POLICY UPTAKE AS POLITICAL BEHAVIOR

Recent estimates suggest that large proportions of Americans do not enroll in public programs through which they are eligible to receive substantial benefits. For instance, according to a report by the U.S. Department of Health and Human Services (2007), uptake ranges from 75% of the eligible population for the Earned Income Tax Credit (EITC) to 55% for the Supplemental Nutrition Assistance Program (SNAP), 46% for Supplemental Security Income (SSI), and just 42% for Temporary Assistance for Needy Families (TANF). These choices can be extremely consequential; for example, the average individual who fails to claim the EITC loses out on about $1,096, which is equal to roughly a month of income (Bhargava and Manoli Reference Bhargava and Manoli2015).

A rational model of policy uptake suggests that eligible individuals choose not to enroll in a given program because they perceive the potential benefits to be outweighed by the associated costs (e.g., Moffitt Reference Moffitt1983). In the case of health insurance enrollment, we can understand the choice to abstain by considering risk tolerance and intertemporal substitution; those who choose not to enroll may simply prefer to take their chances with illness or injury, rather than making regular payments into an insurance pool (Barsky et al. Reference Barsky, Juster, Kimball and Shapiro1997).

In addition, studies of uptake have pointed to a variety of cognitive factors, or “psychological frictions” (Bhargava and Manoli Reference Bhargava and Manoli2015), that shape whether individuals enroll in programs for which they are eligible (Currie Reference Currie, Auerbach, Card and Quigley2006). First and most obviously, individuals must be aware a program exists and know they are eligible in order to take part in it (Chetty, Looney, and Kroft Reference Chetty, Looney and Kroft2009; Chetty and Saez Reference Chetty and Saez2013; Smeeding, Phillips, and O'Connor Reference Smeeding, Phillips and O'Connor2000). They must subsequently sign up for the program, rather than succumbing to procrastination or being unable to successfully navigate the enrollment process (Madrian and Shea Reference Madrian and Shea2001). For many public programs, the enrollment process can involve substantial transaction costs and so requires individuals to possess the cognitive tools, disposition, and capacity to navigate required paperwork (Bertrand, Mullainathan, and Shafir Reference Bertrand, Mullainathan and Shafir2006; O'Donoghue and Rabin Reference O'Donoghue and Rabin1999; Mullainathan and Shafir Reference Mullainathan and Shafir2013; Ericson and Starc Reference Ericson and Starc2012). These factors are not independent of the objective costs and benefits associated with a given program. Individuals are more likely to be aware of and enroll in a program when it offers larger benefits (Daponte, Sanders, and Taylor Reference Daponte, Sanders and Taylor1999; Blank and Ruggles Reference Blank and Ruggles1996).

Additionally, the social construction of a policy's target population might either encourage or discourage enrollment (Moffitt Reference Moffitt2003; Currie Reference Currie and Borjas2000). Perceptions of stigma associated with a given program can reduce the likelihood that eligible individuals will apply. Conversely, positive social evaluations might encourage eligible groups to take advantage of benefits. For example, we might expect the GI Bill, which communicates public esteem for veterans (Mettler Reference Mettler2007), to have a higher uptake rate than cash assistance programs, which negatively stereotype low-income participants (Gilens Reference Gilens2000). Again, a theory of policy uptake based on social construction is not mutually exclusive from one that emphasizes the role of transaction costs, since enrollment processes that require individuals to divulge a great deal of personal information may increase feelings of stigmatization among potential recipients and thereby decrease uptake (Currie Reference Currie, Auerbach, Card and Quigley2006).

While these existing explanations for incomplete uptake are useful, they ignore the important role that political partisanship might play in predicting whether and when individuals take advantage of public benefits. Political scientists have definitively shown that partisanship is a fundamental component of individual orientations toward the policy world. One's partisan loyalties influence patterns of information seeking and information processing (e.g., Zaller Reference Zaller1992), shape political identities and networks (e.g., Jennings and Stoker Reference Jennings, Stoker and Zuckerman2005; Weatherford Reference Weatherford1982; Green, Palmquist, and Schickler Reference Green, Palmquist and Schickler2004), and predict campaign contributions, vote choice, and a host of other important outcomes (e.g., Campbell et al. Reference Campbell, Converse, Miller and Stokes1960). Moreover, research has shown that partisanship is not dynamic, but is most frequently an “enduring attachment” (Campbell et al. Reference Campbell, Converse, Miller and Stokes1960)—a social identity akin to a religious affiliation or ethnic group membership (Green, Palmquist, and Schickler Reference Green, Palmquist and Schickler2004). This identity is established at an early age, is extremely stable throughout the life course, and is largely unaffected by other political attitudes (Niemi and Jennings Reference Niemi and Jennings1991).

