing environment, in which two sides or opposing arguments compete with each other in the public sphere (Chong and Druckman 2007).
There is evidence that situational frames affect support for health policies such as M4A and the ACA. For example, M4A in particular is susceptible to a number of common forms of attack; polling often shows high initial support followed by a decline in enthusiasm as policy details are framed in unflattering ways (KFF 2020). Certain counterarguments tend to depress support for M4A—for instance, the idea that a single-payer system could increase wait times for appointments, lead to large tax increases and a doubling of the government budget, and constitute a “government take-over” of health care (KFF 2020).
Conversely, support for health reforms such as M4A or the ACA can be strengthened through positive situational frames. Jason Barabas, Benjamin Carter, and Kevin Shan (2020) find that providing survey respondents with policy “analogies” for various health programs increased support (such as using car insurance analogies to describe the individual mandate of the Affordable Care Act). Other recent survey experiments find that simpler framing elements can also increase support for the policy—for example, by including the policy name “Medicare for All” with a description of the policy (Karra and Sandoe 2020).
More fundamentally than situational or framing effects, crises (such as the COVID-19 pandemic that struck the United States starting in early 2020) can act as shocks that could theoretically disrupt equilibria and lead to more structural changes in public opinion as well as changes in political alignments that may facilitate policy change (Baumgartner and Jones 1993). However, situational frames may still be invoked by policy elites to counteract these shifts in public opinion at critical junctures.
Research Questions and Hypotheses
Our study design allows for examination of both situational and structural elements of opinion formation and change around M4A. First, given the broader context of the pandemic, we examine the association between recent insurance loss and attitudes toward M4A. Second, we examine the situational framing of attitudes toward M4A with a survey experiment, by measuring how priming respondents about the effects of job loss on insurance coverage affects their attitudes toward M4A. Through this experiment, we randomly expose readers to either no vignette (control) or one of two emotive vignettes of job and insurance loss; we present identical, sympathetic victims who experience no-fault job loss—because of either COVID-19 or technological and market changes.
We hypothesized that both personal experience with insurance loss and exposure to vicarious insurance loss, via vignette, would increase support for M4A. We further hypothesized that framing effects would vary based on political partisanship. This moderating effect was prespecified in an Evidence in Governance and Politics (EGAP)–registered analysis plan. Strong partisans may have more rigid attitudes and therefore be less susceptible to priming. This view aligns with theories of motivated reasoning, which suggest that strong political partisans will be unlikely to change their core positions and may even dig in their heels more firmly in the face of counterevidence (Strickland, Taber, and Lodge 2011). We therefore hypothesized that among strong partisans of either political party, the priming treatment would have limited impact. By contrast, we hypothesized that self-described Independent voters would be more likely to shift opinions in response to priming.
Sampling and Data Collection
We conducted an online opinion survey with a national sample of 1,211 Americans between June 3 and June 8, 2020, during the height of the COVID-19 lockdowns in the United States. Respondents had to be at least 18 years of age and consent to completing the survey. The project underwent ethical review and received approval from the University at Albany Institutional Review Board. Respondents provided informed consent before participating.
We used the third-party firm Qualtrics to administer the survey. Qualtrics is an internet survey provider that recruits respondents who have signed up to take online surveys in exchange for incentives such as cash, airline miles, and gift cards. Qualtrics aggregates respondents initially recruited by other firms. Recruitment and compensation are handled by the third-party firm, but researchers may define the audience and specify certain quotas.
While Qualtrics does not provide a probability sample of the US population, a recent study found that among internet survey providers, a Qualtrics-recruited sample came closest to a national probability sample on most variables relative to samples recruited through Amazon MTurk or Facebook (Boas, Christenson, and Glick 2018). We report on key characteristics of our sample in table 2. Notably, when compared to the US population, our sample has higher proportions of Republicans and Democrats and is more likely to be younger (table A5).
On average, the survey took 15 minutes for respondents to complete. Qualtrics provides quality-control measures to weed out those who do not complete the survey and who do not appear to be taking the survey seriously (such as “speeders”) as well as those who appear to be bots based on input provided in open-ended questions. Twenty percent of the starting sample was dropped through the quality checks, leaving us with an analytic sample of 1,211 high-quality responders.
