The Origins and Consequences of Racialized Schemas about U.S. Parties

Abstract Two parallel processes structure American politics in the current moment: partisan polarization and the increasing linkage between racial attitudes and issue preferences of all sorts. We develop a novel theory that roots these two trends in historical changes in party coalitions. Changing racial compositions of the two major parties led to the formation of racialized images about Democrats and Republicans in people’s minds—and these images now structure Americans’ partisan loyalties and policy preferences. We test this theory in three empirical studies. First, using the American National Election Studies we trace the growing racial gap in party coalitions as well as the increasing overlap between racial and partisan affect. Then, in two original survey studies we directly measure race–party schemas and explore their political consequences. We demonstrate that race–party schemas are linked to partisan affect and issue preferences—with clear implications for the recent developments in U.S. politics.

By the turn of the 21st century, partisan polarization had profoundly altered the political landscape in the United States (Fiorina and Abrams 2008;Hetherington 2009;Layman, Carsey, and Horowitz 2006). Parties' positions in Congress have polarized significantly over the last decades (McCarty, Poole, and Rosenthal 2006). At the same time, there is less agreement on whether the divergence on policy preferences across party lines has been happening within the mass electorate (Abramowitz and Saunders 2008;Fiorina Abrams, and Pope 2006). However, the affective divide across party lines is unambiguous: Democrats and Republicans have increasingly disliked each other since the 1970s until the present (Iyengar, Sood, and Lelkes 2012).
Polarization between Democrats and Republicans coincided with partisan sorting on the basis of racial identity and racial attitudes: beginning in the 1970s, Democratic candidates in presidential elections started to attract large shares on nonwhite voters whereas Republicans increasingly relied on votes of racially conservative whites (Abramowitz and Webster 2018). Over the same period, voters' positions on seemingly non-racial political issues have gradually become more intertwined with racial resentment (Enders and Scott 2019). These developments are often attributed to the reactionary politics among whites that emerged as a response, first, to the civil rights movement and, then, to growing racial diversity in society at large (Parker and Barreto 2015). The current intensity of racial and ethnic politics in the United States is also often linked to the polarizing candidacies of Barack Obama and Donald Trump (Buyuker et al. 2021;Sides, Tesler, and Vavreck 2018;Tesler and Sears 2010).
This paper advances a novel argument with respect to the nexus of race, partisanship, and polarization in the United States. Specifically, we suspect the images of American parties in the average citizen's mind have become racialized but still vary considerably. A combination of partisan realignment on the basis of race and objective demographic changes have created a large racial gap between the two partisan coalitions: the Democratic Party is increasingly racially diverse, whereas Republicans have remained a party of whites. Drawing upon the group-schematic approach to political cognition (Conover 1988), we predict that changes in the racial compositions of the two major parties catalyzed racialization of mental images of Democrats and Republicans among Americans. In the United States, the two major parties are primarily responsible for bundling political issues into ideological packages that are, then, offered to citizens in the electoral marketplace (Bawn et al. 2012). Therefore, the racialization of political issues, even ones that do not explicitly target racial groups, may occur as the result of racialized images of the parties themselves.
Our approach, informed by the schematic model, echoes the existing accounts of partisan sorting and polarization in the United States (Mason 2018). Still, our contribution differs in some important aspects. Real racial sorting has indeed been an important ingredient of partisan realignment in the United States over the last three or more decades. However, the original sorting argument does not consider individual-level variation in the degree to which people imagine party coalitions to be racialized. Instead, it implicitly assumes that the images of parties are highly racially distinct and constant across society, at least at any given moment. We suspect that the mental images of parties can vary substantially (Ahler and Sood 2018;Rothschild et al. 2019), and this variation can have political consequences above and beyond the importance of real-world sorting. In other words, the paper brings together sociological and psychological accounts of partisan sorting in order to predict which voters will feel most strongly about the other party and, perhaps as a result, will adopt more extreme positions on issues.
Importantly, we are not the first to show that Americans' racial affect now spills over into partisan animosity (Westwood and Peterson 2020). However, the exact reasons for this relationship are not yet clear. Is it driven by issue preferences that push citizens into different parties, and then coincidentally lead racial affect to be correlated with partisan attitudes? Or, perhaps, did the increasing salience of race during the presidencies of Barack Obama and Donald Trump lead racial affect to be much more important in evaluating the parties?
We propose a third explanation. The Democratic Party and the Republican Party are now perceived as racially distinct-a perception that has changed incrementally with actual racial sorting in the party coalitions. But variation in these perceptions is still substantial and highly consequential, regardless of the actual level of racial sorting at any given time. To test whether variation in the perceived racial coalitions of each party is indeed at work, we first trace the growing overlap between racial and partisan affect from the 1980s to the present. We also develop original measures of the mental connections between racial groups and political parties and show that these measures are non-trivially related not just to partisan affect but to positions on non-racial issues as well-something that has not been demonstrated previously. Overall, our results show that American parties are increasingly seen as distinct racial and ethnic camps rather than institutions for delivering unique policy bundles, and this has major implications for understanding current political processes in the United States.
We test our argument in three interconnected studies. Using time-series data from the American National Election Studies (ANES), we first document objective changes in racial compositions of the two major parties. The growth in racial diversity within the Democratic Party has outpaced demographic changes in the nation-while very little has changed in the Republican Party's coalition. In other words, the relative racial distinctiveness of the two parties has grown substantially, and it has been reflected in the growing association between racial and partisan affect. For instance, we find that warmth toward blacks has become increasingly linked to positive feelings toward Democrats and negative feelings toward Republicans, whereas the changes in the link between feelings toward whites and the parties have been the opposite.
However, with data from the ANES we cannot directly measure mental images of parties. Instead, we must simply infer them from the observed relationships between race and party affect. To address this limitation, we next run two original survey studies. We start by measuring racialized perceptions about the two major parties using an implicit technique and find that white respondents who think of the Democratic Party as black and of the Republican Party as white feel colder toward the Democrats and warmer toward the Republicans. The choice in favor of the implicit measurement strategy was driven the expectation that respondents who think of the Democratic Party as black might underreport these perceptions in order to avoid appearing racially insensitive. Given that racialized rhetoric seems to have become more acceptable, however, we replicate these findings using an explicit measure of racialized party images as well. In this second survey study, we also show that white respondents who perceive the "typical Democrat" as black hold more conservative views on several political issues.

