This study considers five important questions related to the role of race in state-level public support for the Medicaid expansion: (1) whether public support for the Medicaid expansion varies across the American states; (2) whether public support is positively related to state adoption; (3) whether this support is racialized; (4) whether, if racialized, there is evidence of more state responsiveness to white support than to nonwhite (black and/or Latino) support; and (5) does the size of the nonwhite population matter more when white support is relatively low? Our findings suggest that while public support for the Medicaid expansion is high at the state level, especially in comparison to public support for the ACA, there are important variations across the states. Although overall public support is positively related to state adoption, we find that public support for the Medicaid expansion is racialized in two ways. First, there are large differences in support levels by race; and second, state adoption decisions are positively related to white opinion and do not respond to nonwhite support levels. Most importantly, there is evidence that when the size of the black population increases and white support levels are relatively low, the state is significantly less likely to expand the Medicaid program. Our discussion highlights the democratic deficits and racial bias at the state level around this important coverage policy.
Every once in a while you hear a chorus of voices declaring that race is no longer a problem in America. That's wishful thinking; we are still haunted by our nation's original sin.—Paul Krugman, New York Times, June 22, 2015
This is the concluding sentence written by Paul Krugman in his New York Times column about the role of race in states refusing to pass the Medicaid expansion. He notes, as have several others, that with the exception of Arkansas (and very recently Louisiana), all the southern states in the United States refused to expand Medicaid; and 80 percent of those left uninsured due to this lack of coverage reside in the South where a large proportion are African American (Krugman 2015). Despite attention in the popular press and studies confirming the racialization of the Affordable Care Act (ACA) (Tesler 2012), there has been almost no empirical focus on the role of race in states' decisions to expand Medicaid.
After the Supreme Court's decision giving states the option to expand Medicaid, twenty-five states (including the District of Columbia) elected to expand the program during the first year of implementation, six more states did so as of July 2015, and one more as of July 2016, for a total of thirty-two states adopting the Medicaid expansion (Heberlein et al. 2013; Buettgens, Holahan, and Recht 2015). This means that, despite ACA's original intent to expand coverage to near universalism (undocumented immigrants were never included under the bill), about 3 million Americans remain uninsured due to this gap in coverage (Buettgens, Holahan, and Recht 2015). In the non-expansion states, over half of low-income Americans who were uninsured in 2010 remain without access to affordable coverage (Garfield et al. 2015; KFF 2014b). These Americans fall in the so-called coverage gap: they are not poor enough to qualify for traditional Medicaid, and yet do not earn enough to qualify for subsidies on the exchange (Garfield et al. 2015).
Although it is difficult to argue that this distribution of subsidies is equitable, some might argue that states' decisions are fair based on notions of democratic responsiveness. Indeed, underlying the Supreme Court's claim that the federal ACA Medicaid mandate was too coercive is the invocation of states' rights and the fundamental belief that states should be able to exercise democratic accountability, thereby responding to citizen desires. Yet, while Americans value the benefits of democratic responsiveness that a federalist structure helps attain, several studies confirm that welfare policies enacted and implemented at the state and local level are racially biased (Soss et al. 2001; Soss, Fording, and Schram 2011; Zhu and Clark 2015). Because we know that state discretion with regard to the Medicaid expansion means that many are left uninsured, and given past studies suggesting racial bias, it is important to interrogate whether state-level public opinion about the ACA Medicaid expansion is racialized and, if so, whether racialized opinion has an impact on states' decisions to adopt the Medicaid expansion.
The ACA, Medicaid, and Race
While there are numerous historical accounts that explicate how the development of the US health care system has been racially biased,1 there is recent evidence that public opinion about health care reform has become more racially polarized. In particular, using survey data, Tesler (2012) documents that the racial gap in support for public insurance—with whites being much less supportive than blacks or Latinos—has significantly widened over time. Tesler (2012) also finds, using experimental techniques, that health care policies are significantly more racialized when attributed to President Obama than when they were when attributed to Clinton's reform proposals in 1993–94.