Our main outcome of interest is support for M4A. Our primary outcome variable is the response to the following question, which is the same question wording used by the Kaiser Family Foundation's recurring survey “Public Opinion on Single Payer, National Health Plans, and Expanding Access to Medicare Coverage” (KFF 2020): “As of today, do you favor or oppose a national health plan or ‘Medicare for All’ plan, in which all Americans would get their health insurance from a single government plan?” Respondents could select: strongly favor, somewhat favor, somewhat oppose, or strongly oppose this statement, or report that they do not know. We show the breakdown of responses in table 1. For analysis, we recoded this 4-point Likert scale into a binary variable capturing support for M4A for those who reported “strongly” or “somewhat” favoring M4A.
We also ask about support for other health care reform options using alternative survey items that gave respondents the choice of other health policies, such as expansion or repeal of the Affordable Care Act. We explore the robustness of our findings by using these additional questions to gauge opinion about health insurance expansion via differing question wording, response options, and issue framing.
Actual Insurance Loss
We also leverage variation in pandemic-associated insurance loss to examine the association between having lost one's own health insurance and support for M4A. Our survey collected information about health insurance loss by asking whether the respondent had lost their health insurance in the last 6 months for any reason. We examine the effect of insurance loss on support for M4A through regressions controlling for age, race/ethnicity, gender, previous year income, and political partisanship.
Vicarious Insurance and Job Loss: Experimental Conditions and Randomization Procedure
Our experimental condition is a vignette about job and insurance loss, intended to prime the reader to think about job loss and consequent loss of employer-sponsored health coverage. Respondents were randomly assigned to one of three groups with equal probability: the control group (no vignette), a COVID-19 vignette, or an Airbnb vignette, described below.
In each of the experimental conditions, we present the job-loss vignettes as brief newspaper articles at the beginning of the survey, narrating the story of a white, male former football player (“Sean McGuire”) who gets laid off from his job as hotel concierge in Philadelphia and loses his employer-sponsored health coverage. In one vignette (hereafter “the COVID-19 vignette”), Sean is laid off as a result of COVID-19–induced economic downturns; a plausible scenario, as COVID-19 caused major job losses in the hospitality industry. In the second experimental condition, the layoff is the result of competition from Airbnb (“the Airbnb vignette”). We take this second condition as a “normal” unemployment condition related to market changes. Please see the online appendix for the full vignettes.
We chose to use a newspaper article to present the vignette in order to simulate how people might receive information in the real world. The article was adapted from an actual news story. We chose for the protagonist in the vignette to be a white male to avoid known racial biases/empathy gaps in redistributive politics (e.g., Alesina, Glaeser, and Sacerdote 2001). In both conditions, we take Sean to be a generally sympathetic victim and his job loss to be not his fault. At the end of our survey, respondents were informed that the newspaper article they had read was fictitious but that the information provided in it was accurate. We included two comprehension questions to ensure respondents actually read and understood the vignettes, which respondents had to pass to proceed in the survey.
Prior to data collection, the survey experiment was preregistered with EGAP, and experimental results are reported according to the original study design. Observational analyses of the association between insurance loss and M4A support were not preregistered. All analyses were completed in Stata 15. In all of our main analyses, we control for sex, age, previous year's income, political party identification, and race/ethnicity. To explore the moderating effects of political party identification, we interacted Democratic, Republican, or Independent party identification variables with the pooled treatment (exposure to either job loss vignette) to estimate the impact on support for M4A, in unadjusted models as well as models that controlled for gender, age, income, and race/ethnicity. Party identification was measured by asking respondents, “In politics today, do you consider yourself a Republican, Democrat, or Independent?” We repeat these models with different question wordings and with the “do not know” responses dropped (results available in the online appendix).
Sample and Descriptive Findings
In the 6 months prior to our survey, 22% of respondents lost health insurance. More than half (13% of total) of these respondents lost health insurance because of losing their job, while the remaining 9% lost health insurance for other reasons. Another 23% report that someone close to them lost health insurance.
Our sample is comparable to the US population on gender balance (52% female) and the percentage of respondents who lack health insurance (9%); however, our sample is younger, more likely to be white (72%), and less likely to be Hispanic (6%). Our sample contains more self-described Democrats and Independents, as well as fewer Republicans, than Gallup's data on party affiliation from the same week that our survey was fielded (Gallup 2021b). Compared to the national unemployment rate in June 2020 (11.2%), 24.74% of our under-65 sample reported being currently unemployed (Economic Daily2020). While the sample has representation from all 50 states as well as Washington, DC, roughly proportional to the population in each state, New York is overrepresented in our sample.
