Inequality poses a challenge for nations across the globe – yet the United States is often seen as exceptional. Not only is the US experiencing high levels of income and wealth inequality historically, but it is one of the most unequal nations in the developed world. How did one of the world’s richest democracies also become one of the most unequal? The state of economic inequality is shaped largely by government policy and parties in power (Bartels Reference Bartels2016; Kelly Reference Kelly2009). Scholars have theorized that the US lacks the institutional structure for larger-scale redistribution (Alesina and Glaeser Reference Alesina and Glaeser2004) and that policymakers are more responsive to the interests of the rich and organized interests than the average citizen (Gilens and Page Reference Gilens and Page2014). However, what is particularly puzzling is the response of the American public: over time citizens have not met rising inequality with strong demand for government efforts aimed at decreasing inequality that we would expect. However, there leaves more work to be done to understand why the public has met rising inequality with limited liberalism.
I address this puzzle by pointing to the role of increasing racial and ethnic diversity. Specifically, I theorize that citizens prefer conditional equality, where Americans support equality-enhancing policies conditional on the notion that they perceive fellow citizens as “people like me.” The more citizens identify with one another, the more likely they are to empathize with and in turn support policies that benefit others. However, as the racial fabric in various corners of the US steadily becomes less homogeneous, the likelihood of this condition declines. The less often people see others around them as people like themselves, the less likely it is that the public supports policies aimed at helping others and equalizing resources.
To test my theory, I build on recent important work (Franko Reference Franko2016; Franko and Witko Reference Franko and Witko2017) by highlighting the relevance of subnational context in shaping redistributive preferences. Specifically, I leverage variation in both inequality and racial homogeneity of the American states. Using time-series cross-sectional data for an almost 45-year time span, I find evidence that as racial diversity increases, the less likely the mass public responds to increasing inequality by wanting the government to do more. This is true for opinion toward redistributive policies collectively, as well as policy-specific attitudes – including support for welfare and public education spending. Moreover, I find this pattern is specific to economic liberalism, not liberalism on social issues (e.g., gay rights, prayer in school). Finally, I demonstrate that both rising levels and changes in racial diversity shape public opinion in this way. While these findings fit into a large literature in comparative politics that identifies the relationship between national diversity and social policy, this study’s contribution lies in providing evidence of the typically implied (but not always tested) intermediary between diversity and redistributive policies: aggregate preferences.
The puzzle
Scholars continue to be puzzled by the fact that as economic inequality in the US has risen drastically in the past few decades, the public has not demanded more expansive redistributive policies at a rate we would expect given these extreme disparities. When considering the nation as a whole, some research suggests the public has in fact become more conservative (i.e. wants the government to do less) in response to rising inequality beginning in the 1970s (Kelly and Enns Reference Kelly and Enns2010; Luttig Reference Luttig2013). More so puzzling, this trend in policy preferences transcends class divides. Even among many low-income voters – those on the losing end of inequality who have more to gain from greater redistribution – there is continuing resistance to government efforts (Cramer Reference Cramer2016; Hochschild Reference Hochschild2018; Kelly Reference Kelly2020; Kelly and Enns Reference Kelly and Enns2010). Recently, researchers have drawn attention to important factors that come to play a role in redistributive preferences in light of rising inequality, including subnational economic contexts and the types of redistribution being considered (Cavaillé and Trump Reference Cavaillé and Trump2015; Franko and Witko Reference Franko and Witko2017). For example, Franko (Reference Franko2016) finds that when examining inequality at the state level, growth in income inequality leads to greater levels of public support for redistribution, and importantly, this relationship is most pronounced in states with lower average income. Additionally, McCall (Reference McCall2013) notes that nontraditional forms of redistribution that focus on opportunity expansion, such as education, are a more preferred avenue for addressing inequalities among the public than government assistance programs, such as welfare. But how can we reconcile these findings with the trends we see at the national level? On the one hand, there is evidence state inequality leads citizens to become more supportive of redistribution, particularly for education (Franko Reference Franko2016); yet the opposite is true when shifting the unit of analysis to the nation (Kelly and Enns Reference Kelly and Enns2010). Indeed, even aggregate support for education spending has in fact not risen since the late 1980s.Footnote 1 Ultimately, there is still more to understand concerning how the public responds to rising inequality, and just as importantly, why.
Why should we expect Americans to respond to growing inequality by supporting policies that decrease inequality? And why is it so puzzling that they do not? From a rationalist economic perspective, as the proportion of total income continues to funnel into a narrow segment of society at the top, the rest of Americans are relatively worse off. Thus, as inequality increases, the number of people who stand to benefit economically from policies that help to equalize resources also increases. Second, Americans are increasingly aware of inequality (Franko Reference Franko2017; Stimson and Wager Reference Stimson and Wager2020), dislike inequality, and want the gap between the rich and poor to narrow (McCall Reference McCall2013; Piston Reference Piston2018). For example, in 2016, when asked if they believe the government has gone too far, not gone far enough, or just about right to address income inequality, only 7% of Americans reported that government had gone too far.Footnote 2 This indicates that, at least in the abstract, most Americans are not opposed to reducing inequality. Lastly, public preferences are not inconsequential. Aggregate opinion toward policy shapes election and policy outcomes in ways that can effectively reduce disparities (Kelly Reference Kelly2009). Thus Americans’ weak support for the expansion of government efforts to target inequality has in fact allowed for greater inequality, contributing to a pernicious self-reinforcing cycle (Kelly Reference Kelly2020).
Popular economic models of redistributive preferences (Benabou Reference Benabou2000; Meltzer and Richard Reference Meltzer and Richard1981), while having some explanatory power in other national contexts, are unable to fully explain the trend we observe in the US. It is possible these models lack predictive power in the American context because they ignore characteristics specific to the US electorate. For example, although Americans in principle endorse equality, they often fail to translate this sentiment into support for the appropriate policies (Bartels Reference Bartels2016; Kluegel and Smith Reference Kluegel and Smith1986). Early work in political science attributed this disconnect to the value conflict that emerges between egalitarianism and traditional American values, such as economic individualism and meritocracy (McClosky and Zaller Reference McClosky and Zaller1984). However, recent scholarship has sought to reevaluate some of these characterizations. For example, McCall (Reference McCall2013) demonstrates that the belief one can get ahead with hard work is not related to beliefs about the causes of inequality. This implies that simply endorsing meritocracy does not necessarily mean one perceives the state of inequality as deserved. Further, in contrast to meritocratic notions that the rich and poor are deserving of their status, Piston (Reference Piston2018) illustrates there exists resentment toward the rich (and sympathy for the poor) among the public. Taken together, American values, ignorance, or apathy toward disparities cannot fully explain Americans’ economic conservatism in the face of rising inequality.