Although these findings are very important, one cannot assume that racialized opinion about the ACA carries over to the Medicaid expansion since the public seems to view the Medicaid expansion as separate and different from the ACA overall. National public opinion polls, for example, report consistently higher levels of support for the Medicaid expansion than support for the ACA overall. In 2013, only 32 percent of Americans had a favorable view of the ACA, whereas over half supported the Medicaid expansion (55 percent) (KFF 2013b). Nonetheless, despite this relative popularity of Medicaid at the national level, it is unclear whether this level of support persists across the states. Although Barrilleaux and Rainey (2014) estimate state-level public opinion, they estimate support for the ACA overall (using national poll data) to determine its influence on state governors' support for the Medicaid expansion. Perhaps not surprisingly, because public support for the ACA overall is uniformly low, they find that state-level public support for the ACA has an insignificant impact on governor support for the Medicaid expansion.2 Because we know there is a significant difference in the level of support for the ACA and the Medicaid expansion at the national level, it is important to look specifically at national polls for the Medicaid expansion (not the ACA overall) when creating state-level estimates. Work by Lax and Phillips (2012) and Pacheco (2011) also highlights the importance of analyzing policy-specific public opinion. Lax and Phillips (2012) estimate state-level support for thirty-nine policies across eight issue areas and find that state policy decisions are strongly associated with policy-specific opinion, much more so than a measure of state liberalism or another more general proxy for public opinion. While state-specific public opinion is generally positively associated with state adoption of the specific policy, Lax and Phillips also show that, under highly salient, controversial issues, such as gay rights policies, states have been shown to be less responsive to public opinion where high levels of support (often requiring supermajorities) are needed before a state will adopt (Lax and Phillips 2009a; Lax and Phillips 2012). Because the Medicaid expansion decision is also highly salient and ideologically charged, it is important to not only look at whether public support for the Medicaid expansion is positively associated with state adoption, but the level of state responsiveness as well.
Moreover, national polls on the Medicaid expansion also reveal important differences by race. Over 82 percent of blacks and over 65 percent of Latinos support the Medicaid expansion, compared to only 46 percent of whites (KFF 2013a). These large differences by race raise questions as to whether public opinion about the Medicaid expansion may—similar to the ACA overall—also be racialized.
In sum, to date, we know little about how public support for the Medicaid expansion varies across the fifty states, whether state adoption of the Medicaid expansion is responsive to state-level support for expansion, whether state-level public support is racialized and, if so, whether racialized public support impacts state adoption decisions.
The Role of Race in State Medicaid Expansions: Integrated Racialized Backlash Theory
The theoretical and empirical work that seeks to understand the sources of racial policy attitudes is substantial—see Krysan's (2000) review and, more recently, Tesler (2013) and Neblo (2008). For simplicity and efficiency reasons, here we focus on two main theoretical perspectives—symbolic racism and racial backlash theories—which are most applicable to understanding the impact of racial prejudice on policy attitudes and, ultimately, policy outcomes.
In the 1970s and 1980s new theories of racial attitudes and opinions emerged, in part to explain why whites showed nearly universal support for the abstract principles of racial equality, while at the same time opposing a number of equal opportunity policies such as affirmative action, fair housing, and school integration (Krysan 2000). Work by Sears and colleagues developed what is now well known as the symbolic racism perspective, which argues that white Americans' opinions related to policies they view as disproportionately impacting racial minorities are driven by “the belief that racial discrimination is largely a thing of the past, that blacks should just work harder to overcome their disadvantages, and that blacks are making excessive demands for special treatment and get too much attention from elites, so their gains are often undeserved” (Sears et al. 1997: 22).
Kinder and Sanders (1996) extended symbolic racism to take account of a strong belief in individualism among whites combined with growing racial anger and indignation, which they argue has resulted in growing racial resentment among whites. Since that time, a number of studies have created validated measures of racial resentment and found strong empirical support for the relationship between racial resentment and opinions regarded as racial policies (see the first arrow on the left in fig. 1a). A recent study reconfirms that racial resentment among white Americans persists, has remained largely unchanged since the 1980s when it was first studied, and continues to influence racial policy attitudes (Knowles et al. 2010; Tuch and Hughes 2011).
This effect also extends to attitudes regarding social welfare programs. In particular, symbolic racism portrays racial minorities as “demanding,” and “undeserving” individuals with a “lack of work ethic and responsibility,” especially in association with social welfare programs (Kinder and Sanders 1996; Huddy and Feldman 2009). Several studies provide empirical support for this aspect of symbolic racism theory—that racial attitudes are a key predictor of white Americans' social policy preferences at the national and state level (Carmines and Stimson 1989; Edsall and Edsall 1991; Kinder and Sanders 1996), and impact white opinion about crime policy (Green, Staerklé, and Sears 2006; Matsueda and Drakulich 2009) and welfare policy (Sniderman, Crosby, and Howell 2000; Tolbert and Hero 2001).
Although there is substantial evidence for the impact of racial resentment on white opinion about welfare policies, there are few studies that test for the impact of racialized opinion on state Medicaid policy outcomes. Most applicable to this study, Lanford and Quadagno (2015) test for the impact of state-level racial resentment on state adoption of the Medicaid expansion, and find a significant negative relationship—higher racial resentment is associated with a lower likelihood of states' adopting. While an important finding, because racial resentment theory argues that resentment impacts white attitudes toward welfare policies, it is important to control and incorporate public opinion in the model, which they do not account for.
