Context: A decade after passage, a majority of Americans now support the Affordable Care Act (ACA), and Republican efforts to repeal it outright have failed. This article investigates whether the policy itself, through its beneficiaries, changed public opinion and sowed the seeds of its defense.
Methods: This study used an individual-level panel design to estimate the causal effect of implementation on opinion and electoral outcomes for ACA beneficiaries during the first year of open enrollment.
Findings: Individuals who enrolled in plans on the health insurance marketplaces had significantly more positive opinions of the ACA after implementation. Previously uninsured Medicaid enrollees also reported improved opinions, though results were not statistically significant. In contrast, uninsured individuals residing in states that did not expand Medicaid became significantly less supportive of the law. Changes in opinion persisted up to the 2014 midterm elections, and there is evidence that individuals with marketplace insurance became more supportive of Democratic candidates, although not more likely to vote for them.
Conclusions: Public support for the ACA was enhanced when its beneficiaries became more positive toward it during implementation. Recent changes to key ACA provisions have the potential to undermine the law's effectiveness, potentially leading to political action as benefits slowly begin to disappear.
In an era of political polarization, partisanship is a primary determinant of American opinion on policy. Voters increasingly have policy preferences that are in line with their preferred party's positions across a broad range of issues (Abramowitz and Saunders 2008; Hare et al. 2015; Layman and Carsey 2002; Layman, Carsey, and Menasce Horowitz 2006). The Affordable Care Act (ACA) is no exception; support for the law divides along party lines (Chattopadhyay 2017; Gelman, Lee, and Ghitza 2010; Jacobs and Mettler, 2016; Tesler 2012). Until recently, a bare majority of Americans consistently opposed the law (Fingerhut 2017). Only a fraction of Americans expect to benefit from the ACA (Chattopadhyay 2017). Republicans running on a platform to dismantle the ACA made gains in the 2014 and 2016 national elections, enabling them to secure the presidency and majorities in both chambers of Congress (Nyhan et al. 2012). Nevertheless, the 2017 Republican effort to repeal the ACA was unsuccessful.
This article provides a rare empirical test of whether policy, through implementation, can reshape partisan attitudes. Using longitudinal data from the American Life Panel (ALP), I followed the same individuals repeatedly over the first year of ACA implementation, from September 2013 until the 2014 midterm election. I observed whether individuals enrolled in health insurance and the type of insurance they obtained and then how this experience affected their opinion of the law and voting behavior in the subsequent midterm election. This rich panel data allowed me to estimate the causal effect of policy implementation on policy preferences and vote choice while controlling for previous ACA opinion.
Of primary interest are those who bought insurance on the marketplaces and enrolled in Medicaid between the start and end of the 2014 open enrollment period. Since the primary goal of the ACA was to increase the rates of insurance through the creation of the marketplaces and Medicaid expansion, I also consider how changes in attitudes for these groups depended on whether an individual was previously uninsured. In addition, the ACA implemented a small reduction in the prescription drug doughnut hole during the open enrollment period, which may have caused Medicare enrollees to become more favorable toward the law. There were no significant impacts to other insurance groups, so we should not expect attitudes to have changed for other insurance types during the open enrollment period.
Consistent with these predictions, I found that individuals who enrolled in plans on the health insurance marketplaces had significantly more positive opinions of the ACA by the end of the enrollment period than they did before enrollment began. To a lesser extent, the opinions of Medicare enrollees also significantly improved. Medicaid enrollees who were previously uninsured reported improved opinions, but this change was only statistically significant at the 10% level, possibly due to sample size. In contrast, individuals who resided in states that did not expand Medicaid and failed to obtain insurance developed significantly more negative opinions of the ACA. No other defined insurance group stood to gain as these groups did, and none of these groups reported a significant change in ACA opinion. Changes in opinion persisted into August 2014, and there is evidence that individuals with marketplace insurance became more supportive of Democratic candidates in the 2014 midterm elections, although not more likely to vote for Democratic candidates.
ACA Benefit and Policy Preferences
The ACA represents the most significant change in health care policy in the past 40 years, enabling millions of Americans to obtain health insurance. The ACA provided a new source of health insurance with the creation of the ACA marketplaces (also called exchanges), and it reformed and extended other elements of the existing health insurance system. This article considers whether individuals who obtained benefits under the law changed their opinion of the law after receiving those benefits during the 2014 open enrollment period. I estimated the causal effect among individuals who gained insurance and insured individuals who also benefited from ACA implementation. Democrats were counting on attitudes toward the ACA, which were overall negative, improving once implementation began as individuals received new benefits under the law, an important indicator of the law's success.
Studies suggest that ACA beneficiaries have more positive views of the ACA compared to the general population (Jacobs and Mettler 2016; McCabe 2016). Cross-sectional data, however, cannot disentangle the effect of policy benefits on policy preferences from factors such as partisanship (Campbell 2012). Several studies have used creative designs that attempt to overcome this causal inference problem. Lerman and McCabe (2017) compared the attitudes of individuals just above the age of 65 to those just below and found that individuals just above the Medicare eligibility threshold had more favorable opinions of Medicare and the ACA before implementation. Medicare enrollees are one group that stood to benefit from ACA policy, so their positive attitudes may reflect anticipation of new benefits under the law. Using state-level panel data, Sances and Clinton (2017) found that high school graduates in expansion states and other groups that were likely to benefit from the law had marginally more favorable opinions of the ACA after implementation, although there was no difference in the overall population. Jacobs and Mettler (2016) confirmed that there did not appear to be a significant difference in ACA opinion over time between residents of Medicaid expansion and nonexpansion states.