Race and redistribution
Racial and ethnic divisions provide one of the most formidable challenges to stronger welfare states. For example, greater racial diversity is related to lower levels of welfare spending at both the national and subnational levels (Alesina, Baqir, and Easterly Reference Alesina, Baqir and Easterly1999; Alesina and Glaeser Reference Alesina and Glaeser2004; Desmet, Weber, and Ortuño-Ortn Reference Desmet, Weber and Ortuño-Ortn2009; Filindra Reference Filindra2013; Hero and Tolbert Reference Hero and Tolbert1996). Scholars attribute this association frequently (and implicitly) to voter preferences: the majority group will oppose government spending the more they perceive other racial groups as the beneficiaries. However, scholars examining the causal effect of diversity on public preferences across the globe have come to mixed conclusions, where evidence supporting the diversity theory is often drawn from the US context (Steele Reference Steele2016). Specifically, the American public’s opposition to welfare has been traced back to the racialized discourse of welfare policies, which reinforced the stereotype that blacks were disproportionate and undeserving beneficiaries (Gilens Reference Gilens1999; Quadagno Reference Quadagno1994). However, welfare spending is not unwaveringly opposed: individual support for welfare rises when the perceived number of beneficiaries from their own racial group increases (Luttmer Reference Luttmer2001). But while there is evidence demonstrating racial considerations play a significant role in Americans’ attitudes toward welfare, welfare constitutes a only small fraction of domestic spending (Kelly Reference Kelly2009). The central puzzle of this article is why the American public responds to rising inequality with less liberalism overall – not just attitudes toward welfare. We thus need to understand what common thread connects the public’s perception of various redistributive policies that target inequality.
While scholarship discussing the racialization of government programs frequently points to welfare, other programs that target income inequality are subject to racialization as well. Social Security, Medicare, and public education have historically experienced more support from voters than welfare (Stimson Reference Stimson2015). However, as the perceived beneficiaries of these policies change, so too should the level of public support. For example, social scientists have found evidence of a “Florida effect,” where the state’s average public school student is Latino, but the average taxpayer is white. In turn, there is less support for education spending in states like Florida than in states where students and taxpayers are more likely to be of the same race (Harris, Evans, and Schwab Reference Harris, Evans and Schwab2001; Poterba Reference Poterba1997). These studies are illustrative of a large literature demonstrating social identities shape redistributive preferences, where group members are more likely to support policies when they share a social identity with the perceived beneficiaries (see Costa-Font and Cowell Reference Costa-Font and Cowell2015 for a review). This pattern of behavior is explained by the tendency for group members to have favoritism toward their ingroup (Tajfel and Turner Reference Tajfel and Turner1986). Indeed, people often perceive ingroup members as more similar to themselves than outgroup members; because we feel closer and more connected to ingroup members, we also tend to feel greater empathy and responsibility toward ingroup members than we do for outgroup members (Mullen, Brown, and Smith Reference Mullen, Brown and Smith1992). Although other social identities splinter the American public, race is undoubtedly one of the most consistently salient and divisive (Hutchings and Valentino Reference Hutchings and Valentino2004). This suggests that support for equalizing policies is shaped to some extent by perceptions of who benefits, and this is informed – at least partially – by the racial identities of those beneficiaries.
People like us? Why preferences for equality are conditional?
While whites continue to constitute a narrow majority in the US, the racial fabric of the country has become decreasingly homogeneous. But what effect should this have on Americans’ support for measures targeting inequality? We know that Americans are increasingly aware of inequality, dislike inequality, and would like to see the US become a more equal nation. Given this, as income disparities continue to grow, we would presume that Americans would respond by supporting government policies that serve to narrow this gap. However, I pose that what citizens really believe in is conditional equality, where they support equalizing policies as long as they perceive the policy beneficiaries as “people like us.”
Experimental research demonstrates that inducing economic anxiety can have significant implications for individuals’ willingness to help others. Specifically, contexts of economic scarcity or inequality shift the way people respond to members of other social groups. For example, scarcity can lead people to devalue others’ deservingness and can foster antipathy toward other social groups, strengthen negative out-group stereotypes (Krosch and Amodio Reference Krosch and Amodio2014), and increase the likelihood of discriminatory resource allocation (LeVine and Campbell Reference LeVine and Campbell1972; Sherif Reference Sherif1966). Recently, Condon and Wichowsky (Reference Condon and Wichowsky2020) found that priming inequality increases social comparison, particularly downward comparison; and when one feels unempathetic to the comparison, they are less likely to support helping them.
The idea that redistributive preferences are shaped by empathy is consistent with the social affinity literature, which suggests that the degree to which we care about others is shaped by how closely we perceive them as like ourselves. This phenomenon is highlighted by work demonstrating support for redistribution declines when people perceive that it benefits undeserving members of outgroups, particularly racial outgroups (Alesina and Glaeser Reference Alesina and Glaeser2004; Gilens Reference Gilens1999; Luttmer Reference Luttmer2001). Notably, researchers have argued for several different theories predicting the mechanism through which increasing diversity shapes public opinion. For example, when members of a dominant group, such as whites in the US, perceive their dominant status as those at the top of the racial hierarchy to be threatened by racial others, this leads to the activation of racial prejudice (Blumer Reference Blumer1958; Bobo and Hutchings Reference Bobo and Hutchings1996). While racial group threat can be felt by whites for a variety of reasons, a growing minority population is one driver of perceptions of threat (Christiani Reference Christiani2020; Jardina Reference Jardina2019). Though empathy can be linked to threat, there are cases where low levels of racial empathy can still be found among those who do not perceive racial others as significant threats. In addition to racial threat, increases in diversity are also associated with lower levels of social trust and social capital (Hero Reference Hero2007; Putnam Reference Putnam2007). But these constructs – trust, threat, empathy – should not be seen as mutually exclusive. While I do not directly test the underlying mechanism in this study, previous work has demonstrated that empathy is one of the most powerful predictors of peoples’ willingness or refusal to help others (Condon and Wichowsky Reference Condon and Wichowsky2020). Moreover, though Americans’ conceptions of “people like us,” or their community, can expand beyond members of their own racial in-group and include other out-group members (Wong Reference Wong2010), race and ethnicity remain a critical factor in shaping perceptions of closeness to other people (Wilkinson Reference Wilkinson2015).
When people consider their environment and see it is composed of members of racial out-groups, it signals that there are more of “them” and fewer of “us.” They will have a heightened perception that the public is composed of others, and in turn, they will be less likely to perceive “us” as the beneficiaries of social policies. Since people are more likely to give to members of their own racial group, they will be less likely to support government intervention when racial homogeneity declines. In this case, the public will be less willing to support equalizing policies in response to increases in inequality. However, when contexts are more homogeneous, citizens should be more likely to empathize with those around them and in turn policy beneficiaries. They then will respond to inequality by wanting the government to do more. Thus, voters’ redistributive preferences should be a function of actual income inequality conditional on the racial diversity of their contexts. Formally, I expect that as racial diversity increases, the less likely the mass public should respond to increasing inequality with support greater redistribution of resources (H1). This accounts for those in homogeneous contexts, who on average should respond to inequality with demand for greater redistribution. This hypothesis also accounts for those in diverse contexts, who I expect to respond to inequality with less demand for redistribution.