Racial Backlash Theory
Alongside studies of racial attitudes have been attempts to look at nonattitudinal measures of racial bias. As Krysan (2000: 157) points out in her review, “It has long been argued that as the size of the black population increases, whites respond with increasingly negative racial attitudes” (Blalock 1957; Pettigrew 1957). This idea has been developed into a group threat theory (Taylor 1998; Taylor 2000), which argues that as the size of the minority population increases, whites feel more threatened, which creates negative attitudes toward perceived race-based policies (Johnson 2001; Matsubayashi and Rocha 2012). While the black proportion does not predict equally well across all policy types, a number of studies indicate that states are less generous in the provision of social benefits, and more likely to enact punitive policies, as the percent of racial minorities among the total state population increases (Plotnick and Winters 1985; Grogan 1994; Quadagno 1996; Tolbert and Hero 2001; Soss, Fording, and Schram 2011; Zhu and Clark 2015) (see fig. 1b).
Latinos have also been subject to racial bias in welfare policy development in the last few decades (Zimmermann and Tumlin 1999). Particularly due to politicized concerns about illegal immigration and misperceptions about welfare dependency among immigrant families, the presence of a large Latino population tends to be associated with lower welfare program benefits and restrictive eligibility criteria for welfare programs, such as Temporary Assistance for Needy Families (Kilty and Vidal de Haymes 2000; Soss et al. 2001; Fellowes and Rowe 2004; Hero and Preuhs 2007).
Integrated Racialized Backlash Theory
While there is a plethora of empirical support for backlash theory, this theory fails to explicate the mechanisms through which size of the black population matters (Taylor 1998; Krysan 2000). The theory suggests that when size increases it is viewed as a threat, but when and under what circumstances? Moreover, as mentioned above, there is substantial evidence that racial resentment impacts white attitudes on social welfare policies, but few studies consider if racialized policy-specific attitudes impact state policy outcomes. We attempt to address these two concerns by bringing the two models together in what we call an Integrated Racialized Backlash Theory (see fig. 1c).
We argue that the size of the racial group (blacks and/or Latinos) will be perceived as a threat when whites are less supportive of the policy under consideration due to underlying racial resentment. Since most of the theories of racial resentment show that it impacts white public opinion, then white opinion should have a significant impact on policy outcomes. However, if racialized white opinion exists for a particular public policy, it is more likely to matter if there is a relatively high proportion of nonwhites in the state. That is, bringing in backlash theory, we argue that a relatively higher proportion of nonwhites will more likely be perceived as a threat if white opinion is more negative against the policy. To test this theory, we argue that an interaction term is important to put in the model. Moreover, several studies note the importance of bringing in opinions of nonwhites instead of just modeling white opinion when considering the role of race in state policy outcomes (Bobo and Hutchings 1996; Dawson 2000). In response to this criticism, our model explicitly incorporates black and Latino opinion.
This integration of theories, along with incorporating policy-specific, state-level public opinion, enables us to examine five important questions related to the role of race in state-level public support for the Medicaid expansion: (1) whether state support for the Medicaid expansion varies across the American states; (2) whether state support is positively related to state adoption; (3) whether this support is racialized; (4) whether, if racialized, there is evidence of more state responsiveness to white support than to nonwhite (black and/or Latino) support; and (5) does the size of the nonwhite population matter more (backlash) when white support is relatively low?
To examine these five questions, we first estimate state-level public opinion, and then conduct logistical regression analysis on the dichotomous Medicaid expansion variable. Data for the dependent variable comes from the Kaiser Family Foundation's (2013c) summary of state Medicaid expansion decisions as of October 2013 when they defined states as either “moving forward” or “not moving forward” (i.e., expansion/non-expansion state). At that time, the decision to expand was exactly half (excluding Washington, DC): twenty-five states adopting and twenty-five states not adopting the Medicaid expansion. Although we could use more recent data on state decisions (seven additional states have now expanded as of March 2016), we continue to use the 2013 data because the timing of our public opinion data (discussed below) corresponds more closely to the 2013 decisions.
To estimate state-level public opinion, we used four national-level survey datasets from the Kaiser Family Foundation's monthly tracking poll, which asked specifically about support for the Medicaid expansion:
As you may know, the health care law expands Medicaid to provide health insurance to more low-income uninsured adults. The federal government will initially pay the entire cost of this expansion, and after several years, states will pay 10 percent and the federal government will pay 90 percent. The Supreme Court ruled that states might choose whether or not to participate in this expansion. What do you think your state should do? Do you think your state should keep Medicaid as it is today or expand Medicaid to cover more low-income uninsured people?
The question was asked four times over ten months (July 2012,3 January 2013, March 2013, and April 2013), allowing us to pool the data for a sample size of 4,516 across all four datasets.
