Context: The authors examined whether participation in the Affordable Care Act (ACA) differed by political partisanship. Answering this question is important for understanding how contentious elite-level decision making and discourse may affect policy uptake, and the ability of the ACA to create a constituency of beneficiaries invested in its support.
Methods: The authors combined county- and individual-level data on health insurance gains, political partisanship, and variation in the ACA's implementation to test whether Democratic areas and individuals are more likely to gain health insurance after implementation of the ACA.
Findings: While there was some evidence that Democrats experienced larger insurance gains than Republicans, the major partisan divide was in how insurance was obtained: Democrats were more likely than Republicans to enroll in private marketplace plans, but there were no partisan differences in uptake among those gaining insurance via the Medicaid expansions.
Conclusions: These results provide some important context for the well-known and persistent partisan divide in support for the ACA, an attitudinal divide whose reflection in program participation appears conditional on the salience of different aspects of the ACA.
Public policies often have important effects beyond the proximate problem they are intended to address. By providing program participants with new benefits, these benefits and their distribution may also affect public opinion and political participation. The extent to which public policies affect politically relevant behaviors—whether “new policies can create new politics” (Schattschneider 1935)—is important for judging a program's prospects for creating a constituency of beneficiaries invested in the policy's long-term survival (Schattschneider 1935; Pierson 1993; Soss 1999; Mettler 2005; Soss and Schram 2007; Patashnik 2008; Weaver and Lerman 2010; Campbell 2003, 2012). For a public policy to create a “new politics,” not only must the policy have a perceptible impact on beneficiaries, but this impact must also extend beyond the set of initial supporters. If the benefits of a policy accrue primarily to those who already support the policy, and if opponents of the policy either cannot or will not participate, existing disagreements about the policy will likely be exacerbated.
We explored partisan differences in program participation in the most consequential and controversial social policy enacted in recent years: the Patient Protection and Affordable Care Act of 2010 (ACA). The signature policy of Democratic president Barack Obama, the ACA had wide-ranging impacts on nearly every aspect of the health care infrastructure in the United States. Most important, about 20 million Americans gained health insurance as a direct result of the ACA (Sommers and Epstein 2017). Rather than examining the relationship between policy impact and political changes, as several others have done (Haselswerdt 2017; McCabe 2016; Clinton and Sances 2018; Hopkins and Hobbs 2017; Sances and Clinton 2018), we focused on the first-order question of whether program participation varied along partisan lines. In particular, we asked if Democrats, Republicans, or independents were more or less likely to gain health insurance as a result of the law.
Many have documented the ways Republican-led states resisted the creation of health care exchanges and the expansion of Medicaid under the ACA (e.g., Beland, Rocco, and Wadden 2016; Hertel-Fernandez, Skocpol, and Lynch 2016; Jones 2017; Michener 2018; Patashnik and Oberlander 2018). What is less clear, however, is whether these efforts had downstream consequences on the relative likelihood of Democrats and Republicans actually participating in the ACA. Did the resistance of political elites to Medicaid expansion affect health insurance uptake along partisan lines? Insofar as the policy effects were influenced by the level of political resistance in a state, we might expect Democratic and Republican individuals to participate at different rates.
Examining the partisan distribution of ACA participation is important for at least three additional reasons. From the point of view of a policy designer who also cares about influencing future politics, it is important to create a bipartisan coalition of recipients to help constrain elected officials—in this case, Republicans—seeking to repeal the policy. Moreover, the stability and partisan divide of ACA opinions are well known, with Republicans and Democrats differing in their ACA approval ratings by nearly 60 percentage points. Establishing whether the partisan difference in support for the ACA is related to partisan differences in uptake helps assess the relative importance of program participation for public opinion. Lastly, recent work points to unequal uptake of private marketplace insurance, a result that suggests program participation itself is a form of political behavior (Lerman, Sadin, and Trachtman 2017). Using a wider-ranging analysis, we helped assess whether differential partisan uptake might extend to the ACA broadly construed, as well as the ACA-induced expansion of Medicaid.1
We used what we believe is the most comprehensive data on health insurance gains by voter partisanship. We began by analyzing the geographic distribution of health insurance coverage before and after the ACA's implementation at the county level. Combining administrative data on health insurance gains with election results in all 50 states revealed a positive relationship between gains in health insurance and Democratic vote share. However, the relationship is imprecisely estimated, and it is driven entirely by counties in states where Medicaid was not expanded. In states that expanded Medicaid, the raw relationship between insurance gains and Democratic vote share is flat or even negative. Probing the relationship further using a geographic discontinuity design revealed roughly equal effects of Medicaid expansion in both more Republican and more Democratic areas, again suggesting that any partisan differences in health insurance gains are driven by nonexpansion states.