This theory aims to reconcile scholars’ findings in different units of analysis. Recall that at the state level rising inequality leads to demand for greater redistribution (Franko Reference Franko2016), yet this is seemingly a direct contradiction to Kelly and Enns’ (Reference Kelly and Enns2010) findings at the national level. My expectations indicate that in some contexts inequality can indeed lead to demand for redistribution – under the right conditions. But these conditions (racial homogeneity) have become increasingly rare over the past few decades.Footnote 3
Data and methods
The notion that we are more likely to help people that we perceive as people like ourselves is not new to the political science literature. Recent work demonstrates that Americans’ perceptions of the “other” have implications for their redistributive preferences, where we are more likely to want to help others that we perceive are like us (Theiss-Morse Reference Theiss-Morse2009; Wong Reference Wong2010). However, this scholarship does not consider how rising levels of inequality, combined with racial diversity, shapes citizens’ willingness to help other citizens. Moreover, these findings often rely on cross-sectional data at the individual level. I opt to test my expectations overtime at the aggregate level, where there are direct implications for policy outcomes at both the state and federal levels (Caughey and Warshaw Reference Caughey and Warshaw2017; Erikson, Wright, and McIver Reference Erikson, Wright and McIver1993; Stimson, MacKuen, and Erikson Reference Stimson, MacKuen and Erikson1995). Given that scholarship examining the consequences of diversity for redistributive politics often looks to policy output or government spending as the outcome variable, I shift the attention to the presumed intermediary between diversity and policy – aggregate preferences.
To test the conditional effect of rising diversity on the aggregate public’s response to inequality, I turn to the American states. While scholarship examining the public’s response to inequality has used the nation as the unit of analysis, there is substantial variation in both income inequality and racial diversity across states. Further, at the national level economic inequality and racial diversity trend upwards over time and are highly correlated; thus, a national-level analysis risks spurious inference. For this reason, variation across states is particularly useful in assessing the interactive effect of diversity and inequality. States also have authority over various policy domains, and in turn, have power to shape economic outcomes. From Medicaid expansion to welfare benefit requirements to public education, state decisions on spending and taxation shape actual levels of inequality within their jurisdiction (Franko and Witko Reference Franko and Witko2017; Kelly and Witko Reference Kelly and Witko2012). As described earlier, recent scholarship on inequality has also directed our attention to the states, demonstrating that people accurately perceive changes in inequality in their state over time (Franko Reference Franko2017) and policy preferences are responsive to inequality (Franko Reference Franko2016).
To illustrate the changing racial makeup of the US, Figure 1 maps the racial diversity in each state in 1970 and 2014, respectively. Higher values on the diversity index (darker shades) indicate there is a lower probability that two randomly selected individuals in a state are of the same race. A comparison of the figures suggests there is variance both across time and across states. While Southern states on average are typically more racially diverse than the majority of the country, over time most states across the nation have become more diverse.Footnote 4 Variations in income inequality over time are shown at the bottom of Figure 1. The maps illustrate the share of income that goes to the top 1% of earners in each state in both 1970 and 2014, where higher values indicate more inequality. In 1970, almost all states’ top earners earned less than 10% of the total income earned in the state. By 2014, almost all states’ top 1% earn 15%–30% of the total income earned in the state.Footnote 5
Figure 1. Racial diversity and income inequality by state over time. Darker shades indicate more diversity and more inequality.
One of the primary drawbacks of prior scholarship that tests the relationship between diversity and opinion is the inability to identify causation, often a result of studying a single point in time (Steele Reference Steele2016; Stichnoth and Van der Straeten Reference Stichnoth and Van der Straeten2013). While previous work has been informative, a dynamic analysis offers greater certainty that we are testing the presence of causal processes (Stimson Reference Stimson1985). To determine the effect of state-level inequality and racial diversity on state public opinion, I make use of time-series cross-sectional data from 1970 through 2014 for all American states. By examining how variation in racial diversity shapes the public’s response to inequality, this approach improves upon and complements other work that relies on cross-sectional data. This dataset is also valuable in its breadth by allowing me to model public opinion across every American state over several decades.
Measuring economic liberalism
Liberalism, broadly defined, is the preference for bigger government (Ellis and Stimson Reference Ellis and Stimson2012). According to my theory, racial diversity should shape voter liberalism for a range of policies targeting inequality. As long as the policy in question – whether it is aimed more at equalizing outcomes or opportunity – is ultimately about who gets what, it falls under the economic issue umbrella. To measure overall attitudes toward larger government at the state level, I rely on the state economic liberalism measure from provided by Caughey and Warshaw (Reference Caughey and Warshaw2017), who use a dynamic, hierarchical group-level item-response (IRT) model to infer latent measures of public opinion by state. The measure is similar in nature to Stimson’s (Reference Stimson2018) public policy mood, which aggregates thousands of responses to various survey questions across policy issues, providing a general indicator of sentiment for preferences for policy change over time. Although others have developed estimates of public opinion and ideology at the state level (Berry et al. Reference Berry, Ringquist, Fording and Hanson2007; Enns and Koch Reference Enns and Koch2013), I opt to use the Caughey and Warshaw estimates for several reasons. First, because this study involves assessing economic preferences, and not necessarily overall liberalism, these estimates of state public opinion are useful because they have been divided into separate economic and social dimensions. Although mood for social and economic issues often move in close parallel over time at the national level (Stimson Reference Stimson2015), there is variation between the two at the state level. Economic survey items address issues like taxes, social welfare, and labor regulation, while social issue items measure attitudes toward abortion, gay rights, and other cultural – though not racial – issues. My objective is to measure preferences for more or less government intervention in economic outcomes – thus, using economic liberalism as the outcome variable is the appropriate choice.Footnote 6 Further, I choose to use the IRT estimates as my dependent variable because this approach to modeling opinion data accounts for the paucity of questions asked consistently over time. Finally, this data is also available over an extensive period of time for all states. I refer to this variable as Public Economic Liberalism, where higher values indicate increases in economic liberalism. To ensure that state rising diversity and inequality are specifically affecting economic liberalism and not overall liberalism, I also estimate a model predicting public social liberalism.
In addition to overall economic liberalism, I include two other measures, support for welfare and education spending, as alternative outcome variables to evaluate the influence of inequality and racial diversity on public support for specific redistributive policies. Following Franko (Reference Franko2016), I include these variables to determine if and how the public is responsive to state racial and economic contexts for certain types of redistributive policies. Education and welfare offer strong foils to one another. While both serve to narrow the equality gap, they do so in different ways. Further, Americans notoriously dislike welfare, while they simultaneously perceive public education as a more preferable way to address rising inequality (McCall Reference McCall2013). Public opinion toward spending on education and welfare at the state level was estimated through multilevel regression with post-stratification by Pacheco (Reference Pacheco2014). The author provides the percent in each state population that supports more spending annually for welfare (1974–2000) and education (1975–2000). These two variables I refer to as Welfare support and Education support. These variables were recoded to a 0 to 1 scale where higher values indicate preferences for more government spending.