Multilevel Regression and Post-stratification. Because it is cost-prohibitive to conduct state-level public opinion surveys, they are rarely, if ever, done. To our knowledge, there have been no state-level public opinion surveys across all fifty states on questions pertaining to Medicaid policy. As a result, the most commonly used method for estimating state-level opinion has been aggregation, which requires a large set of national polls to acquire the opinion percentages by state. Yet, even with large samples, the disproportionate sample distributions of those surveys have proven to be problematic for accurate state-level estimation (Lax and Phillips 2009b; Pacheco 2011). As mentioned above, due to these limitations, an alternative method, Multilevel Regression and Post-stratification (MRP), has been developed and provides more accurate state-level opinion estimates with smaller errors and higher correlations, particularly when national samples are smaller (Lax and Phillips 2009b; Pacheco 2011).
The MRP method requires two main stages: (1) estimating demographic subgroup responses through utilization of a multilevel model with individual and state-level information (Multilevel-Regression part); and (2) weighting individual responses by the actual state population for each respondent type (Post-stratification part) (see appendix A for our MRP equations, and Lax and Phillips [2009b] and Pacheco  for detailed discussions on the MRP method). This two-step approach produces the estimated percentage of residents who take a particular position—in this case, a public support level on Medicaid expansion for all fifty states.
Three sets of measures for state-level public opinion are estimated for this study. In addition to overall public support, we estimated race-specific support levels as indicated in our theoretical model. The first set estimates overall state residents' support for Medicaid expansion, using all available individual and state-level information described above. Then, we ran modified MRPs4 without using racial information to estimate state-level Medicaid expansion support levels for white and nonwhite residents.5
The validity and efficiency of MRPs are well documented (Lax and Phillips 2009b; Warshaw and Rodden 2012; Caughey and Warshaw 2015). However, to improve internal validity, we performed a bootstrap with five hundred iterations of random sampling with replacements. Based on the bootstrap distribution of individual state parameters, we estimated the average support levels and the 95 percent error margins for fifty states. To confirm the external validity, we compared the MRP estimates with three independent state-level surveys on Medicaid expansion conducted in seven states, including Alabama, California, Georgia, Louisiana, Mississippi, North Carolina, and South Carolina. The confidence intervals for all three surveys and our MRP estimates overlap, a strong indication of external validity (see appendix B for further discussion on the external validity of our estimates).
Modeling State Medicaid Expansion Adoption
As discussed above, in addition to public opinion and race, several other variables have been found to be important predictors in models of state Medicaid policy (and welfare policy more generally), including state poverty rates, political party control (including state legislature party control and the governor's party affiliation), the cost of the health or welfare program, region, and the strength of interest groups (Barrilleaux and Bernick 2003; Bartels 2008; Grogan and Rigby 2009; Gray et al. 2010; Kelly and Witko 2012; Barrilleaux and Rainey 2014).
Most applicable to this study, Barrilleaux and Rainey (2014) conducted a similar study estimating state governors' support for the Medicaid expansion. Their study asks an important question about whether governors are motivated by state-level health care needs (as measured by the percent uninsured in the state). We do not include a separate measure of the percent uninsured because it is incorporated in our poverty rate and state Medicaid cost estimate. Indeed, measures of need (or health need, which they use as an alternative measure) can also be conceptualized as increasing the cost function to the state, which is hypothesized to have a negative impact on state adoption. So, although Barrilleaux and Rainey conceptualize need as having a positive effect on state governors' decisions, a measure of need could logically have a significant negative impact on state policy decision making because it increases the cost of expenditures to the state (Grogan 1994). Indeed, their finding that “need” plays an insignificant role supports this interpretation of an ambiguous effect, because any measure of need will also increase costs to the state.
To capture state-level partisanship, Enns and Koch (2015) offer a validated measure of state residents' degree of support for states' welfare-type programs, called policy preference or policy mood. However, we decided not to use this variable in our models because it incorporates measures of public opinion, and so it was highly correlated with our overall support (corr. = 0.71, p < 0.01) and white support estimates (corr. = 0.74, p < 0.01). Given that public support and race is the main focus of our study, we did not want to muddy this effect. Nonetheless, we did test the policy mood variable in our models to see if it changed the estimates, and our main findings remained robust.
Interest Group Pressure. In addition to public opinion, political demands for the Medicaid expansion also come from provider groups who have consistently advocated for Medicaid benefit and reimbursement expansions because Medicaid payments go directly to provider groups, not as cash payments to reimburse enrollees. Safety-net providers, including public and nonprofit hospitals and community health centers, provide the bulk of care to Medicaid recipients. Although state Medicaid payments to Medicaid providers are notoriously low relative to rates paid by private insurance companies, many public and community nonprofit hospitals and primary care safety-net providers have a cost structure such that Medicaid funds are not only desirable but create financial sustainability (Swan et al. 2000; Wynn et al. 2002; Dranove, Garthwaite and Ody 2016). Given this, safety-net providers have a vested interest in state Medicaid policy and often play an active role in state-level Medicaid policy decisions (Grogan and Gusmano 2007; Olson 2010; Hall and Rosenbaum 2012; Mickey 2012).