The sample was mostly balanced across experimental conditions on key covariates with the exception of age (see table 2). The control condition was significantly younger, with 20% of respondents in that condition younger than 25, compared to between 10% and 12% in other study arms. Party identification, ethnicity, income, and gender were balanced across treatment arms. We present both unadjusted models as well as those that adjust for age and other covariates (gender, race/ethnicity, previous calendar year income, and political party identification). As is standard practice, we account for sample imbalance by controlling for these characteristics in the regression. We also reweight our sample to account for age-related imbalance using inverse probability weighting (IPW) methods. This is discussed further in the robustness checks section; the results of this reweighting are presented in the online appendix.
Situational Framing: Vicarious Insurance Loss through Experimentally Assigned Vignettes
In bivariate analysis, the COVID-19 vignette increased M4A support by 6.2 pp (p = 0.06), while the Airbnb arm increased support for M4A by 4.8 pp (p = 0.14). A combined treatment indicator pooling both vignettes increased support by 5.5 pp (p = 0.05) (fig. 1).
In multivariate analysis, we again examined the impact of each treatment (vignette) study arm separately relative to control, then pooled both vignettes into a single treatment. Priming respondents with the COVID-19 vignette increased stated support for M4A by 5.7 pp (p = 0.07). Priming with the Airbnb vignette increased support for M4A by 4.6 pp (p = 0.14) (table 3). We cannot reject the null hypothesis that the two treatment arms are equivalent (p = 0.68), thus we reported pooled treatment effects going forward. The pooled effect of any prime on M4A support is 5.1 pp (p = 0.057). Treatment effects drop to 2.6–3.1 pp when “do not know” responses are excluded (table A2). This implies that the priming treatment affects both “oppose” and “do not know” groups.
Structural Framing: Personal Insurance Loss
Next, we estimate the association between recent health insurance loss on support for M4A (table 4). In columns 1 and 2, the independent variable is any health insurance loss within the previous 6 months, with controls for age, gender, race/ethnicity, political party identification, and previous-year income. Recent health insurance loss is associated with a 10 pp increase in M4A support. In columns 3 and 4 we restrict this to respondents who lost health insurance specifically as a result of losing their job; in these specifications, insurance and job loss is associated with a 15 pp increase in support for M4A. In columns 2 and 4 we restrict the sample to respondents not currently on Medicare, since job loss should not be strongly related to insurance status for respondents older than 65. We find similar results with respondents on Medicare removed.
Moderating Effects of Reported Political Party Identification
Next, we examine the moderating variables of political party identification on both the experimental treatment and on real-life job and insurance loss (see table 5). In the experimental component, we find no significant differences in the impact of priming by party identification. Independents were not more likely to switch their positions, counter to our prespecified hypothesis. By contrast, in the observational analysis of the association between insurance loss and M4A support, political party identification is an important effect moderator. Virtually all of the increased support for M4A among those who have lost health insurance comes from Republican respondents. The additional effect of the interaction of insurance loss with GOP identification is 20 pp (p < 0.05); for insurance loss specifically because of job loss it is slightly smaller, and we cannot reject the null hypothesis of zero differential effect (17–18 pp, p = 0.12). In all analysis of partisanship, results are unchanged whether we include strong partisans only or whether we include those who consider themselves Independent but acknowledge “leaning” Republican and Democratic when pushed.
As robustness checks, we examine alternative closely related outcome variables for both the structural insurance loss and situational frame outcome variables. We compare the results of the main outcome measure with two other measures of support for M4A (tables 6 and 7). The first alternative is a series of questions that ask respondents whether they have a mostly positive, or mostly negative, impression of a series of labels: Medicare for All, Medicare for those who want it, universal health coverage (UHC), national health insurance (NHI), and Obamacare. Notably, the experiment only increases the percentage reporting “mostly positive” opinions significantly about M4A (7.1 pp, p < 0.05) and to a lesser extent “Medicare for those who want it” (4.5 pp, p = 0.11) (see table 6, panel A). By contrast, losing one's health insurance is associated with a mostly positive view of Obamacare (7–8 pp, p < 0.05) and, in covariate-adjusted models, is associated with a more negative view of M4A (−8.4 pp, p < 0.05) (see table 6, panel B).