Explanatory variables
The first key explanatory variable used to test my expectations discussed above is the Top 1% income share in each state by year (Frank Reference Frank2009). Top income share is the proportion of total income held by the top 1% of income earners in each state, where higher values indicate more state-level inequality. There are many ways to estimate economic inequality (Frank Reference Frank2009), but I choose this variable because it reflects the skew of the changing income distribution, as rising income disparities are largely a result of the rapid expansion of incomes at the top (Piketty Reference Piketty2017). Further, recent work has shown that the state publics’ perceptions of inequality accurately follow objective measures, including top 1% and 10% income shares; however, perceptions do not as closely follow changes in the Gini coefficient, a discipline standard for estimating inequality (Franko Reference Franko2017).Footnote 7
The second explanatory variable is Racial Diversity. To calculate racial diversity into a single index, I rely upon the Herfindahl index, which is the estimated probability that two randomly selected individuals are from the same group. This measure has been widely used in research examining the consequences of diversity (Alesina, Baqir, and Easterly Reference Alesina, Baqir and Easterly1999; Hopkins Reference Hopkins2011; Putnam Reference Putnam2007; Trounstine Reference Trounstine2016). Diversity is calculated as 1 minus the Herfindahl index. The formula is defined below:
$$ D=1-\sum \limits_{i=1}^n{r}_i^2, $$
where D, diversity, N, number of groups, and r, the size of each group as a percentage of the population. The variable is coded from 0 to 1, and higher values on the index indicate a declining probability that two randomly selected residents are of the same racial group.
Obtaining reliable data on race over a long time period that includes multiple racial/ethnic groups is difficult. Over the decades, the Census Bureau’s method for collecting data on race has evolved substantially, increasing the number of racial and ethnic categories one can identify with and allowing for identification with more than one category. In order to use a measure that is consistent over time, I rely on data provided by the Survey of Epidemiology and End Results Program (SEER). Using race estimates provided by the census to the National Cancer Institute, SEER has calculated race estimates at the state level since 1969. This measure categorizes people into one of three groups – White, Black, and “other.” In this categorization scheme, “white” includes both Hispanic and non-Hispanic whites, and thus my indicator for diversity is an underestimation of actual diversity (in some places more than others). However, given its long time span, I use this racial data as my main measure of racial diversity. While this measure cannot capture the proportion of those who are Latino, it does account for percent Black. Though the proportion of the population that is Black at the national level has remained level, there is variation over time across states, such as in Maryland, New York, and Georgia.Footnote 8 Given that this measure cannot perfectly capture racial diversity, for robustness I also look to more recent data on race that accounts for multiple racial and ethnic groups. Specifically, I rely on recent Census data that reports the proportion of Whites, Blacks, Asians, Latinos, and Native Americans for every state year between 1990 and 2014. This data is available for a significantly shorter time period than my original measure. But when diversity is calculated (Herfindahl index) using this alternative data on race, I find the two diversity indicators are highly correlated (ranging from .75 to .99 depending on the state). I also estimate models using this more detailed indicator of diversity, and the findings help to corroborate the main results presented here. More details on these measures of racial diversity are discussed in later sections of this study as well as in the Appendix.
One question that arises is whether the public is even aware of the racial makeup of their environment. However, Americans may be better at picking up on objective realities over time than those skeptical assume (Stimson and Wager Reference Stimson and Wager2020). While I do not expect the average citizen can estimate the racial make-up of their state with precision, I do assume residents of Vermont perceive their state as whiter than residents of Louisiana. This assumption is supported by research demonstrating individuals are able to perceive objective trends in both economic and racial contexts in local contexts (Newman et al. Reference Newman, Velez, Hartman and Bankert2015). Further, there is debate over whether it is levels of racial diversity or changes in racial diversity that are really the driving force behind changing political preferences. Both higher levels and changes in racial diversity can signal diversity, and I see no reason to treat the two as mutually exclusive in terms of their effect on liberalism. Thus, I consider not only baseline levels but also change. To create a measure of change in diversity, I simply subtract the lagged value of racial diversity from its current value. For this variable, Racial Diversity
$ \Delta $
, negative values indicate a decrease in the diversity indicator (less diverse) from the previous year and positive values indicate an increase in diversity.
Similar to recent work on public responsiveness to inequality (Franko Reference Franko2016; Macdonald Reference Macdonald2019; Wright Reference Wright2018), I include a series of control variables that influence public opinion on economic issues. These include the unemployment rate,Footnote 9 the percent of state residents who are part of a union, the natural log of the population, and the natural log of mean income as controls. I also control for state macropartisanship defined as the percent of each states’ population that identifies as a Democrat, using estimates provided by Caughey and Warshaw (Reference Caughey and Warshaw2017). Further, to control for the potential of a policy feedback effect (Soroka and Wlezien Reference Soroka and Wlezien2010), a measure of government Economic Policy Liberalism calculated by Caughey and Warshaw (Reference Caughey and Warshaw2016), is included. This measure of policy liberalism is based on hundreds of state policies and was estimated using a dynamic Bayesian factor-analytic model for mixed data. The authors again split policy liberalism into two dimensions: economic and social. For my main models, I only include the former as a control. Economic government liberalism covers a wide range of policy areas, including social welfare, taxation, and labor but excludes policies on cultural and social issues.Footnote 10 Higher values indicate greater government expansion. Finally, to account for the possibility that preferences for spending may be a response to actual levels of spending for specific policies, I control for per capita spending in each state annually for both public education and welfare (see Franko Reference Franko2016). All variables were measured at the state level by year. The descriptive statistics for the main variables of interests can be seen in Table 1.Footnote 11
Table 1. Summary statistics
Model specification
To determine the effect of state-level inequality and racial diversity on public opinion, I estimate a time-series, cross-sectional model, with both state and year-fixed effects. The year-fixed effects account for national-level shocks and the state-fixed effects are intended to account for time-invariant differences across states. The inclusion of fixed effects is a conservative choice, placing a more rigorous screen on spuriousness.Footnote 12 The first model is defined below, where s and t index the states and years in my dataset,
$ {y}_{st} $
is an aggregate opinion,
$ {I}_{st} $
is an indicator of inequality, and
$ {D}_{st} $
is an indicator of racial diversity in that year.