Measuring the Strength of Safety-Net Interest Groups. Unfortunately, there is very limited data on interest group strength at the state level, especially for this particularly nuanced sector of health care interests. Although Gray, Lowery, and Benz (2013) conducted a groundbreaking multiyear study of state-level health interest groups, it does not separate out the safety-net institutions. This is important because, even though hospitals, nursing and residential care facilities provide significant levels of care to Medicaid recipients, their interests vary according to ownership type. Moreover, Gray and colleagues' study is based on data from the 1990s, and attempts at updating their earlier estimations would be inaccurate due to limited state disclosure and tracking requirements for lobbyists (Gray, Lowery, and Benz 2013).
Thus, given these limitations, we attempt to capture safety-net interest groups' influences on the Medicaid expansion decision by measuring the size of the safety-net providers in each state as a proxy. For community health centers, we use the total number of patients served in 2011 divided by total state population in 2011 (National Association of Community Health Centers 2014).6 On average, the community health centers provided services for 7 percent of the state population, ranging from 2 percent in Nevada to 21 percent in West Virginia (see table 1).
Another measurement which attempts to capture the size of a state's safety net is the amount of payments that care providers receive for providing a disproportionate share of their care to Medicaid and Medicare recipients (anachronistically called “disproportionate share hospital [DSH] payments”). Presumably, if states have high DSH payments relative to their population size prior to 2014, one would think that hospitals in the state would have a very strong interest in advocating for the Medicaid expansion, especially since provisions under the ACA were supposed to eliminate DSH payment provisions when Medicaid expansion coverage began in 2014. Clearly, safety-net hospitals in states without a Medicaid expansion are much worse off since they will continue to take care of the uninsured (those who fall in the Medicaid gap in coverage) with no compensation in funding, and will lose their DSH payments at the same time (Price and Eibner 2013). In 2011 the average DSH payment per capita was $80 dollars, and ranged from about $4 dollars in Wyoming to almost $500 dollars in Tennessee (see table 1) (KFF 2014a).
The measurements for the other variables in the model are fairly straightforward and well established in the literature. Therefore, with the exception of interest group strength, measurement details are provided in appendix C, and descriptive statistics in table 1.7
Public Support across the American States
While public support for the Medicaid expansion varies across the American states, most states are clustered fairly close to the 50 percent mark, indicating that the public is equally divided in many states (see fig. 2). For example, the average across all the states is 51.0 percent in favor of the expansion (sd = 5.0 percent), and the majority of states (thirty-one) have support levels between 45 and 55 percent. Even the maximum and minimum levels do not suggest extreme levels of support or opposition: the highest public support is 65.2 percent in Hawaii, and the lowest is 42.7 percent in Nebraska. Note that public support for the Medicaid expansion across the states is high relative to the ACA overall—even the lowest level of state support for the Medicaid expansion is still higher than the average national support for the ACA at only 32 percent (KFF 2013b).
In addition to geographic variation in support for the Medicaid expansion, the racial differences found in national surveys hold at the state level as well, confirming a state-level racial opinion gap (Tesler 2012). When employing the MRP methodology to estimate overall state-level support for the Medicaid expansion, race is the key predictive factor where blacks, Latinos, and other racial groups have significantly higher levels of support compared to whites (as we discuss in more detail below).
Overall Public Support and State Adoption
To test whether there is support for our integrated model, we first test the counterfactual: controlling for overall public support for the Medicaid expansion, is there evidence of a direct effect of racial resentment and backlash (size of the nonwhite population) on state decisions to adopt the Medicaid expansion?
As predicted, the likelihood of a state adopting the Medicaid expansion is positively related to state-level public support and partisanship.8 These relationships are robust across several different model specifications (see table 2). While overall public support is positively related to state adoption, indicating policy responsiveness, when considering democratic accountability it is also important to analyze whether policy adoption is congruent with popular will (Lax and Phillips 2012). Using a simple majority (51 percent in favor) as a cutoff point, and taking account of the confidence intervals around the point estimates,9 we find that nineteen states are congruent with popular will: nine adopted the Medicaid expansion with majority support, and ten states with minority support did not adopt (see the dark shaded areas in fig. 3). However, eight states are incongruent with majority support, where four states adopted the Medicaid expansion under minority support and four states did not adopt despite majority support (see lightly shaded area in fig. 3). While a positive relationship with overall support holds, it is clear that other factors are also related to state adoption.