In the second alternative set of questions, respondents were asked to choose among three mutually exclusive options that best described their opinion about which direction the United States should go in health policy reform: “incrementally building on the Affordable Care Act,” “reversing the Affordable Care Act and moving towards more private health insurance coverage,” or “creating a universal M4A system that would replace employer-sponsored health insurance coverage.” The experimental treatment did not shift views on any of these significantly; however, personal experience of insurance loss is associated with more favorable views of extending the ACA (19.3 pp, p < 0.01), more opposition to repealing the ACA (−7.6 pp, p < 0.01), and less favorable views of M4A (−9.8 pp, p < 0.01) (table 7). Full question wording and descriptive statistics from these questions are in the online appendix.
A final robustness check involves addressing the imbalance by age in the experimental sample. In addition to controlling for age in main regressions, we also implement an inverse probability weight (IPW) correction to account for age-related sample imbalance (online appendix table A3). Results are qualitatively similar after this reweighting.
We have examined opinion toward a proposed major reform of the US health system, including expansion of health coverage (M4A) in the context of large-scale job and health insurance loss during the COVID-19 pandemic. The experience of health insurance loss—a “structural” factor—is associated with 10–15 pp higher support for M4A. This association is moderated by political party identification, as the effect is driven by respondents who identify as Republicans. We also find a modest impact of an emotive vignette of no-fault job loss on support for M4A: experimental priming increased support for M4A by 5.5 pp. The effect appeared to primarily work through moving people who would otherwise have had ill-formed preferences on M4A into the more supportive category. Political party identification did not moderate the effect of the situational frame. Alternative question wordings revealed that the situational experimental frame was strongest in moving people to have a more favorable view of M4A, whereas personal (“structural”) insurance loss was associated with a more positive view of ACA/“Obamacare,” support for expansion of the ACA, and corresponding reductions in support for M4A. However, we note that the associations between job loss and opinion are observational estimates and despite extensive controls may be biased by unmeasured confounding variables.
Taken together, these key results suggest that both situational framing and structural effects can increase support for universal health coverage policies, but that structural effects, although nonexperimentally identified, appear larger and stronger. While it remains too early to tell, those who have lost valuable employer-sponsored insurance may serve as a future constituency in support of programs to expand access to health insurance. Within the sample, there was quite broad support for M4A when asked as a stand-alone question—nearly 70% of the sample reported strong or moderate support. Likewise, nearly 54% of the sample reported that their support for M4A had increased following COVID-19 (see table A4 in the online appendix). This level of support is higher than national polls in which, in October 2020 (pre-COVID), 53% favored a national Medicare-for-All plan (KFF 2020), and where, in 2021, 56% of people thought that it is the government's responsibility to make sure all Americans have health care coverage (Gallup 2021a). However, the lived experience of insurance loss was associated with more support for the ACA in alternative question framings in which the ACA was offered as an alternative to M4A. Among respondents who lost health insurance, a plurality remained on employer-sponsored insurance (either from new employment, their spouse, or their parents), but more than one in four reported purchasing private insurance plans using government subsidies (i.e., benefiting from the ACA). While the sample sizes are too small for reliable inference about these subgroups, we hypothesize that this direct experience with the benefits of the ACA may have led these respondents to favor it instead of the less familiar option of M4A. Thus it is also possible that pandemic-driven insurance loss will build a larger structural coalition in support of the ACA.
Given the role of partisanship as a driver of Americans’ policy views, we find it notable that the association of personal health insurance loss and M4A support was stronger among Republicans, suggesting that insurance loss may be powerful enough—at least in the short run—to change the opinion of those with more entrenched oppositional beliefs toward government involvement in health care. Whether these changes can be sustained and ultimately converted into support for candidates who propose expanded government programs remains a challenge in a deeply polarized electorate.
We also find that situational frames, which can provide additional information linking the impact of job loss to insurance loss, may help solidify people's views on Medicare for All. Practically speaking, this demonstrates that advocacy efforts may be effective at moving opinion on M4A, at least temporarily. However, since counterframes were not directly tested, we cannot assess how similar subjects respond to competing frames.