$ {\beta}_1 $
and
$ {\beta}_2 $
are the effects of inequality and diversity, respectively. This model also includes an interaction between the two explanatory variables: inequality and racial diversity, captured by
$ {\beta}_3{I}_{st}{D}_{st} $
.Footnote 13 Lastly, to account for overtime dynamics (Beck and Katz Reference Beck and Katz1995; Reference Beck and Katz2011), I include a lagged dependent variable, denoted by
$ {\beta}_4{y}_{st-1} $
, in the model.Footnote 14
$$ {y}_{s,t}={\beta}_1{I}_{st}+{\beta}_2{D}_{st}+{\beta}_3{I}_{st}{D}_{st}+{\beta}_4{y}_{st-1}+{\alpha}_s+{\varepsilon}_t+{\varepsilon}_{st}. $$
Results
The regression model results are shown in Table 2, where the dependent variable for the first three models is the public’s economic liberalism. My theory predicts that racial diversity should condition the effect of inequality on citizens preferences’ for greater government. The first model contains an interaction term between inequality and racial diversity, excluding control variables. If my hypothesis is correct, this interaction term should be significant and negative. Indeed, confirming expectations, the interaction term in the first model is significant at the
$ p<0.01 $
level, suggesting the effect of inequality does vary by the racial make-up of the state. Specifically, the effect of inequality on public economic liberalism decreases by .05 for every unit increase in diversity. When the long-run multiplier effect is fully accounted for, the total impact is −.07 (−.05/.67). Model 2 is identical to the first but includes control variables that also shape public opinion, which due to data availability leads to a shorter time series (1976–2014). The coefficient of the interaction term between inequality and diversity also is statistically significant and negative.
Table 2. Effect of state inequality and racial diversity on public liberalism (1970–2014)
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses. Durbin Watson tests for all models do not reject the null of no autocorrelation. ***
$ p<0.01 $
, **
$ p<0.05 $
, and *
$ p<0.1 $
.
For clearer interpretation, I plot the estimated marginal effect from Model 2 and the 90% confidence interval over the range of racial diversity in Figure 2. Higher values of racial diversity indicate a state has more racial diversity. The solid sloping line in Figure 2 indicates how the marginal effect of inequality on state public economic liberalism varies by racial diversity. The confidence intervals around the line suggest the conditions under which increases in inequality have a statistically significant effect on liberalism – they have a statistically significant effect whenever the upper and lower bounds of the confidence interval are both above (or below) the zero line. The dashes on the rug plot on the x-axis indicate observations. The figure suggests that for states whose population is about .32 on the diversity scale or greater, the effect of inequality on economic public liberalism is statistically significant and negative. On the other hand, when states are more racially homogeneous (less than .15 on the diversity scale), the relationship between inequality and economic public liberalism is significant and positive.
Figure 2. Conditional effect of state diversity on the effect of inequality on public economic liberalism.
Notably, state years that are .32 on the diversity scale constitute about 26% of the total number of observations in this analysis. State-years that are under .15 on the diversity scale constitute less than 33% of the total number of observations. To gain a better handle on the substantive size of this effect, I take a deeper look at specific points along the range of racial diversity. When states are the most homogeneous, a one-unit increase in the percent of income held by the Top 1% leads to about a .013 unit increase in public economic liberalism. Alternatively, when states are at the highest end of the diverse index, a one-unit increase in Top 1% share results in over a .016 unit decrease in liberalism. To get a sense of the overall impact, note that the observed range of Top 1% income share is 4% to 36%. When states are homogeneous, an increase in inequality representing a shift from the bottom of its observed range to the top would produce a .42 unit increase in public liberalism. On the other hand, when states are more racially diverse, a shift in inequality from the bottom to the top of the range would produce a nearly .51 unit decrease. These effects are sizable, given the range of public economic liberalism (−2.9 to 3.1). In terms of the effect of inequality on the nation, this imbalance should tip the scale – as states grow increasingly diverse, there is a greater likelihood for states to respond to inequality with less economic liberalism (as opposed to more). This finding can help explain the forces that may be driving Kelly and Enns’ (Reference Kelly and Enns2010) findings. Overall, these findings support my expectation that as diversity rises, the less likely it is the mass public will respond to increasing inequality with support for bigger government. In response to an increase in inequality, state publics with lower levels of diversity are more likely to want the government to do more. State publics that have higher levels of diversity are in fact more likely to want the government to do less.
The third model in Table 2 uses annual change in racial diversity (“Racial diversity
$ \Delta $
”) as an alternative indicator of diversity. The interaction term between Top 1% income share and diversity change is not statistically significant. The marginal effects plot is shown in Figure 3. Though the model indicates the relationship is not significant, on average the trend appears to show that states with increases in diversity from the year before (higher values) respond to inequality with significantly less liberalism.
Figure 3. Conditional effect of state diversity annual change on the effect of inequality on public economic liberalism.
Could it be possible that rising racial diversity and inequality are not simply affecting attitudes toward policies that target economic inequality, but are leading the public mood to shift toward conservatism overall? To ensure that it is economic liberalism, and not all forms of liberalism, that are being shaped by these two factors, I estimate the effect of inequality and diversity on public social liberalism in the fourth model of Table 3. As anticipated, the interaction term is not statistically significant.Footnote 15
Table 3. Effect of state inequality and racial diversity on public support for welfare and education spending (1975–2000)
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses using Cochrane–Orcutt Estimation. Durbin Watson tests indicate there is no serial correlation. ***
$ p<0.01 $
, **
$ p<0.05 $
, and *
$ p<0.1 $
.
Next, I test if the interactive effect of racial diversity and inequality functions similarly for preferences for welfare and education spending. In the first model, I show the results of the interactive effect of state inequality and racial diversity on public support for welfare spending. The second model is similar but uses education support as the outcome variable. Durbin-Watson tests for both models indicated to reject the null of no autocorrelation. To account for serial correlation in the error term, I estimated the same models using the Cochrane-Orcutt estimator, an iterative Generalized Least Squares estimation procedure. The output of these models can be seen in Table 3.Footnote 16
Similar to my earlier findings, the coefficient for the interaction term is negative and significant at the
$ p<0.01 $
level. In the second model, when support for education spending is the outcome variable, the coefficient for the interaction term is negative and statistically significant at the
$ p<0.1 $
level. Using the original regression models, I plot the estimated marginal effect of inequality and the 90% confidence interval over the range of racial diversity. The results are illustrated in Figure 4. The figure tells a similar story: for states whose population is racially homogeneous, the effect of inequality on welfare support is statistically significant and positive (Figure 4a). As racial diversity rises, the public’s likelihood to respond with support for welfare spending declines. I replicate the same analysis for public education spending (Figure 4b). The slope and direction are similar to those produced from other models. As predicted, the public’s preferences for education in response to inequality vary by state racial diversity. However, the solid sloped line does not cross the line at 0, suggesting there is no value of racial diversity at which the public responds to inequality with more support for education spending.
Figure 4. Conditional effect of state diversity on the effect of inequality on support for welfare and education spending.
Taken together, these results support my expectations. As inequality rises, states that are racially diverse are more likely to respond by wanting the government to do less. This is true for economic public liberalism as well as policy-specific attitudes, although the size of the effect varies across policy domains. Ultimately, though, the public’s response to inequality is conditional on racial diversity.