Indeed, consistent with many other studies of state Medicaid and welfare policy, the percent of the black population is also consistently significant across several model specifications (see table 2). In particular, consistent with backlash theory, as the percent of the black population increases, the likelihood of adoption decreases. It is noteworthy that it is only percent black that is significant; percent Latino and percent nonwhite are insignificant (see models 3 and 4 in table 2). While racial resentment is initially significant (see model 2 in table 2), it loses significance once backlash variables are also considered (see models 6–8 in table 2).10
Consistent with theories on racial resentment, our results suggest that there is not a direct effect of racial resentment on state adoption. Instead, it is more likely that racial resentment is related to racialized support for the Medicaid expansion, which may have a direct effect on state adoption (as depicted in fig. 1a). As such, we turn next to consider evidence for racialized support for the Medicaid expansion and its impact on state adoption.
Test of Integrated Racialized Backlash Theory
When we include estimates of white and nonwhite public support,11 white opinion is significantly associated with expansion decisions, whereas nonwhite opinion is insignificant (see models 1 and 2 in table 3). This finding suggests that if whites have a relatively high level of support for the Medicaid expansion, the state has a significantly higher likelihood of adoption, whereas nonwhite support does not matter. Or stated differently, when whites have lower support levels, states are less likely to adopt the Medicaid expansion regardless of the nonwhite level of support. Although we do not have the data to test whether racial resentment affects white opinion for the Medicaid expansion, it is noteworthy that these two variables are highly correlated (corr. = −0.73, p < 0.01). This provides support for the notion that white opinion on the Medicaid expansion is racialized.
Consistent with our previous test, when we include the proportion of black population in the model, it is also significantly associated with Medicaid expansion decisions (see model 2 in table 3). Yet, when we include the interaction term—percent black population with white support—it is strongly significant and positive, consistent with our integrated theory of racialized backlash (see column 3 in table 3). Notice how percent black alone is no longer significant (note also that we dropped nonwhite support since it is highly correlated with percent black at 0.58, p < 0.01). The interaction term reveals that, as the proportion of the nonwhite population increases combined with relatively low levels of white support for the Medicaid expansion, the state is less likely to adopt the Medicaid expansion. Also note that this significant relationship for both white support and the interaction holds across different model specifications—even when we control for the south regional dummy and Medicaid expansion cost (see columns 4–5 in table 3).
Racialized Opinion and Tipping Points
To better understand the impact of race on states' Medicaid expansion decisions, we examine three predicted likelihood plots of state Medicaid expansion adoption, with public support estimates for whites (far left curve), nonwhites (far right curve), and the entire population (including both white and nonwhite) (see fig. 4 and table 4). Theses curves illustrate the policy adoption “tipping points” by racial groups: where each curve crosses the middle of the y-axis denotes the average level of public support needed to adopt the policy (Lax and Phillips 2012).12 For whites, the Medicaid expansion has a 50 percent chance of adoption with 45 percent white support. However, because nonwhite support is insignificant in the model, there is no level which nonwhite support could reach to have an impact on the chance of state adoption. That said, the curve suggests that if nonwhites' support were to play a role, their support levels would have to increase to 73 percent to reach the same 50 percent chance of adoption.
Another way to illustrate the difference in state responsiveness between white and nonwhite is to examine the point estimates by race and by state according to state expansion decisions (see fig. 5). The first clear finding is that 95 percent confidence intervals among white support and nonwhite support are so significantly different that they do not overlap. The second takeaway is a difference in within-group variation and its relationship to state adoption. While the support levels for nonwhites is consistently high across all the states (an average of 73 percent), it has no relationship to state adoption decisions. In contrast, white support was relatively spread out, with a 45 percent average of support, and the two graphs together clearly illustrate how white opinion is the key driver in state responsiveness.
This study examined five main questions: (1) whether state support for the Medicaid expansion varies across the American states; (2) whether state support is positively related to state adoption; (3) whether this support is racialized; (4) whether, if racialized, there is evidence of more state responsiveness to white support than to nonwhite (black and/or Latino) support; and (5) does the size of the nonwhite population matter more (backlash) when white support is relatively low?
While public support for the Medicaid expansion is relatively high even at the state level compared to support for the ACA overall, there is some variation across the states, and, at least on first glance, it appears that states are responsive to overall public support. However, by creating support estimates for whites and nonwhites, we are able to demonstrate that states are only responsive to white opinion. Most importantly, similar to the ACA, there is some evidence that public support for the Medicaid expansion is racialized. Not only is there a large and significant difference between white and nonwhite support levels across the American states, but white opinion is also highly correlated with measures of state-level racial resentment. There is also support for the idea that symbolic racism and backlash theories should be integrated and reconsidered as racialized backlash theory. In particular, there is support for the idea that the size of the nonwhite population matters more (backlash) when white support is relatively low.
The finding that race matters in state-level policy making is not new. Race has also been found to be a predictive factor in many other studies looking specifically at social welfare policies targeted at the poor (Grogan 1994; Tolbert and Hero 2001; Soss, Fording, and Schram 2011; Lanford and Quadagno 2015; Zhu and Clark 2015). What our study adds to this consistent finding is revealing in its connection to racialized public opinion. In particular, the lower tipping points for white support levels, the interaction of low white support with a higher proportion of racial minority creating racialized backlash, and an insignificant relationship between state Medicaid expansion decisions and nonwhite opinion highlights a lack of democratic accountability for nonwhites, especially blacks.