Though rare, studies using individual-level panel data enable estimation of the change in opinion as a result of an event while controlling for preexisting opinion. In one such study, Lenz (2013) observed that individuals conform their own opinions on policies to those of their preferred party upon receiving new information about the party's position, once those policies become salient (often in response to a news event). In contrast, Milazzo, Adams, and Green (2012) found that British citizens update their party evaluations to align with their existing beliefs about the government's fundamental role in social services, industry, inflation and unemployment, and redistribution. Jacoby (2014) found these measures to be closely aligned with long-held ideological preferences, so it is not surprising that they are less resistant to elite persuasion.
Regarding the ACA, McCabe (2016) used panel data from the 2012 and 2014 waves of the Cooperative Congressional Elections Study and found that Republicans who became insured during this 2-year period, which covers the first year of implementation, had more favorable opinions of the law than did individuals insured in both waves. Most individuals who became insured during the 2-year time period gained insurance through Medicaid or with the support of federal subsidies to buy insurance on the ACA marketplaces, so increased support for the law among the newly insured is likely a result of the ACA benefits that many in this group received. McCabe also reported that individuals who lost insurance became more negative toward the ACA.
Jacobs and Mettler (2018) fielded an individual-level panel survey with three waves in 2010, 2012, and 2014 to assess the effects of ACA implementation on public opinion. They found that individuals without insurance in 2014, either because they lost their insurance during the 4-year period or remained uninsured, were less favorable of the ACA than in 2012. In addition, they found that individuals who gained insurance between 2010 and 2014 or who obtained 2014 coverage through a government program were more likely to believe that the ACA improved access to health insurance and medical care. However, they did not find that subsidies to buy insurance or prescription drug cost reductions increased overall favorability of the ACA among individuals who benefited from these policies.
The Jacobs and Mettler 2018 result contrasts with the finding of this study that individuals who enrolled in marketplace insurance, usually with support from subsidies, increased their support of the ACA in the period between the start and end of the 2013–14 open enrollment period. Differences in the panel designs and statistical models may explain the difference in results, as well as differences in how the surveys define beneficiary groups. Jacobs and Mettler considered individuals who reported having benefited from “tax credits or other subsidies” stemming from the new health care law, while in this study I observed the type of insurance individuals obtained for 2014. Even though most individuals who purchased marketplace insurance benefited from the premium tax credit and cost-sharing subsidies, they may not have been aware that their insurance was made more affordable as a result of these benefits. In addition, those with Medicare or Medicaid insurance or young adults who obtained insurance through a parent's employer (also subject to tax credits) may have reported having benefited from tax credits or other subsides.
The study presented here builds on previous individual-level panel studies of ACA implementation in several ways. The data include direct measures of insurance status and insurance type, including measures for Medicaid, Medicare, and marketplace insurance, allowing for analysis of whether groups covered by insurance made available or enhanced by the ACA had changed opinions about the ACA as a result of implementation. In addition, the panel design encompasses a much tighter time frame around the first period of open enrollment under the ACA. In panel designs, the choice of time period embodies a trade-off between causal identification and the study of longer-term effects. The more closely the panel period encircles the policy change, the more convincingly one can attribute changes in public opinion and political behavior to specific policy effects and not to such characteristics as partisanship and economic circumstances. Both McCabe (2016) and Jacobs and Mettler (2018) had a 2-year lapse between panels, which requires stronger modeling assumptions to disentangle policy effects from partisan trends. However, the longer time span allowed these studies to assess a broader set of policy changes. My analysis provides a tight window around the first open enrollment period but misses the early phase-in of the prescription drug cost reductions and the requirement to extend employer coverage to young adults under age 26. Each study employed a different analysis design and definition of beneficiary groups and are complementary, together contributing to our understanding of the political impact of ACA implementation.
ACA Policy Changes Implemented in 2013–14: Impact on Insurance Types
Under certain conditions, individuals who personally benefit from a policy can discard partisan beliefs and form attitudes in line with their self-interest (Carsey and Layman 2006; Page and Shapiro 2010). The government implemented several policies during the winter of 2013–14, the first period of open enrollment; the roll-out of the marketplaces and Medicaid expansion were arguably the most significant reforms. Some states opened online health insurance marketplaces, and the federal government opened a marketplace on behalf of the remaining states. These marketplaces guaranteed individuals without employer-provided health insurance access to individual policies with comprehensive coverage at group rates, often with the help of federal subsidies. Subsides were provided at incomes up to 400% of the federal poverty line ($45,960 for a single person and $94,200 for a family of four in 2014), and the vast majority of marketplace enrollees qualified for a subsidy. Individuals in poor health could not be denied coverage in the marketplaces. In addition, the law provided additional federal funds to expand Medicaid coverage to individuals and families with incomes below 138% of the federal poverty level, although not all states accepted these additional funds.
These policies benefited the previously uninsured and the insured. An estimated 8.4 million nonelderly adults gained insurance in 2014, almost always through Medicaid expansion or plans individually purchased on the marketplaces (Skopec, Holahan, and Solleveld 2016). For the previously insured, the marketplaces provided individuals at risk of losing their coverage a guaranteed replacement. The individual market was reformed, offering greater coverage and/or reduced rates to some in this market. Medicaid expansion also reduced the risk of becoming uninsured among preexpansion Medicaid recipients. We should expect that individuals who obtained insurance on the marketplaces or through Medicaid to have more positive attitudes of the law, particularly if they were previously uninsured.
Sizable benefits may not change attitudes or behavior if beneficiaries are unaware of the benefits they receive and the government's role in providing the benefits. Mettler (2011) found that benefits distributed through the tax code or by private industry are less visible to beneficiaries than are direct service programs. In addition, expanding benefits in an existing program may have less effect on attitudes. Campbell (2003) argues that Medicare created a powerful and engaged voting bloc of elderly Americans dedicated to the preservation of this program. But subsequent changes to Medicare that expanded benefits did not lead to more positive attitudes (Morgan and Campbell 2011).