To move beyond ecological relationships, we also analyzed individual-level data using the Cooperative Congressional Election Study (CCES), a panel survey conducted between 2012 and 2014. Our within-individual design revealed that political independents experienced the largest gains in insurance, followed by Democrats. However, Medicaid expansion was critically important for program participation—in states choosing to expand Medicaid independents and Republicans experienced the biggest gains in insurance coverage, but in nonexpansion states Democrats and independents gained most. While these results are suggestive, we were unable to reject the null hypothesis that these differences could have arisen due to chance, in part because of the limited sample size.
To increase our sample, we shifted from a within-individual panel analysis to a between-individual time-series analysis using the Kaiser Family Foundation (KFF)'s monthly tracking polls. This alternative design confirmed the patterns suggested by the geographic and panel data sets: in nonexpansion states Democrats were more likely to become insured than Republicans, but there was essentially no difference in expansion states. Decomposing these gains by insurance type revealed that Democrats were far more likely to enroll in individual market plans, especially in nonexpansion states, but that Democrats and Republicans were equally likely to enroll in Medicaid plans in expansion states.
Using three different research designs and three different data sources gave a consistent portrait: Democrats appeared slightly more likely than Republicans to participate in the overall ACA, but there is interesting and important variation under the surface. In states that opted not to expand Medicaid, Democrats were more likely to participate than Republicans—a result consistent with Lerman, Sadin, and Trachtman (2017) and perhaps reflecting partisans following cues from partisan elites, refusing to participate in a program they strenuously resisted. When we examined states that chose to expand Medicaid, however, we found no partisan differences in uptake. Overall, our results provide some important context for the well-known and persistent partisan divide in support for the ACA, an attitudinal divide whose reflection in program participation appears conditional on the salience of different aspects of the ACA.
The Policy Effects of the ACA
A substantial literature examines how the ACA has affected both the health care infrastructure of the United States and the well-being of its citizens in terms of health and finances (see, e.g., the extensive summary in Antonisse et al. 2017). In addition to examining gains in insurance coverage, scholars have examined how the ACA has affected employer-sponsored health plans (Abraham, Royalty, and Drake 2016), the creditworthiness of low-income Americans (Allen et al. 2017), the financial well-being of low-income families (Miller, Kaestner, and Mazumder 2017; Nikpay, Levy, and Buchmueller 2017), labor market outcomes (Shore-Sheppard, Schmidt, and Watson 2017), and even criminal recidivism (Bird and McConville 2017). Others have studied the consequences of these policy effects for political participation (see, e.g., Haselswerdt 2017; Clinton and Sances 2018) and public support for the ACA (Pacheco 2017; Hopkins and Parish, forthcoming; Hopkins and Hobbs 2017; Lerman and McCabe 2017).
We pivoted to examine the partisan distribution of participation in the ACA itself. To be clear, almost all Americans “participate” in the ACA, as it is an extremely complicated policy that affects nearly every aspect of the health care system in the United States. For our purposes, however, we examined participation in terms of what was arguably the ACA's most salient aspect: the attempt to increase health insurance via regulation of the individual insurance market and the expansion of Medicaid. While participation in other aspects of the ACA (e.g., taxes paid to support the ACA, compliance with regulations regarding insurance coverage) may also vary by partisanship in important ways, we focused on whether the first-order priority for the ACA—increasing the percentage of individuals with health insurance—demonstrated partisan-related differences in participation.
The question of “who participated” can be examined from many angles (Chattopadhyay 2018), but understanding whether policy uptake is equal across partisan groups is important for assessing whether the ACA has the potential to build a bipartisan coalition of beneficiaries (Patashnik 2008; Patashnik and Zelizer 2013). If uptake is concentrated in one party, a policy may become threatened when those opposed gain power. For example, was the inability of the unified Republican Party to repeal and replace the ACA following the election of President Trump in 2017 a failure at the elite level, or was it a failure driven by the reluctance of Republican voters who came to realize how they benefited personally from the ACA (Sommers and Epstein 2017; Hacker and Pierson 2018)?2 Our investigation could not unpack the precise reasons for the failure of the Republican-led repeal efforts, but we could identify whether ACA participation varies in ways that could exacerbate or ameliorate partisan differences in opinions.
Expectations about which partisans participated more in the ACA were not entirely clear. On one hand, we might expect Democratic-leaning citizens to engage with the ACA more because many of the gains in health insurance coverage occurred in the (typically Democrat-led) states that chose to expand Medicaid. If Democrat-led states contained more Democrats than Republicans, we might expect these compositional differences to cause differences in the percentages of Democrats and Republicans participating in the ACA. Democrats may also participate more because Democrats are, on average, worse off financially than Republicans (Leighley and Nagler 2014). Because the ACA created subsidies for the purchase of health insurance for those making up to 400% of the federal poverty limit, then to the extent that there are more Democrats than Republicans below this cutoff, Democrats will be more likely to take advantage of the ACA's tax credits. A third potential reason for greater participation among Democrats was suggested by Lerman, Sadin, and Trachtman (2017): Republicans may be less likely to enroll in the ACA's marketplace coverage due to their opposition to government intervention—perhaps fueled by elite-level resistance.3 This resistance may also have taken the form of televised ads opposing the ACA, which Gollust et al. (2018) showed led to lower marketplace participation.