Individuals also use partisanship as a primary heuristic in determining their public policy preferences. Rather than gather all the relevant information about a policy—such as whether and how it will benefit them and those like them, or what taking advantage of benefits will cost them relative to what they will gain—citizens employ partisanship as a shortcut for political decision-making. Essentially, partisans in the electorate can save time by adopting the policy positions of their preferred political party (or the opposite position from the opposing party) as their own (e.g., Lenz Reference Lenz2012).

Given partisanship's central role in American political behavior, it is not hard to imagine that it might also be brought to bear on other areas of government engagement that have not yet been explored, such as enrolling in a social program or complying with program requirements. There are a variety of ways in which partisanship might operate to influence policy uptake. Most obviously, Republicans might be less likely to sign up for government benefits, all else equal, because they are ideologically opposed to public-sector growth. In a 2012 Gallup poll, a large majority (82%) of Republicans voiced the belief that the government is doing too much; only 15% of this group took the opposite position that government should do more. Among Democrats, patterns were reversed (Newport Reference Newport2012).

Republicans also may be less likely to participate in public programs because they perceive the federal government to be incompetent, inefficient, or corrupt. In a 2010 Pew Research Center survey, 81% of Republicans concurred that government waste is a major problem, compared to 58% of Democrats who gave this response. Partisan differences similarly appear in other evaluations of government performance: in a 2015 survey, Republicans were 32 percentage points more likely than Democrats to believe that the federal government does a poor job running programs, and 31 percentage points more likely to say that government needs major reform (Pew Research Center 2010, 2015).

Apart from ideology, partisanship might also affect uptake through its salience as an in-group identity (Green, Palmquist, and Schickler Reference Green, Palmquist and Schickler2004). Perceptions that co-partisans are abstaining from enrollment or oppose the existence of a program may lead potential beneficiaries to believe that a program is primarily targeted at social groups more strongly aligned with the other party. In addition, co-partisan elites might explicitly discourage enrollment by criticizing the program publicly, attempting to repeal it, or refusing to be associated with it. It is also possible that partisans learn specific facts (or acquire misinformation) from co-partisans about the costs and benefits of a particular program that can either encourage or discourage participation. For instance, Republicans are likely to be more open to information that comes from other Republicans (e.g., Zaller Reference Zaller1992), who may be more likely to emphasize the costs of a program they dislike than to talk about its benefits.

THE AFFORDABLE CARE ACT: A CASE OF PARTISAN UPTAKE

Taken together, previous work on both partisanship and policy framing prompt two hypotheses about the role of party identification in the case of the Affordable Care Act. First, we expect that partisanship will be a strong predictor of policy behavior. In the case of the ACA, we anticipate that Republicans—who on average are much less supportive of the health insurance reform and are generally more resistant to government intervention in the private market—will be less likely than Democrats to take advantage of health insurance options provided by the ACA. Second, we posit that policy framing matters for policy uptake. We expect that a private or free-market frame that de-emphasizes the role of government will increase overall insurance enrollment among Republicans and conservatives, thereby closing the partisan uptake gap.

The Affordable Care Act provides an ideal testing ground for our partisan model of policy uptake. Since the beginning of the healthcare debate, there has been considerable political rancor over the policy reform that both connected directly to ideological schisms over government's reach and signaled in-party identity. As one political pundit concluded, remarking on the particularly bitter debate in Congress over the ACA, Republicans and Democrats “have existentially different views of the world” when it comes to what government can do well, what it should be doing, and how much of a role American government should play in the health policy domain (The Guardian 2011).

It is therefore not surprising that partisans in the mass public hold distinct views of Obamacare; by staking out starkly opposing positions on healthcare reform, partisan elites sent explicit signals to the public about where they should stand. From July 2010 through January 2016, the party gap in overall favorability toward the ACA remained stable and large: approval of the law hovered between 60–70% for Democrats and between 10–20% for Republicans throughout this period. Republicans were also significantly more likely than Democrats to say they would like to see the ACA repealed (Kaiser Family Foundation 2010–2016).