These examples suggest that we may expect Medicaid expansion to have a weaker effect on attitudes to the extent that incremental changes to policies are less visible than newly created benefits. Many individuals already enrolled in Medicaid may not have realized that the ACA reduced their risk of losing eligibility in the future. Individuals who became eligible for Medicaid as a result of the new federal funds to expand the eligibility threshold may not have been aware that they benefited from the ACA. Turnover among Medicaid beneficiaries is high (Ku and Steinmetz 2013). Beneficiaries frequently gain and lose coverage and may not have attached responsibility to the ACA for the change in their eligibility status in 2014. If so, the effect of the ACA on attitudes should be smaller for individuals who became insured though Medicaid than for those who purchased insurance on the marketplace. In the nonexpansion states, it is unclear how attitudes toward the ACA among Medicaid enrollees and individuals who would have qualified for Medicaid expansion were affected by state decisions not to implement expansion.
I also consider how ACA attitudes changed over this period among enrollees in other types of health insurance plans, including Medicare and employer plans. The ACA enacted policies that affected other health insurance plans, but these policies had little effect during the period of open enrollment in 2013–14. Prior to the act, Medicare beneficiaries faced a “doughnut hole” in Part D drug coverage in which they paid 100% of the costs of drugs. Beginning in 2011, the act gradually reduced cost sharing in the doughnut hole to 25% by 2020. During the first open enrollment period, the only effect of this policy change was to reduce cost sharing for generic drugs from 79% to 72%. The ACA also mandated free preventive care in Medicare and employee plans, but this policy took effect earlier in 2011 and is captured by the preenrollment attitude variables in my analysis. Therefore, for the reasons explained above, we might expect at most weak effects of the incremental changes on attitudes for Medicare and no effects for employer plans.
The ACA imposed few costs on individuals during the initial enrollment period in 2013–14. The ACA's employer health insurance mandate, originally scheduled to begin in 2014, was delayed until 2015 and again for midsize employers until 2016. This provision requires firms with 50 or more full-time employees to provide a minimum level of health insurance coverage for their employees or pay a fine. Minimum coverage requirements also affected the insurance for individual coverage. In general, the opening of the marketplaces fundamentally changed the market for individually purchased insurance and the types of plans offered. As a result, individuals with individual coverage before 2014 may not have been able to renew their previous coverage. In addition, some individuals who were previously uninsured may have preferred to remain uninsured rather than pay for health insurance. The ACA's individual insurance mandate (repealed in 2017), which required individuals to obtain a minimum level of health insurance, initially carried a tax penalty of $95 per adult and $47.50 per child (up to $285 for a family), or 1% of household income above the 2014 tax return filing threshold, whichever is greater. The penalty was not due until 2014 taxes were filed in 2015. However, many individuals who did not obtain insurance coverage in 2014 did not face a penalty because the law had an exemption for low-income individuals. Regardless, public discussion of the mandates during the open enrollment period may have caused those most at risk for becoming uninsured to anticipate future costs and have lower opinions of the ACA. Jacobs and Mettler (2018) found that individuals who reported receiving subsidies to help pay for insurance were more likely to believe that the new health care law increased their tax burden.
In summary, the law's benefits in the first period of open enrollment were concentrated on two insurance types: marketplace-purchased health insurance plans and Medicaid. Medicare enrollees experienced a modest benefit. Other insurance types did not receive new benefits or pay significant costs during the first enrollment year, nor did most perceive any benefits or costs aside from a fear that they might face higher health costs in the future. If such a fear existed prior to the start of open enrollment, it would factor into prior attitudes. Since premium contributions for Medicare Part B were unchanged in 2013 and those for employer health plans increased at a rate comparable to previous years (Claxton et al. 2015), there was no reason to suggest that such fears would heighten during the period of open enrollment. Thus, we should not expect cost fears to impact ACA attitudes during the open enrollment period.
The data came from the American Life Panel (ALP), an ongoing panel that started in January 2006 and is representative of the US adult population. Documentation for the ALP can be found in Pollard and Baird (2017). Its probability-based sample consists of 4,200 respondents recruited from several sources between 2002 and 2012: (a) respondents to the University of Michigan Surveys of Consumers, (b) participants in a Stanford University Abt Associates panel study with a representative probability sample, and (c) a vulnerable-population sample selected by RAND using address lists in Zip codes with high percentages of Latino/as or low-income households. Initial participation rates were 30% for the Michigan cohort, 46% for the Stanford sample, and 42% for the vulnerable-population sample. Year-to-year retention rates among participants were high: of those who participated in a survey in 2013, 94% participated again in 2014. Completion rates for individual surveys vary considerably depending on the sample and topic. On average in 2015, participants completed 62% of the surveys they were asked to complete.
The panel includes several series of themed surveys. I primarily drew on the ACA series, which began in August 2013. I constructed a panel consisting of all individuals who participated in the baseline survey conducted from September 16 to October 1 2013, and who appeared in at least one of the monthly surveys conducted over the open enrollment period through May 2014 and again in August 2014. With this setup, I could observe respondent opinions of the ACA in the month before the start of the open enrollment period, the type of insurance they enrolled in during open enrollment, and their opinion of the ACA after enrolling or after open enrollment ended. The ALP also collected information on voting in the November 2014 midterm election. ALP estimates of enrollment decisions and 2012 election outcomes match well to other sources (Carman, Eibner, and Paddock 2015; Gutsche et al. 2014).