State diversity and perceptions of “people like us”
So far, this study has offered evidence to suggest that the effect of inequality on mass opinion has been conditioned by racial diversity over time. I now turn to gain more purchase on the theorized causal mechanism – that perceptions of other citizens as “people like me” vary by actual racial diversity. To do so, I rely on the Perceptions of the American People Survey, administered by the Ohio State University’s Center for Survey Research in the summer of 2002.Footnote 17 Subjects were selected through random-digit dialing and a total of 1,254 interviews took place. The advantage of this original survey is that respondents were asked about their perceptions of group boundaries, including if respondents felt a part of certain groups. In addition to questions concerning national identity, subjects were asked about state boundaries. This allows for an empirical assessment of state racial diversity on perceptions of others as “people like us.”
To measure state racial diversity, I include the same diversity measure used in the previous analyses, which is calculated using percent White, Black, and other. I also estimate models using new racial diversity measures based on higher-quality data on race that the Census has collected in recent years. This data categorizes people into one of four races (Asian or Pacific Islander, Black or African American, American Indian or Alaska Native, White) as well as ethnicity (Hispanic and non-Hispanic). Data was recoded so there were population estimates for one of five racial/ethnic categories: Hispanic, non-Hispanic Asian or Pacific Islander, non-Hispanic Black or African American, non-Hispanic American Indian non-Hispanic Alaska Native, and non-Hispanic White. These five groups were used to compute diversity scores using the Herfindahl index. Using this same data, I also calculated state diversity using a different categorization scheme that is more consistent with the original measure, where data was recorded so that the three racial categories are non-Hispanic Whites, non-Hispanic Blacks, and “other” (the sum of Hispanic, Asian or Pacific Islander and American Indian or Alaska Native). Overall, this leaves me with 3 measures of racial diversity, that I refer to in this section as Racial Diversity (Original), Racial Diversity (5 groups), and Racial Diversity (3 groups).
The outcome variable is subjects’ response to the following item: “How strongly do you feel part of, or identify with, people from your state?” Responses were coded from 1 to 7, where 1 means they do not feel part of the group at all and 7 means they feel very strongly part of the group. At the individual level, I include several control variables that may shape group identification (Theiss-Morse Reference Theiss-Morse2009). This includes if the subject is native to the US (0 = non-native, 1 = native), age, education (coded from 1–4 where higher values indicate more education), party identification as a Democrat or Republican (Independent was the excluded category), family income (rescaled on a 1–5 scale where higher values indicate higher income) and race (White, Black and other).Footnote 18 At the state level, I include controls for logged state population, logged mean income, and percent Democrat.
Not unlike identification with the nation, Americans tend to feel part of the people living in their state (Figure 5). But what factors predict how much of an attachment they feel to other state residents? The results of the regression analyses can be seen in Table 4. In the first model, I estimate the effect of state diversity on respondents’ perceptions of feeling a part of their state for the entire sample. According to theory, we would expect that as state diversity increases, people are less likely to perceive their states’ population as “people like me.” And indeed this is what we find. The regression coefficient for racial diversity is statistically significant (
$ p<0.01 $
) and negative, where a one-unit increase in racial diversity leads to a 1.85 decrease on the state identity scale, on average. The second and third models include alternative diversity indicators that account for ethnicity and different racial groups. Including these measures does lead to a smaller effect size of diversity on identity, though the effect is still statistically significant. To get a better sense of the size of this effect, consider that in 2002, the 5-group racial diversity index among states ranges from .06 to .65. The coefficient on the diversity term in the third model (−.92) suggests that going from the least diverse state to the most diverse results in over a .5 unit decline on the 7-point state identity scale. Overall, this suggests when states are more racially diverse, state residents are less likely to feel they are a part of the people in their state. This is the case when using the original diversity measure from the previous analyses, as well as when using diversity measures that take into account more racial and ethnic groups.
Figure 5. Respondents’ state identity strength. Higher values indicate stronger identity strength.
Table 4. The effect of state racial diversity on feeling a part of the state
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses. Data come from The Perceptions of American People Survey (2002). The Diversity (5 groups) measure is computed based on percent Hispanic, non-Hispanic Asian or Pacific Islander, non-Hispanic Black or African American, non-Hispanic American Indian or Alaska Native, and non-Hispanic White. The Diversity (3 groups) measure is computed based on the percent non-Hispanic white, non-Hispanic black, and all others. ***
$ p<0.01 $
, **
$ p<0.05 $
, and *
$ p<0.1 $
Individual-level data allows us to examine how a state’s racial makeup affects members of different racial groups. Instead of using the racial diversity scale as a measure of racial makeup, the models in Table 5 use the percent White, Black, and “other” in each state. In states that have higher proportions of Blacks and American Indians/ Alaskan Natives, whites are significantly less likely to feel a part of their state (Model 1). Blacks’ sense of state identity does not appear to be significantly affected by the percent of non-blacks in the state (Model 2). In Model 3, I estimate the same regression using respondents that identified with a racial or ethnic group other than Blacks or Whites. Notably, the small number of minority respondents in this sample leads to a substantial decline in statistical power. Respondents’ sense of state identity is only significantly affected by the percent of Black people in their state, where they are less likely to feel apart of their state as this proportion increases. However, there is no statistically significant effect for the proportion of whites in a state on state identity strength for those who are neither White nor Black.
Table 5. The effect of racial group proportions on feeling a part of the state
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses. Data come from The Perceptions of American People Survey (2002). Includes controls for state population, per capita income, and proportion democrat, as well as respondent’s age, partisanship, family income, education level, and if they are native to the US.
These results provide some suggestive evidence for the individual causal mechanism I propose is driving the relationship between inequality, diversity, and political preferences – seeing others as people like ourselves. When states are more racially diverse, state residents are less likely to feel they are a part of the people in their state. However, whites in particular appear to be the most susceptible to the effects of a non-white population. Consistent with recent scholarship (Abrajano and Hajnal Reference Abrajano and Hajnal2017), this suggests that whites may be the driving force behind shifts in aggregate opinion toward economic conservatism. However, higher quality data with over-samples of non-white groups would be beneficial in further exploring how racial diversity shapes feelings of “people like us.”
Conclusion
While scholars have recently made important strides in identifying how the American public has responded to rising inequality, the objective of this study was to offer a deeper understanding of why citizens respond in the way they do. My explanation is built on the notion that rising economic inequality can lead to anxiety and increase the propensity for social comparison. Under such conditions, people will engage in discriminatory behavior, preferring to help others that they see as more like themselves and even less likely to help others perceived as not like them. Thus, our response to inequality hinges upon if we perceive other citizens as “people like us.” While this pattern has been identified at the individual level, this article explores its macro-political implications. As inequality has risen drastically over the decades, simultaneously so has rising racial diversity, decreasing the likelihood we empathize and perceive of others as people like ourselves. Ultimately, as diversity increases, the public is less likely to respond to rising inequality by supporting policies that equalize economic resources to help the “have-nots.” This has important implications: public opinion ultimately shapes policies that either keep the status quo or help produce more inequality, contributing to a cycle of self-reinforcing inequality (Kelly Reference Kelly2020).