Yet it is also possible that Republicans participated as much as or even more than Democrats. To the extent that the social welfare state, the health care infrastructure, or the health and well-being of citizens in Republican-led states are systematically worse than those in Democratic-led states, it is possible that the ACA had more impact for residents of the former. As an illustration of the dramatic differences in health care support between states, consider the status of two nonexpansion states as of January 2018: deeply red Alabama, where the Medicaid eligibility limit for parents was 18% of the federal poverty level ($20,780 for a family of three), and purple Maine, where the limit was 105%. Residents of states like Alabama, with few social services, may be more affected by the ACA's regulations and subsidies than residents of states with a more generous social safety net. If so, states with less expansive Medicaid programs may see larger increases in insurance coverage because of the ACA, even if they choose not to expand Medicaid.
To examine whether there are partisan-related differences in ACA participation, we conducted several analyses at varying levels of aggregation, and using multiple research designs, to assess how the percentage of insured individuals changed after 2014. In particular, we used administrative, county-level data on health insurance coverage and election results, and also self-reported, individual-level data on coverage and partisan affiliation. Neither approach is perfect—the county-level analysis relied on ecological inferences, and the individual-level data relied on the accuracy of self-reports of insurance coverage—but to the extent that we obtained consistent effects from each, we can be more confident in our conclusions.4
We began with a cross-sectional analysis of the relationship between county-level gains in insurance and county-level Democratic vote share. We measured changes in insurance using the Census's Small Area Health Insurance Estimates. To construct our outcome variable, we computed the share of those between ages 18 and 64 with health insurance in each county in 2013, the same share in 2015, and then took the difference. To measure Democratic vote share, we averaged the Democratic share of the two-party vote received in all statewide races (president, governor, and senator) in 2004, 2006, 2008, and 2010 for each county, using election returns obtained from the Congressional Quarterly's elections database.
Figure 1 plots the gains in health insurance against Democratic vote share for all counties and then separately for counties in states that did not and did participate in Medicaid expansion. We weighted observations by population and used a (weighted) regression line to denote the linear relationship between the two variables for each group of counties. The left panel reveals a positive relationship between Democratic vote and gains in insurance coverage, suggesting that more Democratic-leaning areas were more likely to acquire health insurance under the ACA. Separating out the counties by expansion status, however, reveals that the positive relationship between vote share and health insurance gains was driven entirely by counties in nonexpansion states. In expansion states, the relationship is actually negative, suggesting that residents of Republican-leaning counties were more likely to gain insurance due to the ACA.
To formally estimate these relationships, we regressed the results of the change in share insured between 2013 and 2015 on Democratic vote share (weighted by population; table 1). For interpretation, we rescaled the Democratic vote such that zero represents the sample minimum (8.5 points) and one represents the sample maximum (86 points), so the resulting coefficient summarizes the impact of going from the least to most Democratic county. Because counties are not independent within states, and because many important aspects of the ACA vary by state, we clustered standard errors at the state level.
The results in table 1 using all counties regardless of expansion status and other state-level factors show that, compared to the least Democratic leaning county, which saw a gain of 5.93 points in insurance coverage (the regression constant), the most Democratic county saw a 2.49-point gain in insurance coverage. However, we were unable to reject the null hypothesis of no difference at conventional levels, given an estimated standard error of 1.61 points. We repeated the analysis separately by expansion status. In agreement with figure 1, table 1 shows a positive relationship in nonexpansion states (estimate of 3.23, standard error of 1.70) and a negative relationship in expansion states (estimate of −0.98, standard error of 2.21). However, none of these estimates—and thus the relationships shown in figure 1—statistically differs from zero.
We then repeated the analysis with state fixed effects to account for the many other state-level factors that may jointly influence county voting behavior and county gains in health insurance.5 Controlling for stable between-state differences using state fixed effects increased the precision of our estimates: the positive slopes for all states and for nonexpansion states are statistically significant at conventional levels. In expansion states, however, the relationship between Democratic vote share and gains in insurance coverage is substantively and statistically insignificant: the estimate is positive (1.00), but with a large standard error (1.20).