Insurance Coverage, 2014
The first postenrollment surveys were sent to all participants in October and November 2013, and follow-up surveys were sent each month from December 2013 through May 2014 to all participants. The surveys asked respondents about their insurance coverage for 2014. They could select one or more items from a list of thirteen different types of insurance. For individuals who had already reported insurance coverage in a previous wave, the December through May surveys asked participants to confirm their previous choice. I collected the final insurance type(s) each respondent selected, as well as the month in which the individual's final insurance type was recorded. Individuals were categorized as “uninsured” if they indicated that they did not have insurance or selected “no coverage of any type” as their type of insurance coverage.
For analysis purposes, I condensed the 13 insurance types into 9 categories to account for major overlaps in coverage groups, though the results are substantively the same regardless of whether I used the original 13 or condensed 9 categories. Table 1 lists the insurance types as they appear in the survey and my coding for the analysis.
Apart from the marketplaces, insurance enrollment occurred early in the ACA enrollment period. November is the median month that individuals without marketplace insurance report having enrolled for 2014. Individuals who purchased insurance on the marketplaces most often report having obtained insurance in April. The delay reflects the fact that many marketplaces experienced technical difficulty during the early enrollment months, as well as the fact that individuals were signing up for the first time.
Insurance Status, 2013
The ACA series in the ALP fielded two preenrollment period surveys that collected information about respondents' previous insurance histories. I further augmented this information with data from an early October survey on the effects of the financial crises, another ongoing ALP series. It was necessary to draw from all three surveys in order to collect 2013 insurance information for the maximum number of individuals in the sample.
The ALP contained the following three ACA opinion questions in each of the monthly ACA-series surveys from September 2013 through May 2014, and again in August 2014:
As you may know, a health reform bill (the Affordable Care Act, or Obamacare) will take effect in 2014. Given what you know about the reform law, do you have a generally favorable or unfavorable opinion of it?
Do you think you and your family will be better or worse off under the reform law or don't know?
Do you think the country as a whole will be better or worse off under the reform law or don't know?
Respondents could select very favorable, somewhat favorable, somewhat unfavorable, very unfavorable, or don't know for the first question. The second and third questions allowed respondents to select better off, worse off, not much difference, or don't know for their responses. I recoded all variables into categorical variables centered at zero. “Don't know” was coded as zero, or neutral, rather than dropped.1
Key variables in the analysis are pre- and postenrollment ACA opinion and 2014 insurance coverage type. Preenrollment ACA opinion came from the baseline September survey. Postenrollment ACA opinion came from the last survey in which the panelists participated during the postenrollment period. Although the possible dates for this last survey could range from November through May, the April or May surveys included over 70% of all individuals in each insurance type.
Data on preenrollment political variables came from several surveys fielded around the 2012 presidential election. Again, drawing data from multiple surveys ensured that the analysis included preenrollment data for as many panelists as possible. Information on party ID came from the ALP's Kimball pre- and postelection surveys and global warming survey, which asked about Democratic or Republican political affiliation and leaning. I combined these variables to create an indicator for whether or not an individual identifies or leans Democrat. Variables on vote choice and candidate choice were combined to create a variable indicating support for Obama in the presidential election over the other candidates. This information, along with reported voter turnout in the 2012 presidential election, was drawn from the three election surveys mentioned above, as well as another postelection survey. These surveys were fielded between October 2012 and May 2013.
I drew postenrollment political variables from the midterm election series conducted between October 10 and November 13, 2014. In surveys leading up to election day, respondents were asked to report the percent chance that they would support the Democratic, Republican, or Independent candidate in the House elections. This type of question has been shown to more accurately predict election outcomes than questions that simply ask for the respondent's top choice (Delavande and Manski 2010). The final survey of the series, conducted after the election, asked respondents who voted which candidate they voted for. I used their reported vote choice to create an indicator of vote for the Democratic candidate. I also used the party identification questions from this series to create a postenrollment Democratic Party identification dummy.
The ALP collects standard socioeconomic information from participants on a quarterly basis. In addition to insurance coverage and ACA opinion, the ACA series also included questions that changed each month about respondents' expectations of the law's impact on health care access and cost and the likelihood of having to pay a tax penalty.
Many factors affect individuals' attitudes toward the ACA, and these factors are also associated with whether they were previously insured and which type of insurance, if any, they obtained in 2014. Figure 1 shows average ACA opinion in the month prior to open enrollment by postenrollment insurance type. Individuals who were insured in 2014 through Medicaid or marketplaces were most positive about the law on all three questions in the preenrollment period. These individuals had lower income, were less likely to be previously insured, and were more likely to be Democrat. In comparison, higher-income, more Republican groups such as those enrolled in self-pay insurance or a military health care plan were less favorable to the law before the enrollment period.
I used an individual-level panel design to control for other factors, such as political variables, that affect ACA opinion and are correlated with insurance status and type. The panel design requires fewer assumptions to establish causality than do cross-sectional designs that include controls for preexisting characteristics or difference-in-difference designs that compare state trends over multiple periods. A key assumption of the panel design is that, conditional on preenrollment ACA opinion, changes in an individual's ACA opinion over the enrollment period results from the benefits (or costs) received as a result of ACA implementation. By this assumption, the difference between individual-level preenrollment opinion and postenrollment opinion subtracts out the effects of factors that are constant throughout the time period, such as political affiliation, ideology, and region, as well as the effect of anticipated benefits on ACA opinion. Individuals who were aware of the law and expected to benefit from it during the open enrollment period were likely to have already had favorable attitudes of the law. To the extent that their experience during open enrollment met their expectations, the attitudes of those who anticipated benefits would have remained constant through the period. Rather, the panel design allows us to understand how attitudes changed as a result of the law exceeding or falling short of individual expectations.