Race in U.S. politics: current accounts
The phenomenon of racialization-whereby racial affect increasingly predicts voters' positions on ostensibly race-neutral issues-seems currently to be on the rise in U.S. politics. Racial attitudes are now important predictors of opinions about electoral fairness (Appleby and Federico 2018), gun control (Filindra and Kaplan 2016), policing (Jefferson, Neuner, and Pasek 2021), international trade (Mutz, Mansfield, and Kim 2021), and healthcare (Tesler 2012).
Two perspectives have emerged to explain this increasing prominence of racial divides in contemporary U.S. politics. The first concerns demographic changes in American society, most importantly the increase in racial and ethnic diversity. Whites' decreasing share of the total U.S. population, and especially the prospect of losing their majority status, represent threats to many white Americans (Myers and Levy 2018). Perceived threats to their group status also mobilize conservative whites' support for policies aimed at maintaining the status quo, such as opposition to income redistribution (Abrajano and Hajnal 2015). As a result, the political importance of racial identity among whites has increased over the last two decades (Jardina 2019).
The second explanation emphasizes the personalities and political styles of Barack Obama and Donald Trump, and their special role in bringing race to the forefront of American politics. Optimists hoped the election of Obama, the nation's first African American president, would improve race relations in the United States. Instead, it seemed to further polarize the country by priming racial resentment in evaluations of almost any political cause Obama championed (Tesler 2016). Racial resentment thus came to influence not only evaluations of Obama himself (Wilson and Davis 2018), but also those of many other Democratic candidates on the national level during the Obama presidency (Luttig and Motta 2017).
Similar effects apparently played a role in the political rise of Donald Trump. According to a popular argument, Trump's campaign successfully weaponized perceptions of racial status threat, group identity, and racial resentment among whites (Abramowitz and McCoy 2019). Some even characterize Trump's political platform as "white protectionism" (Smith and King 2021). The appeal to racially conservative whites clearly played an important role in Trump's success in 2016 (Reny, Collingwood, and Valenzuela 2019). This strategy also seemed to benefit down-ballot Republican candidates, and the effect persisted through at least the 2018 midterms (Algara and Hale 2020).