This persistent gap in attitudes provides an opportunity to test whether political partisanship influences individual decisions to take advantage of the benefits offered by the ACA. Certainly, Republicans have been more likely to claim that they will not participate: A 2013 Gallup poll of uninsured Americans found that 45% of Republicans reported they would choose to remain uninsured despite the financial penalty, relative to only 15% of Democrats. Descriptive evidence suggests that these attitudes may well translate into (in)action: a blog post by Michael Tesler shows that Democrats’ uninsured rates were halved from 2013 to 2015, while Republicans’ uninsured rates held fairly steady. In fact, Tesler finds that by mid-2015, Republicans actually surpassed Democrats in their likelihood of being uninsured, despite a long history of having more coverage (Tesler Reference Tesler2015).

We confirm and substantially extend this analysis using the Kaiser Family Foundation (KFF) Health Tracking Poll from 2010 through 2016. The KFF surveys, conducted every month since 2010, are fielded among a nationally representative, random digit dial telephone sample of adults ages 18 and older living in the United States. They thus provide an excellent source for examining cross-sectional trends in insurance coverage and post-ACA uptake. Figure 1 illustrates the percentage of Democrats and Republicans (including leaners) reporting they are uninsured at the time of each monthly poll.Footnote ¹ As in previous work, we find that the uninsured rate for Democrats fell faster than for Republicans. However, our data suggest that by the later period, which was not included in previous analyses, there is again a larger share of Democrats than Republicans who remain uninsured (Kaiser Family Foundation 2010–2016).

FIGURE 1.

Percent Uninsured by Party Identification (2010–2016)

Source: Kaiser Family Foundation Health Tracking Surveys, 2010–2016. (N = 35,965 Democrats; 31,162 Republicans, weighted for national representation). Figure represents average number of respondents, by party identification, that report being uninsured. Democrats’ uninsured rates dropped quickly at the opening of the ACA marketplaces, while Republican uninsured rates remain steadier.

Using these same KFF survey data, but restricting to the period after establishment of ACA marketplaces in 2014, we next examine the relationship between partisanship and health insurance enrollment (relative to the option of remaining uninsured). We focus our analysis exclusively on the uptake behavior of individuals who do not receive health insurance coverage through the government or through their employers, since this population is most likely to use the marketplace. Moreover, this helps control for the fact that Republicans are more likely to be covered by their employers than Democrats.Footnote ²

Individuals without an existing source of coverage like employer-sponsored insurance, Medicare, or Medicaid must choose amongst three options under the ACA. First, they can opt to refrain from obtaining health insurance and pay the tax penalty, which has increased in each year that the marketplaces have been operational (Kaiser Family Foundation 2015). Second, they can purchase insurance through the marketplaces established by the ACA. Third, they can purchase insurance off-marketplace, either directly from insurers or through brokers selling off-marketplace plans.Footnote ³

Using a multinomial logistic regression,Footnote ⁴ we estimate the effects of partisanship on uptake behavior for this group of individuals. Because there are a variety of differences between Democrats and Republicans that might predict insurance status, we control for potential confounders, including age, race and ethnicity, gender, state of residence, employment, education, and income (as well as the date of the poll). Previous research has illustrated that these factors are strongly related to health insurance status, which makes sense given existing models of insurance uptake. For instance, if individuals are not taking advantage of the ACA because the cost of coverage is too high, we might expect income to predict insurance enrollment. Employment status might be similarly implicated, either because the cost is too high for those who lack steady income or because they anticipate gaining coverage in the near future from a prospective employer.

In addition, our control variables reflect considerations that have been shown to matter for uptake rates more broadly. For instance, if uptake is primarily about “psychological frictions” and cognitive capacity (Bhargava and Manoli Reference Bhargava and Manoli2015; Currie Reference Currie, Auerbach, Card and Quigley2006), we might reasonably expect education to predict insurance status. Similarly, if average uptake within one's social group influences individual decision-making (Aizer and Currie Reference Aizer and Currie2004), we might expect to see variation across age categories or racial groups. Including these control variables helps us to evaluate these alternative hypotheses in our data. Table A2 in the Online Appendix reports the full set of marginal effects of the model covariates on individuals’ uptake behavior.