Panel attrition is mitigated to a certain extent by a tighter time frame in this study. Panel attrition could bias the results if individuals who exit the panel are systematically different from those who remain in ways that relate to insurance status and type, as well as ACA opinion. I compared individuals who dropped out of the panel to those who remained in the panel on preenrollment ACA opinion, 2013 insurance type, and socioeconomic and political characteristics for the three periods investigated in the analysis: to the end of open enrollment period, through August, and up to the 2014 midterm elections.
Between the preenrollment and postenrollment period, 6% of the sample dropped out of the panel. Individuals who left the panel had significantly more favorable preenrollment ACA attitudes, although there were no significant differences in socioeconomic and political characteristics among those who left and remained in the panel. There were no differences in 2013 insurance coverage, except that individuals with employer-provided insurance were somewhat more likely to leave the panel. The ACA did not have a direct impact on this group during the first enrollment period, and it is unlikely that their departure correlated with a change in ACA opinion over the enrollment period.
Between the preenrollment and August surveys, 22% of the sample left the panel. I found no difference in preenrollment ACA attitudes among those who left and those who remained or for any insurance, socioeconomic, and political characteristic.
Not all ACA-series panelists were asked to participate in the elections series. As a result, and because the panel spans a longer time period, the percent missing rises to 34% or 56% in the midterm election period, depending on whether chance of support for or vote for the Democratic candidate is the dependent variable. The missing panelists were significantly more likely to be Republican and to have more negative opinions of the ACA at the start of the open enrollment period. With a one-year time lapse between the start of open enrollment in October 2013 and the November 2014 midterm election, the election results are only suggestive of real effects.
Startup problems with the marketplace websites, including notable difficulties with the federal website, pose another potential challenge to the panel design (Oberlander and Weaver 2015). These problems were largely resolved before the end of 2013, but they may have caused postenrollment ACA opinion for respondents in November to be lower than it would have been in later months. These respondents were less likely to obtain marketplace insurance. I reran the analysis including only individuals whose opinion came from the April or May surveys, when the exchanges were functioning smoothly, and found no meaningful change in the results.
I modeled the relationship between type of insurance and postenrollment ACA opinion as
where yitv represents one of three ACA postenrollment variables and yi(t−1) is a vector of all three ACA preenrollment variables. hit contains eight dummy variables for whether or not the panelist obtained 2014 insurance through the following avenues: the marketplaces, Medicare, Medicaid, other government programs, individually not on the marketplace (self-pay), the US military, or other means. In addition, there is an indicator for whether the individual was unable to obtain insurance. Employer-provided insurance is the omitted variable. It made sense to make this group the baseline group since the majority of Americans receive their insurance through their or their spouse's employer. The ACA was designed to address gaps in the insurance system, and the law had no direct impact on this group during the period under investigation. Indeed, t-tests and bootstrap Kolmogorov-Smirnov tests indicate that there was no significant change in the attitudes of this group during the open enrollment period; the base group is stable, and the effects that we observe with respect to the included insurance groups are differential effects from a stationary base. The term zi(t−1) is a vector of preenrollment characteristics, including voted in 2012, supported Obama in 2012, black, female, and the log of income. The error term is ɛitv. The intercept term is αv, and βv, δv, and ηv are the coefficients associated with 2014 health insurance type, preenrollment ACA opinion, and preenrollment characteristics, respectively.
In addition, I considered whether the attitudes of those who obtained insurance through the marketplaces or Medicaid depended on previous insurance status and state of residence. We might expect the most significant change in attitudes to occur among those who were previously uninsured in these groups since they arguably benefited the most from ACA implementation of these policies. In this model, I omitted from the regressions individuals who enrolled in Medicare in 2014, because they were certain of being insured if over the age of 65 and in the near future if under 65 at baseline.
The propensity to become insured and the type of insurance enrolled in depend in part on state of residence. States varied considerably in their implementation efforts, and the basic model estimates the average treatment effects over all states. Thus, treatment is a function of the individual's experience during the enrollment period as a result of federal, state, and local ACA implementation efforts. In an extension of the basic model, I considered whether residence in a state that chose not to accept federal funds to expand Medicaid mediated the effects of being uninsured or obtaining insurance through the marketplaces, Medicaid, or Medicare.
I begin with an analysis of the change in ACA opinion during open enrollment and in August 2014. I then discuss how changes in ACA attitudes as a result of implementation may have affected support for Democratic candidates in the 2014 House midterm elections. I display all results as recycled predictions.2
Figure 2 plots the recycled predictions for ACA opinion postenrollment and in August 2014 from the regressions of ACA opinion in these two time periods on eight insurance types, controlling for preenrollment ACA opinion and other characteristics. Figure 2 has panels for ACA favorability, effect on family, and effect on country, in which the vertical lines represent the average recycled predictions for the omitted insurance type: employer-provided insurance. The average recycled prediction for the other insurance types is shown by data points: dark for postenrollment and light for August ACA opinion. A data point lying farther to the right (left) of the vertical line indicates a more (less) positive ACA opinion relative to the omitted type. The results show little difference between the postenrollment and August recycled predictions for this insurance type. As a result, the impact for each of the other insurance types shown in figure 2 also reflects the difference for that type relative to the preenrollment period. The horizontal lines through the markers are the 95% confidence intervals for the recycled predictions, a different concept than the statistical significance of the insurance type, holding other variables constant. However, in work not shown, I found a high degree of similarity between the significance of the recycled prediction and the significance of the coefficients on the insurance-type variables.