The findings from this study suggest the relationship between inequality and mass preferences is conditioned by subnational racial diversity. When states are more diverse, there is a weaker perception that the citizenry is composed of “people like us,” and in turn citizens on average are less willing to support equalizing policies in response to inequality. However, when states are less racially diverse, the public is more likely to share a common social identity. In turn, they are more likely to empathize with others as people like themselves and thus more likely to support government intervention to help others. The findings from this study also may suggest that diversity has this effect when measured in levels, but also may have this effect when measured as changes, consistent with recent scholarship (Hopkins Reference Hopkins2009; Reference Hopkins2011).
This research contributes to the literature in several ways. First, it demonstrates that public opinion is responsive to changing economic conditions at the subnational level, in line with recent research (Franko Reference Franko2016; Franko and Witko Reference Franko and Witko2017). However, public responsiveness to state-level inequality is not absolute. Racial diversity conditions the effect of rising inequality on redistributive policies. This finding may also shed light on why scholars have come to conflicting conclusions when studying the public’s response to inequality at different units of analysis, where rising inequality leads to demand for greater redistribution at the state level (Franko Reference Franko2016), but leads to less support for redistribution nationally (Kelly and Enns Reference Kelly and Enns2010). Inequality can be met with increasing support for redistribution under certain conditions. But as these conditions (racial homogeneity) have become increasingly rare over the past few decades, we inevitably see weaker mass support for redistribution.
Second, though these findings are supported by an influential literature suggesting social heterogeneity poses challenges to government spending (Stichnoth and Van der Straeten Reference Stichnoth and Van der Straeten2013), this study identifies the impact of subnational racial diversity on aggregate opinion toward redistribution. Thus, this article offers evidence for the mechanism usually implied but rarely tested in studies of diversity and policy output: mass preferences. Based on these analyses, public opinion is likely one of the driving factors that has led to limited government spending in the US. Additionally, although the racialization of welfare through elite framing is well-documented, this article addresses how race plays a role in shaping public opinion on a wide range of policies that effectively reduce economic inequality. Recall that despite the fact that public funding of education is significantly more popular among voters than other forms of downward redistribution (McCall Reference McCall2013), support for increased education spending has plateaued since 1990. This study offers one explanation as to why: public education, and potentially other policies, are likely to become more subject to racial considerations as the nation continues to become increasingly diverse.
Finally, following prior scholarship that finds people pick up on their racial context (Newman et al. Reference Newman, Velez, Hartman and Bankert2015; Wong Reference Wong2007), this study also poses that the public is able to perceive differences in the magnitude of racial diversity in their states and this shapes perceptions of other citizens as “people like us.” When states have higher proportions of people that are not in their own racial group, they are less likely to identify with the people of the state. However, more work still is to be done to form a deeper understanding of the extent to which these perceptions vary across geographic units, racial/ethnic identities, and what informs these perceptions over time.
The theory I present suggests that a decline in empathy is the causal link between rising inequality and aggregate opinion, consistent with the social affinity literature. However, others scholarship points to several other alternative mechanisms through which diversity affects public opinion. Proximity to other sizable racial groups can trigger feelings of threat and competition (Blumer Reference Blumer1958; Key Reference Key1949) and community diversity has been found to erode trust, social capital, and civic engagement (Hero Reference Hero2007; Putnam Reference Putnam2007). So though I identify empathy as a plausible explanation for the decline in public liberalism, future experimental and qualitative research may better evaluate and explore the causal mechanism linking diversity, inequality, and preferences.
While the analyses presented here suggest that the puzzle we see is mass preferences over time is driven by whites, this may not be the whole story. I find suggestive evidence that larger shares of blacks in state populations lead other racial and ethnic minorities to feel less a part of the people in their state. This is consistent with other work suggesting that racial minorities – not just whites – can be inclined to see other racial groups, other people who are not like them, in a negative light (Gay Reference Gay2006; Kim Reference Kim2000; McClain et al. Reference McClain, Carter, De Francesco Soto, Lyle, Grynaviski, Nunnally, Scotto, Kendrick, Lackey and Cotton2006; Wilkinson Reference Wilkinson2015). Aside from Whites, Black people have also become increasingly opposed to government redistribution (Ashok, Kuziemko, and Washington Reference Ashok, Kuziemko and Washington2015; Tate Reference Tate2010). However, public opinion scholars studying the intersection of race and redistribution have almost exclusively focused on white attitudes (Harris-Lacewell Reference Harris-Lacewell2003). Future scholarship would benefit from more deeply exploring the unique factors that shape redistributive preferences among members of various racial and ethnic groups.
Finally, though the theory I offer focuses on the implications of increasing racial diversity, there may be one component to a broader phenomenon. As the country grows increasingly diverse – on multiple dimensions – we perceive fellow others in this country as increasingly different, and in turn, are less likely to empathize with one another. Therefore, my account of Americans’ opposition to bigger government is not entirely inconsistent with Cramer’s (Reference Cramer2016) portrayal of the politics of rural America. Cramer’s ethnographic study reveals White rural residents who referred to the “undeserving” recipients of government funds were nearly always talking about urbanites or young people, not racial minorities. While it is unlikely racial considerations are always absent in such contexts, the overall sentiment proposed in this study is present: those people are not “people like us.”
Data availability statement
Replication materials are available on the SPPQ Dataverse at https://doi.org/10.15139/S3/HXUYCZ (Wager Reference Wager2024).
Acknowledgments
The author thanks Jim Stimson, Leah Christiani, Lucy Britt, Eroll Kuhn, Dave Attewell, Rachel Porter, and the State Politics Working Group at UNC-Chapel Hill for helpful comments.
Funding statement
This research was supported by National Science Foundation grant SES-1841184.
Competing interest
The author declares no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Author biography
Emily Wager is a postdoctoral fellow at Rice University.
A. Appendix
A.1. Descriptive statistics
Figure A1. Caughey and Warshaw’s state economic public liberalism scores (1970–2014).
Table A1. Descriptive statistics (1969–2014)
Table A2. Descriptive statistics (2002)
Note. State-level data is from 2002. Individual-level data come from The Perceptions of American People Survey (2002).
A.2. Alternative models predicting public liberalism
Table A3. Effect of state inequality (gini) and racial diversity on public economic liberalism
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses. Durbin Watson tests for all models indicated we cannot reject the null of no autocorrelation. ***
$ p<0.01 $
, **
$ p<0.05 $
, and *
$ p<0.1 $
.
Table A4. Inequality and economic public liberalism: ADL lag specifications
*p < 0.1, **p < 0.05, and ***p < 0.01.
Table A5. Racial diversity and economic public liberalism: ADL lag specifications
*p < 0.1, **p < 0.05, and ***p < 0.01.
A.3. Public support of education spending
Figure A2. Support for public education spending over time. Higher values indicate preference for more government spending or intervention. Data has been rescaled to 0 to 1 to facilitate interpretation. Source: General Social Survey (GSS). Survey Item: “Are we spending too much, too little, or about the right amount on improving the nation’s education system?”