As expected, people with higher incomes are less likely to report being uninsured and more likely to report that they purchased off-marketplace plans. This result makes sense given that higher income individuals are less likely to benefit from federal subsidies available to marketplace participants. In addition, we find that Blacks and Hispanics are more likely than Whites to be uninsured and less likely to purchase either marketplace or off-marketplace coverage.Footnote ⁵ The unemployed are also more likely to be uninsured and less likely to purchase marketplace or exchange plans. Finally, the evidence suggests women are less likely to be uninsured, and more likely to purchase marketplace plans than men. While these findings validate extant research regarding other factors that influence coverage, what is important for our purposes is that party remains significant even when we control for all of these potential confounders.

Results from this model reveal that partisanship is a meaningful factor in policy uptake (presented in Figure 2). First, we find that partisanship has a strong and statistically significant relationship to whether individuals have health insurance. The evidence indicates that, all else equal, Republicans are 6 percentage points more likely than Democrats to be uninsured. Second, we find that partisanship is strongly associated with opting to purchase marketplace plans. With the full set of controls, we find that Republicans are fully 12 percentage points less likely to purchase marketplace plans than Democrats.

FIGURE 2.

Impact of Partisanship on ACA Insurance Enrollment

Point estimates are marginal effects of Republican partisanship on uptake behavior for individuals without group coverage derived from a multinomial logistic regression (N = 3,728), controlling for age, race and ethnicity, gender, state of residence, employment, education, income, as well as the date of the poll. Education is coded as high school or less (1), some college (2), or college + (3); income is coded as an eight-level categorical variable ranging from less than $20k to $100k+. Data are compiled Kaiser Family Foundation Health Tracking Surveys starting in 2014.

Finally, we find that partisanship is strongly associated with whether individuals purchase off-marketplace plans, with Republicans estimated to be 7 percentage points more likely than Democrats to purchase plans sold outside of the ACA marketplaces. In other words, being a Republican (relative to a Democrat) increases the likelihood that an individual will either be uninsured or purchase off-marketplace insurance, as opposed to signing up for an insurance plan offered through the ACA marketplaces.Footnote ⁶

It is difficult to get at the mechanisms underlying the effect of partisanship on uptake in our observational analysis, as this would involve examining whether the partisan effect on enrollment is driven by ideological factors or in-group identity. However, we are able to assess whether party identity and liberal-conservative ideology have independent effects on uptake. We run two additional models with the same full set of controls – one that includes both liberal-conservative ideology and party identity, and one that only includes ideology. We find that the results for conservatives mirror the results for Republicans. Both groups are more likely to remain uninsured, less likely to enroll in the marketplace, and more likely to purchase private insurance. When both party and ideology are included in the model, some of the impact of the ideology variable is absorbed by party. In this model, we find that they both matter, but party matters slightly more. (Republicans are roughly 10 percentage points less likely to enroll in the marketplace compared to Democrats, and conservatives are 6 percentage points less likely to enroll compared to liberals.) Details of this analysis are provided in Table A3 in the Online Appendix.

Finally, we use administrative data to further illustrate the persistence of the partisan policy uptake gap at the county level. We start with a model developed by the Kaiser Family Foundation that estimates the number of marketplace eligible individuals at the Public Use Microdata Area (PUMA) level (Levitt et al. n.d.). The model estimates marketplace-eligible individuals by starting with those who are uninsured or purchased non-group coverage, and then excluding the Medicaid-eligible, the undocumented, uninsured individuals who have offers for employer coverage, and the uninsured with incomes below the poverty level. The model then matches HHS marketplace enrollment data to PUMA-level estimates of eligible population to estimate the share of the eligible population enrolled. We spatially match PUMAs to counties and estimate the relationship between Obama's presidential vote share in 2012Footnote ⁷ and share of eligible individuals who signed up for marketplace coverage in 2015.Footnote ⁸ Figure 3 presents a scatterplot of the PUMA-county matches, with the upward-trending locally weighted smoother demonstrating the positive relationship between Obama's 2012 vote share and marketplace enrollment. Unlike the previous analysis, this model has the advantage of not relying on self-reported insurance status.

FIGURE 3.