Individuals with marketplace insurance, Medicare enrollees, and individuals who selected “other” insurance all had more favorable opinions of the law after enrollment. The change in attitudes for the marketplace and Medicare groups persisted to August, indicating that ACA implementation had a lasting impact on ACA opinion for these groups. Individuals who purchased insurance through the marketplaces were the only group to have significantly improved attitudes on all three measures of ACA opinion. The changes in attitude are substantively large for both the law's effect on family and overall favorability rating, suggesting that the improvement reflects personal gain. Thirty-nine percent of marketplace enrollees did not have insurance in 2013 (see table 2), 83% received a premium subsidy (Levitt, Claxton, and Damico 2014), and the others received the same insurance pooling benefits enjoyed by those who have employer-provided insurance. Reflecting these tangible benefits, the marketplace group on average had a positive predicted rating of the law with respect to all three questions.
Individuals enrolled in Medicare also became significantly more favorable toward the law and optimistic about its effect on the country postenrollment. The attitude shift was small, perhaps a reflection that ACA implementation had a minor impact on this group during the 2014 enrollment period relative to those who were able to purchase affordable insurance on the marketplaces. Notably, Medicare enrollees were not more likely to say that the law made their family better off, suggesting that their change in attitudes may not result from personal benefits but sociotropic considerations.
Individuals who selected “other” as their insurance type were the only other group to have significantly more positive opinions of the law after enrollment. However, the change in opinion, while still more favorable, is no longer significant in August. Analysis of individuals who had initially selected other insurance and switched coverage in a follow-up survey shows that many of them were enrolled in Medicare.
Two groups had significantly less favorable opinions of the ACA after enrollment: individuals with military or veteran coverage and those unable obtain insurance for 2014. The effect for military/veteran coverage is substantively small and disappears by August. In comparison, the uninsured experienced a large drop in favorability of the law, which persisted until August. Despite reporting a median income of $25,000–30,000, the uninsured identified as Democrat at rates similar to the insured, whose incomes were twice as large. They were more conservative than expected because a disproportionate number resided in nonexpansion states, that is, states that chose not to accept Medicaid expansion funds.
These results largely align with the expectation that only groups that benefited from ACA implementation during the open enrollment period would have more positive attitudes of the law upon receiving these benefits. Both the marketplace and Medicare groups had more positive opinions of the law and their increased support of the law persisted until August. The size of the shift is in proportion to the benefit received. In contrast, only the uninsured had sizable and durable declines in support for the law. Possible explanations include an unrealized promise of coverage or the fear of having to pay a tax penalty in the next tax period. Twenty states refused federal funds to expand Medicaid coverage in the first year of implementation, leaving many individuals without insurance. Individuals who fell into the so-called Medicaid gap may have lost faith in the ACA's promise of insurance for all after they did not gain access to affordable insurance during the first enrollment period. Seventy-eight percent of individuals in the panel who were uninsured in 2014 were also uninsured in 2013 (see table 2).
Notably, the opinions of Medicaid enrollees did not change as a result of ACA implementation. While expansion benefited all individuals on Medicaid by reducing the likelihood of being uninsured in the future, clearly the largest benefit accrued to the 8.4 million adults who qualified as a result of expansion. To the extent that beneficiaries were aware of the ACA's role in expanding Medicaid, we should expect individuals who gained insurance through expansion to have more favorable postenrollment opinions of the law. Similarly, individuals who were previously uninsured and who were able to afford insurance on the marketplaces, often with the support of federal subsidies, arguably benefited the most from the creation of the marketplaces. However, it is important to reemphasize that the analyses estimate the unanticipated benefits (or costs) arising from implementation. To the extent that individuals anticipated benefits, their opinions of the law would remain unchanged after enrollment.
I ran regressions with interactions between having no insurance in 2013 and three insurance types: individuals without insurance in 2014, Medicaid enrollees, and individuals with marketplace insurance. Figure 3 plots the recycled predictions for the key interactions. The sample excludes Medicare enrollees since this group was certain to be insured in 2014 (or within the near future if under age 65).3 I observed previous insurance status for only a subset of the sample. As a result, the numbers are too small to draw definitive conclusions from the interaction terms for the marketplace and Medicaid groups. Only 26 of the 249 Medicaid enrollees in the postenrollment sample did not have insurance previously. This number dropped to 19 out of 201 individuals in the August sample. Nevertheless, the combined effect of the main effects for Medicaid and uninsured in 2013 and their interaction is significant at the 90% level and indicates that previously uninsured Medicaid enrollees became more favorable toward the law.
For individuals with marketplace insurance, previous insurance status does not appear to have affected their opinion of the law. Again, the results are only suggestive due to small sample sizes. With respect to ACA favorability and the ACA's effect on family, individuals with marketplace insurance were more positive toward the law regardless of previous insurance status. The difference is significant in all but one case in the postenrollment period and in every case in the August period. On the question of the ACA's effect on country, only individuals who were previously insured changed their opinion. These individuals may have been the most skeptical of the law heading into the enrollment period. Their options for insurance coverage under the law may have exceeded their expectations.
In terms of expectations, individuals who were not able to obtain insurance in 2014 despite the ACA's promise of expanding insurance may have been the most disappointed after the open enrollment period. The results confirm that individuals who were uninsured for both 2013 and 2014 were significantly less favorable toward the law. Those who lost their insurance coverage in 2014 also were less favorable toward the law, but the effect is not significant. Sixty-two out of 403 individuals reported having lost their insurance in the postenrollment sample. The number dropped to 50 out of 194 in the August sample. In this case, variance in the response and not sample size may explain the lack of significance. Many in this group were left uninsured in 2014 as a result of state decisions not to accept additional federal funds to expand Medicaid.