These conclusions raise serious questions about minority representation and whether the denial of access to public health insurance benefits in states that rejected the Medicaid expansion is democratically just. When the ACA was passed in 2010, it was supposed to be a national health care reform in which coverage policies would be consistently implemented across all fifty states. Obviously, the Supreme Court decision on the ACA Medicaid expansion changed that intent. But, the arguments used by the Supreme Court were based on assumptions that states would act in democratically accountable ways. This study raises questions about to whom states are democratically accountable.
Multilevel Regression and Post-stratification Equations for Estimating Public Opinions on Medicaid Expansion
Note: presvote = the proportion of presidential election vote for Democratic candidates. 1 = White; 2 = Black and Latino; 3 = Other.
External Validity of MRP Estimates
To examine the external validity of the MRP estimates, we took an out-of-sample approach, comparing the estimates with other available state-level surveys. We found three state-level surveys asked similar questions on Medicaid expansion across seven states.
(1) The “Deep South and Medicaid Expansion: The View from Alabama, Georgia, Louisiana, Mississippi, and South Carolina”13 is a five-state survey, conducted between March and April 2013. Five hundred adults from each state were asked the following question:
■ “The new health care law expands Medicaid to provide health insurance to more low-income uninsured adults, including adults with no children whose incomes are below about $16,000 a year. The federal government will initially pay the entire cost for three years, and after that states will pay 10 percent and the federal government will pay 90 percent. The Supreme Court ruled that states may choose whether or not to participate in this expansion. What do you think your state should do: keep Medicaid as it is today, with no new funding from the federal government and no change in who will be covered by the program, or expand Medicaid to cover more low-income uninsured people, with the federal government initially paying the entire cost of the expansion and your state eventually paying 10 percent?”
(2) The “2013 TCWF-Field Health Policy Survey (TCWF-Field Poll)”14 is a statewide health policy survey, conducted between June and July 2013 on 1,687 registered voters of California. The survey asked:
■ “The new health care law allows states to expand health insurance programs such as Medi-Cal so they provide health coverage to more low-income uninsured adults. The federal government will initially pay the entire cost of this expansion for three years, and after that they will pay for 90 percent of the costs with the state government paying the difference. Generally speaking, do you favor or oppose California expanding its Medi-Cal program in this way?”
(3) The “New York Times Upshot/Kaiser Family Foundation (NYT/KFF) Polls in Four Southern States”15 was conducted in April 2014 with 4,152 adult residents of four states, including Arkansas, Kentucky, Louisiana, and North Carolina. The following question was asked only for Louisiana and North Carolina residents (n = 2,099):
■ “Which of the following comes closest to your view about what your state SHOULD do? Do you think (Louisiana/North Carolina) should (keep Medicaid as it is today) or (expand Medicaid to cover more low-income people, with the federal government covering most of the cost and the state paying a small portion)?”
The comparison between three independent surveys and the MRP estimates demonstrate an external validity of our state-level Medicaid expansion support estimates (see appendix table B1). All states' error margins of MRP estimates and three independent surveys overlap, and most MRP estimates fall within the range of individual surveys' error margins.
Racial Resentment. Lanford and Quadagno (2015) used the following two statements from the 2012 Cooperative Campaign Election Study to estimate a state's degree of racial resentment: “(CC422a) The Irish, Italians, Jews, and many other minorities overcame prejudice and worked their way up. Blacks should do the same without any special favors,” and “(CC422b) Generations of slavery and discrimination have created conditions that make it difficult for blacks to work their way out of the lower class.” After calculating on an individual-level racial resentment scale—ranging from 1 (most racially sympathetic) to 9 (most racially resentful)—they used the average values as each state's racial resentment measure. The average state-level value was 6.41 (marginally resentful), with values ranging from 5.56 in Hawaii to 7.47 in Louisiana.
Proportion of Racial Minority Population. The American Community Survey's estimate on the state population in 2012 was used to measure the proportion of black and Latino populations of the states. Not surprisingly, there are huge variations in the percentage of nonwhites across the American states. On average, across the states, nonwhites made up 29.0 percent of the population (sd = 15.51 percent). Hawaii showed the highest nonwhite population proportion (77.3 percent), while Maine had only 5.7 percent. Blacks accounted for about 10.1 percent (sd = 9.5 percent) of state populations in 2012, ranging from 0.4 percent in Montana to 37.2 percent in Mississippi. Latinos accounted for about 10.8 percent (sd = 10.1 percent) of state populations in 2012, ranging from 1.2 percent in West Virginia to 46.7 percent in New Mexico.