To further remove possible confounding factors and identify the effects of the Medicaid expansions in particular, we used a regression discontinuity design (Clinton and Sances 2018) to examine the gains in insurance coverage occurring in counties in expansion states that are just over the border from nonexpansion states (for geographic discontinuity designs more generally, see Keele and Titiunik 2015).6 To do so, we focused on the roughly 2,000 counties in states that share a border with a county (or counties) in a state with a contrary expansion status (e.g., Tennessee-Kentucky, Virginia-West Virginia). To visually demonstrate this strategy, we plotted county-level changes in insurance against the distance between the geographic center of the county and the closest border of an opposite-expansion state (figure 2). We show this relationship separately for Republican-leaning counties (those below the median Democratic vote share of 42 points) and Democratic-leaning counties (those above the median).
Consistent with the well-known impact of the Medicaid expansions on insurance coverage, there is a positive jump of about 5 points at the threshold. Compared to counties just inside nonexpansion states, counties just inside expansion states saw a 5-point increase in insurance coverage. Visually, the jump for Republican-leaning counties in the left panel of figure 2 actually appears somewhat larger than that for Democratic-leaning counties in the right panel.
Table 2 reports the results of regressing the county-level change in insurance on expansion status (being just across the border of an expansion state), distance from an expansion state border, and their interaction. Estimates using all counties indicate counties along the border of expansion states saw an average gain in insurance coverage of 2.82 points compared to counties along the border of nonexpansion states. The effect is precisely estimated, with a standard error of 0.93 points. We repeated the analysis by estimating the effect separately for Republican- or Democratic-leaning counties. Republican-leaning counties actually saw a slightly larger gain in insurance, according to these estimates: 3.97 points (error of 1.77 points) versus 2.54 points (0.89 points) in Democratic-leaning counties, but we were unable to reject the null hypothesis that these two estimates could be equal.7 Our inability to detect a differential effect in Democratic- or Republican-leaning counties is not due to our decision to focus on whether a county is above or below the median Democratic vote share: table 2 shows that the effect of the discontinuity estimated separately by quintile of Democratic vote is generally larger in more Republican-leaning counties, but we cannot reject the null hypothesis that the effects are equal across quintiles.8
While our county-level analysis provides suggestive evidence, it relies on ecological inferences and cannot identify which types of individuals participated in the ACA. Even if Republican-leaning areas may have seen insurance coverage increase as much as Democratic-leaning areas, it could still be the case that Democratic voters in both areas participated more than Republican voters. To account for this possibility, we turned to individual-level panel data from the 2010-2012-2014 CCES, analyzing responses to the question, “Do you currently have health insurance?” Following McCabe (2016), we coded respondents as gaining insurance if they did not have insurance in 2012 but did have insurance in 2014, and we used survey weights to correct for the slight nonrepresentativeness of the sample.9 Overall, 7% of the 4,343 CCES respondents reported gaining insurance. To remove those receiving insurance through Medicare, we dropped all CCES respondents over age 65 as of 2014; this reduced our sample size to 2,767 and increased the baseline increase in insurance coverage to roughly 9%.10
Figure 3 plots the percentage gaining insurance by party identification and by state expansion status.11 The results for all states in figure 3 (thick black line) show that independents saw the biggest increases in insurance coverage (roughly 10 percentage points), followed by Democrats and Republicans (both roughly 8 percentage points). Moreover, the effect for Republicans varies considerably depending on whether or not they lived in a state that expanded Medicaid. In nonexpansion states (dashed line), Republicans saw considerably smaller gains in health insurance coverage compared to any other group (5 percentage points). In contrast, in expansion states (dotted line), Republicans actually saw higher gains in insurance coverage (around 10 percentage points) than Democrats.
To examine these relationships in more detail, table 3 reports the results of regressions corresponding to these patterns. We again clustered standard errors at the state level, and the baseline category in these regressions is Democrat. Interpreting the constant terms in table 3 reveals that Democrats saw insurance gains of about 8 percentage points in all states, in nonexpansion states, and in expansion states. The coefficients on “Independent” and “Republican” reveal that independent voters saw larger insurance gains than Democrats regardless of expansion status, but Republicans saw smaller gains than Democrats in nonexpansion states and larger gains in expansion states. However, these group differences are again statistically indistinguishable from zero, as is the difference in differences—how much Republicans gained or lost relative to Democrats in expansion states versus nonexpansion states.12
The imprecision of these estimates is unsurprising given that just 190 of our roughly 2,700 CCES respondents report gaining insurance, but these estimates are qualitatively and reassuringly similar to the patterns we identify in our cross-sectional, between-county analysis. We again found little to no evidence of partisan differences in overall insurance gains under the ACA but some suggestive variation by Medicaid expansion status. Democrats' insurance gains were larger than Republicans' in nonexpansion states, but there were no partisan-related differences in expansion states.13
The CCES data allowed us to examine the ACA's impact using a within-individual panel design that avoids the issue of ecological inference, but the relatively small number of respondents gaining health insurance, as well as the fact that we had only two time periods, limited our ability to precisely estimate differences. To address these issues, we next used individual-level data from the KFF tracking poll to examine the between-individual patterns over time. The KFF polls have been administered to nationally representative samples of US adults on a regular basis since 2010, and although they are not panel data—meaning that we could not observe individual-level changes in insurance coverage—they did allow us to determine whether the proportions of Democrats and Republicans with health insurance changed differentially over time.