I considered whether the change in ACA opinion among the uninsured varied depending on residence or, more specifically, on whether a respondent lived in a Medicaid nonexpansion state. Figure 4 plots the results of the recycled predictions from regressions of insurance type on the three measures of ACA opinion, with the inclusion of interactions with residence in a state that chose not to accept federal Medicaid expansion funds and three insurance types: individuals without insurance in 2014, Medicaid enrollees, and individuals with marketplace insurance; the sample includes all individuals. States that chose not to accept Medicaid expansion funds were also more likely to have less generous Medicaid programs prior to ACA implementation and to not take an active role in creating and promoting the health insurance marketplaces. Most relied on the federally operated marketplace. The results suggest that opinion among marketplace enrollees may have improved most in the nonexpansion states after enrollment on the measures of ACA favorability and the ACA's effect on family. By August, however, predicted opinion is consistently higher for individuals with marketplace insurance in expansion states on all three opinion questions. Though significantly more favorable toward the law, individuals enrolled in the marketplace in Medicaid nonexpansion states became less optimistic about the law's effect on their family and on the country by August.
This reversion suggests that initial optimism may have diminished among marketplace participants in Medicaid nonexpansion states, perhaps in response to issues with coverage or as a result of the anti-ACA politics exposed by state elites. Across all insurance types in nonexpansion states, there was no effect of ACA implementation on ACA opinion. These results provide evidence that the panel design effectively controls for preenrollment confounders and that the decline in ACA attitudes among marketplace enrollees in nonexpansion states reflects personal experience with implementation.
Among the uninsured, the results reveal that the main effect is no longer significant for all three measures of ACA opinion, and all of the decline in ACA opinion is explained by the interaction with residence in a nonexpansion state. These individuals were significantly more likely to have unfavorable opinions of the ACA and believe it would make both their family and the country worse off. These results confirm those of Jacobs and Mettler (2016), who found that ACA opinion did not vary significantly between residents of expansion and nonexpansion states. Rather, it appears that only individuals who were left without insurance in these states had much more negative opinions of the law after the open enrollment period. These results support the conclusion that the uninsured were less supportive of the law as a result of unmet expectations of ACA implementation. Figure 4 also shows predicted postenrollment and August ACA opinion for all individuals in nonexpansion states.
Across these analyses of postenrollment opinion, I found consistently more positive ACA opinions among individuals who enrolled in marketplace insurance. In addition to being more favorable toward the law, they were more likely to say that the law makes their family better off. In contrast, I found that individuals who were unable to obtain insurance in 2014 were less favorable toward the law. This effect appears to be concentrated among uninsured individuals residing in states that chose not to expand Medicaid. Based on their incomes, they are likely to have qualified for Medicaid if the eligibility threshold had been raised in their state. In the next section, I consider how changes in policy attitudes may have affected political opinion and behavior.
House Midterm Elections, 2014
For policy to affect political outcomes, policy benefits must affect political behavior, by turning out new groups of voters, impacting political debate, or influencing candidate support. The panel design does not allow me to evaluate the impact of policy on voter turnout. While I observed turnout in the 2012 presidential election, this variable does not effectively control for likelihood of turnout in the subsequent midterm election. Specifically, older voters were more likely to turn out and are more likely to have benefited from Medicare reforms implemented under the ACA during the first period of open enrollment. It is unlikely, however, that this group's higher turnout in the midterm election was a result of ACA implementation rather than simply a result of their higher propensity to vote. I am able to assess the impact of ACA implementation on party support in the midterm election, although these results are only suggestive due to significant panel attrition between the health and election surveys over the course of the entire year.
First, I reran the basic regression of insurance type on ACA opinion, including interactions with Democratic Party identification for all insurance types. These results, shown in figure 5, are suggestive: the sample size for certain subgroups is too small to make a reliable inference, particularly among individuals with marketplace insurance. There are only 26 Republicans and Independents in the postenrollment marketplace sample and 22 in the August sample. Keeping this in mind, there is evidence to suggest that Republicans and Independents with marketplace insurance had the largest improvement in ACA opinion in the postenrollment period. By August, these gains had diminished, though the effects were still significant for ACA favorability and the ACA's effect on family. Among the uninsured, Democrats were significantly less likely to be favorable toward the law after implementation. This decline persisted into August.
Because the results discussed so far reveal no significant change in ACA opinion for most insurance groups, ACA implementation during the first enrollment period likely had little effect on midterm election choices. Groups not directly impacted by the law had the same opinion of the ACA throughout the first enrollment period and heading into the November midterm election. Partisanship more than any other factor appears to have shaped their position on the ACA, so implementation should not have directly impacted their voting behavior.
There are two groups, however, where implementation did have a sizable effect on policy preferences: individuals who were able to buy insurance on the marketplaces, often with the support of federal subsidies, and individuals left without insurance after the end of the open enrollment period. The question is whether their changing opinion of the law had any influence over how they voted in the election. The ACA featured prominently in the debates surrounding the 2014 midterm election, and many observers described the election as a referendum on Obama and Obamacare. Those who benefited from the law by being able to purchase insurance on the marketplaces had a reason to support Democratic candidates in the House and Senate elections. Republican candidates campaigned on a promise to dismantle the law in the next Congress. It is clear that, for most voters, preserving the ACA was not a priority. Republicans made sizable gains, claiming the majority in the Senate, increasing their majority in the House, and achieving a net gain of two states in gubernatorial elections. Significantly, incumbent Republicans in states that refused to accept federal Medicaid funds retained their seats.
Figure 6 plots the results of regressions of insurance type on vote for and support for the Democratic candidate in the 2014 House midterm elections. I found that individuals with marketplace insurance increased their support for Democratic House candidates, but their more positive assessment did not translate into an increased likelihood to vote for the Democratic candidate in the election. ACA implementation had no significant effect on political attitudes and behavior for the uninsured.