2012 Democratic County Vote Share and Marketplace Enrollment

Points represent PUMAs and counties (N = 852) geographically composed as discussed in Footnote 6. The upward-trending locally weighted smoother demonstrates the descriptive relationship between percentage vote for Obama in 2012 and share of eligible population, as estimated by KFF, enrolled in marketplace plans in 2015. 10

Recognizing that Democratic-voting counties are different from Republican-voting counties in a number of ways, we extend this descriptive analysis by controlling for variables that our previous analysis and existing theory suggest are associated with Democratic vote share and marketplace enrollment.Footnote ⁹ These include median income, rural-urban continuum code, unemployment rate, share of individuals reporting poor or fair health, percent uninsured in 2013, percent Black, percent Hispanic, median years of college, population, percent of population under 18, percent of population over 65, number of plans offered on the marketplace in 2014, benchmark silver-level premium in 2014, and state fixed effects. The descriptive analysis is corroborated by this multivariate regression analysis, reported in Online Appendix Table A4 and Online Appendix Figure A2. We estimate that a 10 percentage point swing towards Obama in 2012 is associated with an approximately 2 percentage point increase in the share of the marketplace-eligible population enrolling. This analysis also indicates that more rural counties tend to have lower enrollment, while income and number of plans offered in 2014 are positively associated with enrollment.

POLICY FRAMES AND PARTISAN UPTAKE

The data make clear that there is a significant gap between the uptake rates of Republicans and Democrats. Additionally, we find that Republicans have been significantly less likely to enroll for health insurance through the government exchanges, even when they do procure health insurance, and these effects continue to hold when we control for potential confounders. We have suggested, however, that this stark inequality in the rate of uptake between Democrats and Republicans is not necessarily a fixed aspect of the debate over healthcare. Instead, asymmetrical partisan uptake might be a reflection of a prevailing policy frame, which stresses both the government's and the Democratic Party's role in regulating and expanding the private insurance marketplace. In framing the ACA, Republican elites have taken pains to link the reform to federal government, the Democratic Party, and President Obama's administration; for instance, the National Review called the reform package “the poster child for Washington arrogance” and an “unprecedented overreach” by government in Washington (Flores, Roe, and Scott Reference Flores, Roe and Scott2015). In a Pew Research Center survey about the ACA, 88% of Republicans expressed concern about the government becoming too involved in health care, compared to just 37% of Democrats (Pew Research Center 2012).

In order to test whether uptake is responsive to policy framing, we partnered with Enroll America, a national, non-partisan health outreach organization whose goal is to maximize the number of Americans enrolled in and retaining health insurance coverage. In a large-scale field experiment, we randomly assigned a sample of uninsured individuals to one of two websites through which they could enroll in the healthcare exchanges: either a private website called HealthSherpa.com, or the government's website, Healthcare.gov. This experiment provides an excellent test for our hypothesis, given that it partners with an external organization in testing actual treatments in a field experimental context.

The processes by which individuals can register through the two sites are parallel and comparable; individuals on both sites are asked to enter their zip codes and a variety of demographic information before an assortment of plans are presented. However, the design of Healthcare.gov makes clear that the healthcare exchanges are government initiated. The website explicitly mentions the “Affordable Care Act,” “Regulatory and Policy Information,” and “Tax Info and Tools”; it emphasizes that the site is a “federal government website managed by the U.S. Centers for Medicare and Medicaid Services”; and it includes a symbol of the White House and USA.gov branding (see Online Appendix Figure A3).

In contrast, HealthSherpa.com emphasizes the private nature of the exchanges, explicitly stating that the web portal “is not affiliated with any lobbying or trade group, or any government agency, and has no political agenda.” Instead, it describes itself as a neutral site for “consumers,” providing choice among “innovative products that help consumers easily understand, sign up for and use health insurance.” Imagery on the private site includes photographs of doctors and patients, families in medical waiting rooms, and the logos of major private insurance providers, including Cigna, UnitedHealthcare, and Kaiser Permanente (see Online Appendix Figure A4).

Our experimental sample consists of individuals drawn from the twelve states (AL, AZ, FL, GA, NC, NJ, OH, PA, TN, TX, IL, MI) in which our partner organization, Enroll America, maintained a field program during the 2014–2015 open enrollment period. These states were chosen by the organization because they were identified as having particularly large proportions of uninsured individuals. Ten of these states used the federal exchange, while two states—Illinois and Michigan—had state-partnership marketplaces.