Figure 4 plots the percentage of 62,579 unique respondents under the age of 65 who reported having health insurance as of each of the 66 KFF time points between 2010 and 2016.14 As in our CCES analysis, we adjusted these percentages using survey weights provided by KFF. The thin lines represent the raw weighted percentages in each survey, and the thick lines represent moving averages computed using local polynomial smoothers. To capture the sharp change in insurance coverage attributable to the ACA's implementation beginning in 2014, we fitted the moving average separately for pre- and post-2014, as in a regression discontinuity design. The vertical dashed lines mark the start of 2014 and the ACA's implementation.
Figure 4a graphs the percentage of insured individuals by party across all states and shows gains in insurance coverage for all groups after implementation of the ACA. However, Democrats appear to have experienced the largest increase (around 5 percentage points), followed by Republicans and independents (who are about 1–2 percentage points lower). This overall pattern obscures important variation, however. Figure 4b graphs the relationship in states that chose to not expand Medicaid. Strikingly, yet consistent with the preceding analyses, gains in insurance coverage are found only among Democrats in nonexpansion states. When we examined expansion states in figure 4c, however, we saw roughly equal gains for all subgroups, based on the magnitude of the jumps in moving averages at the 2014 mark.
The patterns of figure 4 suggest that the failure to expand Medicaid had policy consequences that were largely felt by Republicans and independents in nonexpansion states. Democrats appear to experience an increase in health insurance coverage regardless of whether their state expands Medicaid, but the uptake of Republicans and independents appears contingent on their state's expansion decision. (Of course, it is possible that the gains for Democrats in nonexpansion states would be even larger had expansion occurred there, but this is impossible to determine.)
To test whether these differences are statistically significant, we regressed an indicator for having insurance (multiplied by 100 to ease interpretation) on an indicator for being surveyed in 2014 or later, linear date, and their interaction (table 4). Thus, the “Post” coefficient corresponds to the jump in the percentage insured between the end of 2013 and the beginning of 2014. We clustered standard errors by survey date, although the results were substantively identical when we clustered errors by state. As before, we estimated this regression by partisan subgroup and by state expansion status.
The analysis of all states (table 4, top) shows that, among all respondents, insurance coverage increased by 4.2 percentage points between late 2013 and early 2014. Gains were slightly higher among Democrats (5.94 points), slightly lower among Republicans (3.72), and lowest among independents (3.07). These increases are all statistically distinguishable from zero, meaning that all groups experienced a detectable increase, but the difference in the differences—that is, the degree to which Democrats gained more than the other groups—is not distinguishable from zero. A combined regression with triple interactions failed to reject the null hypothesis of no difference in differences, meaning that we cannot conclude that one group gained more than any other despite the differences in point estimates reported in table 4.15
In examining the relationship by expansion status, the results in table 4 show the gains in insurance were smaller for all groups in nonexpansion states compared to the full sample, and we can distinguish the effects from zero only when using all respondents (gain of 3.19 points) or Democrats (4.57 points). The estimated increases in insurance coverage for independents and Republicans in nonexpansion states is positive (2.46 and 1.63 points, respectively), but we cannot statistically distinguish these from zero. And again, the difference in insurance gains between Democrats and the other groups is indistinguishable from zero.16
Gains were generally higher in expansion states than in nonexpansion states, and we can now reject the null of no significant gain for all respondents (gain of 4.95), Democrats (gain of 7.04), independents (3.37), and also Republicans (5.74). In expansion states we can also conclude that Democrats gained insurance coverage at a higher rate than did independents, but the gains among Democrats and Republicans are statistically indistinguishable.17
Together, these results further highlight the importance of Medicaid expansion for the increase in insurance coverage. Partisan differences in uptake depend critically on whether or not Medicaid expansion occurs: in expansion states, Democrats and Republicans participate equally, while in nonexpansion states, we can detect a nonzero increase in insurance coverage only among Democrats. That said, we cannot statistically distinguish the magnitude of that increase from the small, and imprecise, increase in insurance coverage among Republicans.
Digging deeper, however, we can also examine the source of the insurance coverage to determine if the increase is due to insurance being provided by the insurance marketplace or Medicaid. To do so we took advantage of the fact that the KFF surveys also asked respondents about the source of their coverage: from an employer (or spouse's employer), the individual market, Medicaid, Medicare, or other sources. Thus, we could decompose the gain in insurance by different subgroups reported in figure 4 into transitions into different insurance sources.