The key assumption of the panel design is that, conditional on baseline ACA opinion and other preenrollment variables, insurance type and postenrollment ACA opinion are independent of other factors that are constant over time. To provide support for the model, I conducted a series of placebo tests to demonstrate that there is no correlation between 2014 health insurance type and Democratic Party identification in the 2012 presidential election conditional on the regression model.
I tested three models: the basic model of the regression with insurance type, the model including interactions with 2013 insurance status, and the model including interactions with residence in a Medicaid nonexpansion state. None of the effects are statistically significant or substantively large for the key variables considered in the analyses. The placebo test results, displayed in figure 7, provide further support that I have captured the effect of ACA benefits received through insurance and not political trends.
This study provides rare evidence that, under certain circumstances, policies can affect the policy preferences of impacted groups, moving them away from their partisan views. The ALP's ACA survey series provided the individual-level panel data required to estimate the causal effect of policy benefit receipt on support for the policy. I found that opinions of the ACA among individuals who enrolled in insurance plans on the health insurance marketplaces improved in the few months between the start and close of open enrollment among both Democrats and Republicans. To a lesser extent, individuals on Medicare also reported improved attitudes of the ACA. These changes in opinion indicate that, for these groups, their experiences during the first period of open enrollment exceeded their expectations heading into this period.
Individuals who enrolled in Medicaid reported no significant change in their opinion over the open enrollment period. Among Medicaid enrollees who were previously uninsured, the group most likely to have benefited from Medicaid expansion, there is some evidence that attitudes of the ACA improved over the open enrollment period and into August. However, these changes in ACA opinion are significant only at the 90% level, possibly due to the small number of these individuals observed in the panel data.
In contrast, opinions of the ACA among individuals who did not obtain insurance during this period became significantly more negative. The decline in opinion, however, is observed only for uninsured individuals in states that did not accept federal funds to expand Medicaid. The incomes of these uninsured individuals place them as likely beneficiaries of Medicaid expansion. Individuals in nonexpansion states therefore appear to attach some responsibility to the ACA for not delivering on its promise of affordable health insurance, even though the responsibility lies with state leaders who rejected federal expansion funds. An expectation of having to pay a tax penalty for not having obtained insurance may also explain decreasing support for the law among those who remained uninsured, even though most had incomes low enough to exempt them from the small penalty.
These changes in preferences persisted at least through August 2014, at which time individuals would have had an opportunity to use their insurance. Despite the fact that improvements in ACA favorability persisted, I found no evidence that changes in opinions of the law resulting from implementation impacted the likelihood of voting for a Democrat candidate in the 2014 House midterm elections. Individuals who obtained insurance on the health insurance marketplaces did report a significantly increased percent chance of supporting the Democratic candidate, though the change does not appear to have affected their ultimate vote choice.
The absence of an election effect on party support may reflect the multidimensional nature of a voter's political calculus, with health policy being only one component. Because voters have an enduring allegiance to their party, it would be remarkable for individuals to shift partisan allegiances as a result of a single policy reform. Moreover, the loss of a significant proportion of panelists across different surveys over the course of the year may bias the election results. Further data are needed to provide a definitive answer to whether ACA implementation impacted the 2014 midterm election.
Regardless, voters do not have to vote for a different party in order for a policy to affect politics. It is possible that ACA policy increased turnout among policy beneficiaries, which could have had a large impact on the midterm election, when overall voter turnout was low. In addition, candidates and parties may be forced to change their policy platforms to avoid voter retaliation. The results for marketplace insurance suggest that individuals who benefit from the ACA favor the law and will likely oppose any efforts to repeal it without a plan to replace lost benefits with equally good alternatives. When Medicaid was first enacted in 1965, many states chose not to accept federal funds to establish state Medicaid programs (Engelhard and Olson 2010). Facing increasing pressure to expand access to insurance coverage, all states eventually accepted the federal funds and had Medicaid programs in place by 1982 (Tallon, Rowland, and Lyons 2015).
Since the first year of ACA implementation, the trend has been for states to reverse their initial decision not to accept Medicaid expansion funds. Traditional lobbying groups such as hospitals and doctors, who stand to benefit from the additional funds, have put pressure on holdout states to expand Medicaid and opposed the initial repeal-and-replace legislation. As more and more individuals rely on the ACA (or a largely similar program) for their health care, it is possible that the ACA will become another “untouchable” policy, joining Medicare on the third rail of politics. Democrats, campaigning on the issue of health care, regained control of the House of Representatives in the 2018 midterms (Hall and Tolbert 2018). While aspects of the ACA may well change in the future, the question is whether some form of government-organized and subsidized health insurance market will become a widely accepted method for filling the gaps in the US employer-based insurance system.
I gratefully acknowledge Katherine Carman, Eric Schickler, Jasjeet Sekhon, Erin Hartman, and Francesca Refsum Jensenius for their suggestions and guidance in the development of this article, as well as the invaluable feedback from anonymous reviewers, seminar panelists, and my colleagues at the University of California, Davis.
To test whether the results are invariant to this coding scheme, I reran all the analyses presented in this article after replacing pre- and postenrollment ACA opinion variables with an indicator for positive support of the law. In addition, I included controls for whether an individual responded “don't know” in the preenrollment period to each of the three ACA measures. The results of the sensitivity analyses are consistent with those presented here.
The coefficient estimates and their standard errors, as well as information about subgroup sizes, are given in the appendix.
The Medicaid main and interaction effect estimates are robust to the inclusion/exclusion of Medicare enrollees.