Individuals in these states were brought into the sample via two sources. First, Enroll America developed a proprietary statistical model that predicted the probability that individuals in their database were uninsured. People with a high probability of being uninsured were targeted through the field program and asked to fill out a card committing to enroll in health insurance. These individuals were then recontacted and directed to our partner organization's website. The second source of the sample consisted of people who went online to our partner organization's website of their own accord and filled out a form to get more information about enrollment. From December 2014 through February 2015, roughly 20,000 people went through the digital platform of our partner organization to express interest in getting coverage. All individuals who came to our partner's website were then randomly assigned by zip code to one of the two sources of access to ACA marketplaces: the government website, Healthcare.gov, or HealthSherpa.com, the private site.

The information individuals provided through our partner organization's website was matched back to the organization's database, which contained additional data about the individual from public records and consumer files. In addition, we conducted two follow-up surveys to assess enrollment, the first survey online and the second by phone. The online survey was conducted in February and March 2015. Participants in the online survey received an email notification asking them to participate. Invitations were sent to the full email list maintained by our partner organization, and the survey yielded 1,329 participants. Of survey participants, 388 lived in zip codes assigned to Healthcare.gov and 941 lived in zip codes that were assigned to HealthSherpa.com.

In the subsequent phone survey, we selected a random sample of individuals with active phone numbers and attempted to contact them three months after the close of the open enrollment period in May and June 2015. Calls were made to a total of 4,972 individuals who resided in zip codes assigned to Healthcare.gov; 884 of these individuals completed the phone survey. Calls were made to 6,090 individuals residing in zip codes assigned to HealthSherpa.com, and 1,144 of these individuals completed the survey. (Our research design is summarized in Online Appendix Figure A5. Demographic characteristics of the analysis sample are described in Online Appendix Table A5.) Despite differences in mode, sampling, and response rates, results across the two surveys were substantively equivalent and we therefore combined results whenever possible (i.e., where we asked identical questions across the two surveys).

Our research design relies on the assumption that treatment assignment did not influence participants’ willingness to respond to the survey. This is difficult to evaluate formally, however, as it requires knowledge of individuals whom we were not able to contact in follow-up. We had access to data on partisanship from public records for the individuals contacted by phone, but there is considerable missingness in the data; the majority (76%) of the sample we attempted to contact by phone is labeled as partisanship “Unknown.”

However, for those whose partisanship we can identify in these records, we find little evidence of response bias. Within the phone sample, 20.2% of Democrats made it halfway through the survey, while 22.6% of Republicans did the same (p = 0.28). With respect to treatment assignment, 14.4% of those assigned to Healthcare.gov made it through the mid-point of the survey, as did 15.1% of those assigned to HealthSherpa.com (p = 0.32). There is also no evidence that treatment interacted with partisan identification to determine survey participation: 18.5% of Democrats assigned to Healthcare.gov completed at least half of the survey, while 21.5% of Democrats assigned to HealthSherpa.com did (p = 0.12), and 26% of Republicans assigned to the government site made it to the mid-point, as did 20.4% of Republicans assigned to HealthSherpa.com (p = 0.20).

One potential concern is that Republicans assigned to the HealthSherpa.com website may have been more open to talking about their enrollment experience relative to those assigned to the Healthcare.gov site. However, both follow-up surveys were conducted months after people received the treatment and respondents had no knowledge that they were part of an experiment. Moreover, respondents in the phone poll were told they were being contacted for a study “from UC Berkeley on healthcare” as opposed to the partner organization, Enroll America, further obscuring the connection between respondents’ enrollment experience and the follow-up survey itself. Respondents that agreed to participate were also not aware at the outset that the survey specifically pertained to insurance enrollment.

Differential attrition based on treatment can also be investigated by looking for imbalances on observed covariates across treatment and control groups within the contacted sample. We evaluate balance by regressing treatment assignment on race, income, party, education, gender, age, and orientations towards government. The analysis indicates that treatment assignment is not associated with observed covariates (p = 0.32), suggesting that our randomization strategy was successful.Footnote ¹¹

In our analyses, we restrict the sample to include only individuals who either remained uninsured or enrolled through the online state and federal insurance exchanges (N = 1,837), excluding those who enrolled in insurance through an employer-sponsored plan or off the marketplace.Footnote ¹² This sub-sample is fairly representative of the population from which it was derived. The average age of sampled respondents is 44.5 years, 69% of respondents have a college degree, and 63% are male. As expected, the Republican sample (n = 178) is comparable to the national population of Republicans in that it is slightly older and wealthier than the sample of Democrats and Independents.