Democratic Governor. As a proxy of state political party control, we used state governor's Democratic Party affiliation as of June 2013. The data were drawn from the National Governors Association. Originally, we planned to use the combination of two variables: the percentage of state Democratic legislators across the two chambers, and the state governor's party affiliation. However, the planned measure was highly correlated with white opinion (corr. = 0.65) and racial resentment (corr. = −0.52). The Democratic governor measure was also correlated with those variables, but in lesser degrees, 0.54 and −0.46, respectively. In total, twenty states had Democratic governors in August 2013.
Poverty Rate. As a proxy of state fiscal capacity and Medicaid need, we used the American Community Survey's state-level three-year (2010–12) poverty rate estimate. The US average poverty rate was 15.0 percent (sd = 3.2 percent), and ranges from 8.9 percent in New Hampshire to 23.0 percent in Mississippi.
State's Cost for Medicaid Expansion. To measure the states' expected additional expenses for adopting the ACA Medicaid expansion, we used the Urban Institute's estimate (Holahan et al. 2012). The incremental impact of the Medicaid expansion was estimated and we used the change in the percent share of the state's expenditure. Delaware showed the highest expected saving (−11.0 percent) thru Medicaid expansion, while Mississippi showed the lowest saving (6.6 percent), which means Mississippi was estimated to have a 6.6 percent additional expenditure under ACA with Medicaid expansion. The average incremental cost change was 1.5 percent (sd = 4.0 percent).
1. There are too many to list here, but for starters see: Smith (2016), Wailoo (2001), and Gamble (1995).
2. For details on their public support measure and approach, see their technical appendix at www.carlislerainey.com/papers/need-appendix.pdf.
3. The July 2012 survey wording was slightly different: “As you may know, the health care law expands Medicaid to provide health insurance to more low-income uninsured adults, including adults with no children whose incomes are below about $16,000 a year. The federal government will initially pay the entire cost of this expansion, and after several years, states will pay 10 percent and the federal government will pay 90 percent. The Supreme Court ruled that states may choose whether or not to participate in this expansion. What do you think your state should do? (Keep Medicaid as it is today, with no new funding from the federal government and no change in who will be covered by the program, or expand Medicaid to cover more low-income uninsured people, with the federal government initially paying the entire cost of the expansion and your state eventually paying 10 percent?”
4. Since the sample was segmented by race, the new MRP estimates exclude and components from the equation described in appendix A.
5. Although MRP is a superior method of estimating state-level public opinions with midsized to small nationally representative samples, roughly one thousand responses are needed to perform with little validity issues (Lax and Phillips 2009b). Unfortunately, the numbers of racial minority respondents were not large enough to perform MRP for each group (464 blacks and 738 Latinos and others). As an alternative, we decided to merge the nonwhite respondent groups to compare with the opinions of white respondents (n = 3,314), large enough to perform MRP individually.
6. Because Nevada's total number of patients served in 2011 data was not available from the National Association of Community Health Centers, we used the latest available—2006 data—as a proxy, and divided by Nevada's 2006 total population.
7. Some independent variables are highly correlated. For instance, racial resentment was highly correlated with whites' support (corr. = −73, p < 0.01), cost of Medicaid expansion (corr. = 0.64, p < 0.01), and southern state dummy (corr. = 0.62, p < 0.01). To avoid potential multicollinearity in regression analysis, we ran multiple models with different specifications in tables 2 and 3.
8. Note that we used a common measure of party control, which incorporates party control of the legislature and the governorship, and findings are the same as shown in table 2. However, we show the results using Democratic governorship (whether the governor is a Democrat or Republican) because, as we explain below, the party control variable in table 3 is highly correlated with white opinion. To avoid complications with multicollinearity in our model, we replaced it with Democratic governorship, and we wanted to show consistency across tables 2 and 3 so that the coefficients and significance levels could be compared.
9. The 95 percent error margins for individual state's public support levels also varied from 2.6 percent in Georgia, North Carolina, and South Carolina to 7.3 percent in Utah. Multiple factors can affect the size of error margins, such as sample size and homogeneity of state population.
10. Note that we attempted to control for other nonmeasurable effects in southern states by including a southern dummy variable. However, because it is so highly correlated with the race variables in the model (percent black at 0.72 and racial resentment at 0.62), this created a multicollinearity problem and eliminated all the effects (see model 7 in table 2). Note also that we attempted to control for the cost of the Medicaid program as described in the methods section, but this variable is also highly correlated with racial resentment (0.64). The high odds ratio (wrong direction) strongly suggests that multicollinearity is problem here as well (see model 8 in table 2).
11. Ideally, we would have separate support levels for blacks and Latinos, but the sample sizes were too small to create reliable estimates.
12. Because the nonwhite group's opinion does not have statistically significant association with state decisions, tipping point analysis cannot be applied.