Directions for Future Research
Our research suggests that expressed preferences for health reform can be moved by both structural factors and situational framing. Given the cross-sectional nature of our data, we cannot assess the stability of these opinions. Longitudinal research designs will be needed to demonstrate how preferences evolve over time, including as COVID-19 vaccines are rolled out and the US economy continues to recover. The US unemployment rate, which peaked at 14.8% in April 2020, had recovered to 6.7% by December 2020, suggesting that any increases in support for M4A among those who lost insurance temporarily could gradually fade. The inauguration of President Joe Biden together with unified Democratic control of Congress may also trigger “thermostatic” dynamics in public opinion, pushing some Republicans and Independent voters to rediscover opposition to universal health programs. Thus, while the mass layoffs stemming from the COVID-19–induced recession may have presented an opportunity for proponents of M4A plans to make the case for the need to decouple insurance from employment, it remains unclear whether this message—and the life experiences that can generate receptivity to the message—can enduringly move the needle on public support for M4A or other UHC programs.
The widespread use of situational frames by political elites in a fragmented media market has given rise to concern about how “frame contests” may be contributing to growing political polarization in the United States (e.g., Baum 2011). An additional line of recommended research is to investigate the stability of health reform preferences not just over time but when exposed to counterarguments. That is, are situational frames pointing to problems with tying insurance to employment sufficiently convincing to inoculate against counterframes that paint M4A in a negative light? Future studies will have to gauge how resilient this new framing is to counterarguments (for instance, frames suggesting that countries with universal health coverage have had high mortality from COVID, or have had to ration care during the crisis).
Our findings also highlight that the broad term Medicare for All may mean different things to different people, and it does not necessarily equate with the idea that insurance coverage should be decoupled from employment. The findings of our secondary outcome analysis suggest uniquely positive features of the “Medicare” label, as these were only abstract concepts that gained support in response to experimental priming (compared to “universal health coverage” or “national health programs”). At the same time, however, we observe that actual loss of insurance was associated with increased approval of the Affordable Care Act/Obamacare rather than M4A, again suggesting that respondent experience with actual programs plays a large role in their opinions.
We also find major differences in magnitudes when comparing experimental versus personal experience of insurance loss. This highlights important methodological trade-offs in research design. While survey experiments generate strong internal validity, the larger effects, and differential patterns of heterogeneity, of our nonexperimental estimates are a reminder that real-life exposures are likely more powerful—and of much greater interest—than differences in issue framing generated by researchers. However, residual confounding of these estimates remains a possibility. Longitudinal study designs could shed further light on these questions.
A limitation of the study is that while the “structural” factor—job loss—was unexpected for many, given the unexpected nature of the pandemic, it may be subject to residual confounding; despite extensive covariate adjustment, the associations between health insurance loss and M4A approval cannot be interpreted as causal. In the experimental component, treatment was randomly assigned, although differential attrition with respect to respondent age may also bias point estimates. We mitigate the impact of this imbalance by controlling for age. We also note that, since attrition of younger respondents was higher in the treatment groups, and since younger respondents are on average more favorable to M4A, this imbalance may work against the likelihood of finding treatment effects. This survey experiment is also limited by the controlled environment in which it was implemented: respondents were not exposed to counterframes; as a result, we cannot assess how similar subjects would respond to the COVID-19 priming in the presence of competing frames. A further limitation relates to generalizability: perhaps reflecting the online recruiting modality, the sample in this article is younger, more likely to be unemployed, and more likely to have lost health insurance recently than the US population as a whole. The relationships identified in this sample may be weaker in older and more stably employed populations. A final limitation is that, while all experimental analysis, including subgroup analyses, were preregistered, observational analyses of the association between insurance loss and M4A support were not preregistered and should be interpreted as exploratory in nature.
We find that sympathetic framing of job loss and its association with insurance loss can bolster support for M4A, but that actual experience of insurance loss increases support for universal health coverage options more. Whether COVID-19 might tip the balance toward broader support for Medicare for All, the Affordable Care Act, or similar proposals will likely hinge on whether affected groups begin to perceive a stake in the programs, particularly the millions of people who lost employer-sponsored coverage in 2020 (Jacobs and Mettler 2011). With a new presidential administration, health care policy will continue to evolve. Our research suggests that while appealing framing can help, concrete benefits delivered by programs, rather than more effective messaging, are the most promising path toward generating a broader consensus around universal health coverage programs in the United States.