A.4. Models using alternative racial diversity measures
The Census Bureau’s data collection on race has evolved substantially, increasing the number of racial and ethnic categories one can identify with and allowing for identification with more than one category. In order to use a measure that is consistent over time, I rely on data provided by the Survey of Epidemiology and End Results Program (SEER). For data access, see https://seer.cancer.gov/popdata.thru.2017/download.html. Using race estimates provided by the Census to the National Cancer Institute, SEER has calculated race estimates at the county level since 1969. This measure categorizes people into one of three groups – White, Black, and “other.” An illustration of the percent of the state population that is one of these three groups is shown in Figures A3–A5. As Hispanics can be of any race, those in the White category include those who identify as Hispanic and non-Hispanic whites. The other category comprises all of those who do not identify as White or Black. Though this measure is limited in that it does not capture ethnicity, it is useful for my purposes by allowing me to create a general indicator of diversity over a substantially long time period.
As indicated earlier, accessing high-quality data on race by state over decades is difficult. To compare estimates, I also obtained data from SEER which provides data on race and ethnicity by state from 1990 to 2018. This includes data on the four race categories specified under the 1977 standards (Asian or Pacific Islander, Black or African American, American Indian or Alaska Native, White) as well as ethnicity (Hispanic and non-Hispanic). I recoded data so there were population estimates for one of five racial/ethnic categories: Hispanic, non-Hispanic Asian or Pacific Islander, non-Hispanic Black or African American, non-Hispanic American Indian non-Hispanic Alaska Native, and non-Hispanic White. Again using the Herfindahl index, I calculated a state diversity measure using this different categorization scheme. There is a high positive correlation between the original diversity measure and this shorter diversity measure (ranging from .75 to .99 depending on the state). When data is recorded so that the three racial categories are non-Hispanic whites, non-Hispanic blacks, and “other” (the sum of Hispanic, Asian or Pacific Islander, and American Indian or Alaska Native) there is also a high positive correlation between this measure and the original measure.
To examine if using these alternative measures of diversity supports the results I find using the original, broader measure (1969–2014), I estimate the same models predicting public economic liberalism, welfare support, and education support shown in the main article. I interpret the following results cautiously given the significantly shorter time series of the data (24 years for the public economic liberalism model and only 10 years for welfare/education spending model).
The tables below replicate models in Table 2 of the manuscript using the two different diversity measures which span from 1990 to 2014, including 5-group Racial Diversity (Table A6) and 3-group Racial Diversity 3 groups (Table A7). These two tables are identical, where the only difference is the diversity measure used. For both tables, the first model is the fully specified model including an interaction term between racial diversity and inequality, as well as controls.
Table A6. Effect of state inequality and racial diversity (5 groups) on public economic liberalism (1990–2014)
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses. Durbin Watson tests for all models indicated we cannot reject the null of no autocorrelation. The Racial Diversity (5 groups) measure is computed annually based on percent Hispanic, non-Hispanic Asian or Pacific Islander, non-Hispanic Black or African American, non-Hispanic American Indian or Alaska Native, and non-Hispanic White. ***
$ p<0.01 $
, **
$ p<0.05 $
, and *
$ p<0.1 $
.
Table A7. Effect of state inequality and racial diversity (3 groups) on public economic liberalism (1990–2014)
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses. Durbin Watson tests for all models indicated we cannot reject the null of no autocorrelation. The Diversity (3 groups) measure is computed annually based on percent non-Hispanic white, non-Hispanic black, and all others. ***
$ p<0.01 $
, **
$ p<0.05 $
, and *
$ p<0.1 $
.
For Model 1 in Tables A6 and A7, the interaction term between diversity and inequality does not reach statistically significance. To better understand the interaction term, I include marginal effects plots from Model 1. These plots are shown in Figure A3. Figure A3a plots the marginal effect of inequality on liberalism across the range of diversity values calculated using 5 different racial/ethnic groups, while Figure A3b uses the alternative 3-group diversity measure. Using a 5-group diversity measure with a shorter time frame, the marginal effects plots illustrate that public opinion’s response to inequality does not vary at different levels of racial diversity.
In each table, the second model includes an interaction term between inequality and annual racial diversity change. Instead of annually, the third and fourth models use 3 and 5 year windows for diversity change. Similarly to model 1, for most of the models the interaction term between inequality and diversity change does not reach statistical significance (though comes close for some). However, the coefficient on the interaction term is negative. I include marginal effects plots for 3 year diversity windows shown in Figure A4. Figure A4a,b illustrates the marginal effect of inequality on liberalism across the range of diversity change values (where positive values indicate increases in diversity from 5 years prior and negative values indicate decreases). As anticipated, states with increases in diversity from the three years prior (higher values) respond to inequality with significantly less public economic liberalism. Simply put: positive changes in diversity are associated with negative (less liberal) responses to inequality.
Figure A3. Conditional effect of state diversity on the effect of inequality on public economic liberalism (1990–2014).
Figure A4. Conditional effect of state diversity change (5 years) on the effect of inequality on public economic liberalism (1990–2014).
I estimate the same models for welfare and education spending from Table 4 using these two alternative measures of diversity, for a total of 4 models. The data on race and the public opinion measures from Pacheco (Reference Pacheco2014) only overlap for 10 years, from 1990 to 2000. Similar to the models in the main manuscript, Durbin–Watson tests for the education models indicate to reject the null of no autocorrelation. To account for serial correlation in the error term, I estimated the same models using the Cochrane–Orcutt estimator. The output of these models can be seen in Table 3, which includes the two new diversity measures in each of the models. For the first two models, using the two diversity indicators predicting welfare support, the diversity and inequality interaction term is not statistically significant. However, we do find the significant effects in the expected direction for the education models (Models 3 and 4). What can explain this difference? It may be the case that in recent years, welfare has become so racialized in the media as benefiting minorities (Clawson and Trice Reference Clawson and Trice2000; Gilens Reference Gilens1999) that variations in subnational diversity are not as effective as shaping attitudes on policy areas that are not yet quite as racialized, such as education.
Table A8. Effect of state inequality and racial diversity on public support for welfare and education spending (1990–2000)
Note. Results include regression coefficients and standard errors in parentheses from OLS analyses using Cochrane–Orcutt Estimation. The Diversity (5 groups) measure is computed annually based on percent Hispanic, non-Hispanic Asian or Pacific Islander, non-Hispanic Black or African American, non-Hispanic American Indian or Alaska Native, and non-Hispanic White. The Diversity (3 groups) measure is computed annually based on percent non-Hispanic white, non-Hispanic black, and all others. ***
$ p<0.01 $
, **
$ p<0.05 $
, and *
$ p<0.1 $
.
A.5. Illustration of inequality, race, and public opinion by state
Figure A5. Racial diversity scores by state over time.
Figure A6. Percent white by state over time.
Figure A7. Percent black by state over time.
Figure A8. Percent “other” by state over time.
Figure A9. Income inequality by state over time.
Figure A10. Public liberalism by state over time.
Open access