Figure 5 plots the percentage of KFF respondents receiving health insurance from the individual market, Medicaid, or no source over the period we examined. As the pattern makes clear, the drop in uninsured is matched by a corresponding increase in the share of respondents receiving insurance from Medicaid or the individual market, which is exactly what we would expect given the design of the ACA.
To ease presentation, we proceed directly to regression estimates rather than replicating figure 5 by party and state expansion status. We used a similar specification as before, except now our dependent variable is an indicator for receiving insurance from a given source (equal to 100 if a respondent is in that category, and 0 otherwise). Further, we restricted the sample to those receiving insurance from the individual market, Medicaid, or no source. This restriction allows us to focus in on transitions from uninsured status to one of the two primary channels by which participants became insured under the ACA.
Table 5 summarizes the results from a series of regressions on different subsamples. We present only the coefficient on “post” and its standard error, omitting the estimates for distance and the interaction. The data show trends corresponding to those shown in figure 5: among all respondents, there is a drop of 12 points in the uninsured share (standard error of 1.5 points). This overall increase was driven by a 2.66-point increase in the share with individual insurance and a 9.33-point increase in the share with Medicaid. In changes among Democrats regardless of their state's expansion status, the overall 16.6-point gain in coverage is distributed roughly equally across the individual market (estimate of 8.44, standard error of 2.12) and Medicaid (8.17, 2.58). In contrast, among independents and Republicans, we see statistically significant increases only in the share receiving insurance from Medicaid, with substantively small changes in the shares receiving insurance via the individual market.
The analysis for respondents living in nonexpansion states in table 5 shows, given the lack of Medicaid expansion, sharp increases only in the share receiving insurance from the individual market: the estimate is 4.19 with a standard error of 1.76. Breaking this gain out by party shows that the overall gain in individual market insurance is driven entirely by the 11.37-point increase (standard error of 3.00) among Democrats. In contrast, the gain among independents is smaller and insignificant, and it is negative for Republicans. Unexpectedly, we also see a gain in the share receiving Medicaid of 8.08 points among Republicans, and the estimate is significant, with a standard error of 3.25. The difference in individual market uptake between Democrats and Republicans is, not surprisingly, significant at conventional levels.18
The results for the gains among individuals living in states that chose to expand Medicaid reveal statistically significant gains in the percentage covered by Medicaid for all subgroups, and the largest gains among independents, at 15.28 points (standard error of 2.29 points), followed by Democrats at 15.51 points (3.2 points), and finally by Republicans at 13.48 points (3.76). We also see a statistically significant increase in individual insurance in expansion states, but only among Democrats: in these states the increase is 6.69 points with a standard error of 2.99. However, the difference in differences—the Democratic marketplace gain of 6.69 versus the Republican gain of 6.32—is not significant (p = 0.95).
The individual time series data provided by the KFF surveys reveals relationships that are consistent with the patterns uncovered in our county-level and individual panel-level analysis. Overall, Democrats are slightly more likely than Republicans to participate in the ACA, but important differences emerge depending on whether the respondent resides in a state that expanded Medicaid: in nonexpansion states Democrats are more likely to participate in the individual marketplace than Republicans, but in expansion states Democrats and Republicans are equally likely to sign up for expanded Medicaid benefits under the ACA.
Discussion and Conclusion
Using a variety of data sets and research designs, we explored whether Democratic areas and voters were more likely to participate in the ACA than Republican areas and voters. While we generally found Democrats to be more likely to gain insurance, the differences were never large or precisely estimated. However, substantively interesting differences emerged when we disaggregated access to health insurance under the ACA into two of its major components: the expansion of access to private insurance on the individual marketplace, on one hand, and the expansion of the public Medicaid program, on the other. When we examined the gains in insurance made possible by each aspect of the ACA, we found that Democrats do appear more likely than Republicans to sign up for private marketplace plans, but there are no partisan differences in participation in the expansion of Medicaid.
Our results suggest that the politicization of the ACA had important consequences on the gains in insurance coverage. Because Democrats and Republicans were equally likely to sign up for Medicaid where coverage was expanded, the decision by Republican elites to not expand Medicaid affected the ability of both Republicans and Democrats to gain access to insurance coverage. Moreover, the resistance of Republican elites to Medicaid expansion also seemed to decrease the willingness of Republicans to participate in the insurance marketplace, resulting in Republicans being less likely to gain access to health insurance. While it is possible that these differences are entirely a consequence of Republicans and independents in nonexpansion states being less willing to purchase health insurance for other reasons, the choice to not expand Medicaid may have had a disproportionate impact on these types of voters.
Although we found partisan-related differences in how insurance was obtained, any overall differences in insurance gains between Democrats and Republicans are imprecisely estimated and are substantively small, at most about 2 percentage points when we pooled all states. That both Democrats and Republicans participated in and benefited from a highly politicized policy yet continued to diverge sharply in their attitudes toward the law speaks to the importance of both political contexts and policy contexts for generating feedback effects. Republicans were only slightly less likely to gain insurance from the ACA overall, but their gains came primarily via the preexisting, state-run Medicaid program in expansion states. In contrast, Democrats gained insurance via both the insurance marketplace and the Medicaid expansions.
While our analyses focused primarily on identifying whether partisan-related differences in insurance gains emerged after implementation of the ACA, the results suggest several explanations. The fact that Republicans were less likely to purchase insurance via the insurance marketplace if they lived in nonexpansion states than if they lived in expansion states suggests—but certainly does not prove—that the differences in political context may matter for program participation. The fact that fewer Republicans chose to participate in the insurance marketplace when Republican political elites chose to not participate in Medicaid expansion suggests that elite resistance and discourse may have affected individual behavior. In states that chose to expand Medicaid, however, Republicans were as likely to participate in Medicaid as Democrats and independents, despite the widespread opposition of Republican elites to Medicaid expansion in other states—suggesting that if Republicans were taking cues from political elites, those cues did not cross state borders.
Another reason that Republicans may have participated in Medicaid but not the insurance marketplace is that the connection between Medicaid and the ACA was less clear. With the variety of state Medicaid program names, as well as the efforts of some governors to actively obscure the connection between the federal ACA and the Medicaid expansion (Kliff 2016), it is plausible that Republicans ended up participating in a relatively more “submerged” component of the ACA (Mettler 2010). Tallevi (2018), for example, found tremendous variation in awareness of the connection between private- and public-provided health insurance, and this ambiguity may have weakened Republicans' negative policy feedback effects. Thus, while Republicans may have been less likely to participate in the parts of the ACA they knew they disliked (Lerman, Sadin, and Trachtman, 2017), many may have also been unknowing participants in a significant, if less salient, component of the reform.
This research has been partially supported by award 94-16-06 from the Russell Sage Foundation. All opinions and errors are our own. For feedback, we thank participants at the 2018 JHPPL Special Issue on Health and Political Participation Conference at the University of Missouri, including Eric Patashnik, Julie Pacheco, Sarah Gollust, Chris Ojeda, and Chaegyung Jun. We also thank Andrew Senter for research assistance.
The Obama administration attributed 12 million of the 20 million overall newly insured to Medicaid expansion (Sommers and Epstein 2017).
Of course, it is also possible that the attempt to repeal and replace not only made voters aware of what the ACA actually did but also that the proposed reforms mobilized additional groups, given the changes that were being proposed (e.g., the proposal to deliver Medicaid as a block grant).
While individuals were able to sign up for ACA-approved plans offered by private insurers, they did so via health insurance marketplaces that were often created and run by the federal government via HealthCare.gov once most states chose not to create their own. The existence of these health care exchanges was directly linked to the ACA. As Beland, Rocco, and Wadden (2016: 91) note: “Conservative opponents of the Obama administration and health insurance reform in the states attacked this previously consensual policy instrument [health insurance exchanges] to publicly express their dissent and directly undermine the implementation of the 2010 federal legislation. The logic of ‘guilt by association,’ along with conservative, GOP mobilization against state-run exchanges was related to policy visibility and the high public profile of exchanges as policy instruments, especially when compared with the regulatory instruments.” The public and prominent implementation difficulties with the website of the federal exchange HealthCare.gov also helped connect the federal exchanges to the ACA (Brill 2015).
To be clear, we were interested in estimating differences in participation at various levels. Differences in the overall policy impact of the ACA, however, may be slightly different. For instance, even if Republicans and Democrats have an equal probability of gaining insurance in expansion states, Democrats will be affected more by the expansions if there are more Democrats living in expansion states.
Democratic-leaning counties also have more nonwhite residents, higher population densities, and lower median incomes. Appendix table A1 reports the results including county-level demographics to show that the partisan differences we detected were not due to these demographic differences.
A geographic discontinuity design makes three assumptions: (a) health insurance gains in a county cannot depend on the treatment status of other counties—that is, insurance gains are not caused by individuals sorting into treated counties from untreated counties; (2) there are no other changes that affect county-level insurance coverage that covary by county and by expansion status; and (3) expansion and nonexpansion states are on parallel paths with respect to insurance gains absent the treatment such that the change in insurance coverage in counties in nonexpansion states can be used to identify the impact of expansion in expansion states. Clinton and Sances (2018) examined the plausibility of these assumptions and found no evidence of their violation.
Using a single regression with an interaction between expansion status and an indicator for being above or below the median (as well as an interaction between distance and partisanship, and a triple interaction), the p-value on the relevant coefficient is 0.31.
The p-value on the joint test of equal effects across quintiles is 0.08.
One concern with the CCES is that, as an opt-in Internet survey (Schaffner and Ansolabehere 2015), it may underrepresent Medicaid recipients relative to the population. Appendix table A2 examines whether CCES respondents are more or less likely to receive Medicaid than KFF respondents. We found this is not the case and that the predictors of Medicaid coverage also did not vary across samples.
We made the same exclusion in the Kaiser data below.
We treated “leaners” as independents, rather than partisans, to make the analysis consistent with the KFF data, analyzed below, which does not differentiate partisans beyond a three-category classification. Coding leaners as partisans gave similar results in the CCES data.
In a single regression estimated among the full sample, and including interactions between partisan categories and expansion status, the p-value on the interaction between Republican and expansion is 0.404. In a separate analysis (data not shown), we also found no evidence that different groups were more or less likely to lose insurance over this period. Overall, about 3% of the weighted sample reported that they had insurance in 2012 but not in 2014.
While party identification is highly stable, we tested whether changes in insurance led to changes in party between 2012 and 2014. Appendix table A3 shows that they did not.
We obtained the individual polls from the Roper Center database hosted at Cornell University. As in the CCES data, we dropped those over age 65 (roughly 24,000 cases).
This pooled regression includes a post indicator, linear time, indicators for each party, an interaction between post and time, interactions between post and party, interactions between time and party, and interactions between post, party, and time. The p-value on the coefficient corresponding to the difference in gains between Democrats and Republicans is 0.17; it is 0.08 when comparing Democrats to independents.
The relevant p-values are 0.24 (Democrats vs. Republicans) and 0.48 (Democrats vs. independents).
The relevant p-values are 0.441 (Democrats vs. Republicans) and 0.05 (Democrats vs. independents).
In a regression of individual insurance receipt on party, post, date, and all relevant interactions, estimated among the subsample living in nonexpansion states, the coefficient on post × Republican is −11.92 with a standard error of 5.7 (p < 0.05).
Appendix: Replication of Table 1 with County-Level Covariates
Table A1 replicates the county-level regression from table 1, but including county-level covariates: log median family income, the share of the county population that is white, the log of county population, and the log of population density. Compared to table 1, the results are substantively unchanged. In particular, the only specification with a statistically significant relationship between Democratic vote and gains in insurance is in nonexpansion states when including state fixed effects.
Representation of Medicaid Recipients in the CCES
Given the CCES uses opt-in panelists weighted to match population-based samples, one concern is that Medicaid respondents could be underrepresented. Additionally, the Medicaid respondents who do appear in the CCES could differ systematically from Medicaid recipients in other surveys.
To examine this possibility, we pooled the CCES and KFF individual-level surveys and predicted receipt of government health insurance (1 = Medicaid or Medicare, 0 = other source or uninsured) as a function of individual-level demographics and survey instrument (1 = CCES, 0 = KFF). We pooled Medicaid and Medicare recipients because the CCES question does not differentiate between those with either of these benefits. From the KFF surveys, we selected the tracking surveys from November of 2010, 2012, and 2014 to provide a comparison to the CCES panel. We clustered regression results by survey respondent ID, and we included indicators for survey year.
Table A2 shows the results of a bivariate regression of government insurance receipt on being in the CCES (vs. being in the KFF data). The coefficient suggests that CCES respondents are just 1 percentage point less likely to receive Medicaid or Medicare, but the large standard error of 2 percentage points indicates this difference is very likely to have occurred due to chance. When we added covariates, the sign reversed and the difference between surveys remained small and insignificant. Finally, the predictors of government insurance receipt also did not vary across surveys: in the final column, we included interactions between CCES membership and sex, education, race, and age, and found no meaningful interactions.
Endogeneity of Health Insurance Receipt
One concern with cross-sectional data is that partisanship could be endogenous to insurance receipt. We used the CCES panel to explore this concern, testing whether changes in insurance status were related to changes in party identification. Specifically, we regressed the change in 3-point party identification between 2012 and 2014 on indicators for having gained insurance over this period and for having lost insurance over this period. The change in party variable ranges from −2 to +2, as the original scales are 3 points (with 1 = Democrat, 2 = Independent, 3 = Republican). The mean of the change in party variable is −0.002, with a standard deviation of 0.28, and higher values indicate shifts toward the Republican Party.
Table A3 shows the results of these regressions. Those who gain insurance between 2010 and 2014 shift by 0.04 to the Republican Party, but the large standard error of 0.04 suggests this difference could very likely have arisen by chance. Controlling for lagged party identification did not substantively alter these results. We similarly found that losses of health insurance are positively associated with shifts to the Republican Party, but the associations could have easily arisen due to chance. Given that the standard deviation of the partisan change variable is 0.28, these differences are also substantively quite small. Thus, we conclude that insurance receipt does not influence party identification.