The schematic model: racialized parties
Our argument draws upon these findings but offers a novel perspective on the relationship between race and polarization in American politics by bringing together the pieces described above: changes in party coalitions, racial attitudes, and partisan identification. We start by acknowledging objective changes in the American partisan coalitions caused by several factors. First, the partisan realignment caused by the civil rights movement led African Americans to become one of the most reliable segments of the Democratic Party's coalition, and simultaneously pushed formerly Democratic, racially conservative whites to the Republican Party (Valentino and Sears 2005). Second, the growing Latino population in the 1980s and 1990s caused a defection of non-Hispanic whites out of the Democratic Party (Hajnal and Rivera 2014). Third, ethnic and racial minority groups in the United States tend to be lower in socioeconomic status and therefore more attractive targets for the Democratic Party even beyond the appeal of racial identity (Zingher 2018). As a result, the Democratic Party coalition has become increasingly racially diverse, whereas the Republican Party has remained almost entirely white. Importantly, however, these changes influence, but do not precisely dictate, how racialized each party's coalition is seen by members of the public. It is this variation in the racialization of imagined party coalitions that, we think, ultimately structures political attitudes and identities.
To theorize about the impact of racial changes in partisan coalitions on the public's subjective images of the political parties, we invoke the schematic model of political cognition (Conover and Feldman 1984). Social schemas are a class of cognitive structures-organized assemblages of knowledge about people and social situations-that are stored in memory (Moskowitz 2005). Schemas decrease the cognitive effort required to process what is often a flood of complex social information present in the environment. This simplifying function of schemas makes them particularly relevant for the study of cognition in politics. As shortcuts, schemas assist voters in making up their minds about political issues, parties, and candidates. Among the nearly infinite variety of politics schemas, ones relevant to race and ethnicity might be particularly important, due to the strong emotional attachments that individuals have for these groups. Generally, voters tend to support candidates and policies that promote the interests of the ethnic ingroup, often at the expense of outgroups. In other words, people are generally ethnocentric in their politics (Kinder and Kam 2009).
Schemas are created out of information springing from various sources, both direct (e.g., personal experience) and indirect (e.g., partisan media). We suspect that as the racial compositions of the U.S. partisan coalitions changed, this information diffused through society via personal experiences and mass media representations, gradually altering citizens' partisan schemas. We argue, therefore, that the images of political parties have become strongly linked to race in the minds of Americans over the few last decades. Throughout the paper, we refer to these linkages as the raceparty schemas. The modal mental images of both parties have changed over time, such that nowadays Democrats are not simply viewed as liberal, but as nonwhite. Republicans, on the other hand, are viewed primarily as a party of whites. As generations socialized before the racial realignment pass away, images of the Democratic Party as nonwhite and the Republican Party as white are becoming more common. It is the variation in these imagined coalitions that matters deeply for how people feel about the parties.
Since stereotypes like these often develop at an early age, the racialization of party schemas probably occurs early in life, as party identity itself is forming (Sears and Valentino 1997). In other words, partisan and racial identities do not emerge separately, only to be linked in adulthood. Instead, the link is probably forged during the process of partisan socialization, so that group schemas are automatically called up whenever party is salient. This mechanism predicts a smooth, secular shift over time in race-party schemas despite the fact that party platforms on race shifted rather abruptly during the 1960s (Carmines and Stimson 1989).
Our theory has several observable implications that are tested in the empirical analyses to follow. The racial composition of the U.S. parties over the last decades changed so that the share of nonwhites significantly increased among Democrats but remained small and mostly constant among Republicans. If our theory is correct, this growing demographic racial sorting should lead racial attitudes to become increasingly intertwined with partisan affect over the same period. We would also suspect many American voters at present should imagine the parties as racially distinct: the Democrats will be stereotyped as nonwhite and The Journal of Race, Ethnicity, and Politics Republicans as white. Finally, these imagined party coalitions are important, and variable. The variation in these mental images, or race-party schemas, should be closely associated with both issue positions and partisan affect.
Study 1: the racialization of party images over time To test our first hypotheses about the racialization of party images over time, we employed time-series data from the ANES. We documented changes in (a) racial compositions of partisan coalitions after the World War II and (b) the overlap between racial attitude and partisan affect during the last three decades. In the ANES, respondents' race is recorded by interviewers. For this analysis, we defined partisans as those who openly support one of the two major parties (i.e., leaners were not treated as partisans). To measure partisan affect, we relied on feeling thermometers, where respondents were asked to rate their warmth toward each major party using 100-point scales. To measure respondents' racial attitudes, we used two measures: (a) feeling thermometers toward blacks and whites and (b) the four-item version of the racial resentment scale (Kinder and Sanders 1996). In order to account for common criticisms of the racial resentment scale (Sniderman and Tetlock 1986), we controlled its effect for political ideology. For question wordings, see Supplementary Material. We used full ANES samples but weighted them to be representative of the U.S. adult population for each year.

Results
We first document the profound demographic changes in U.S. partisan coalitions from 1952 to the present. Figure  whites and blacks among self-identified Democrats and Republicans, as well as among all eligible voters. Table 1 presents linear trends in racial compositions of the two major parties and the general electorate estimated as the effects of time on the shares of whites and blacks from simple ordinary least squares (OLS) regressions. Effects presented in the first three rows are estimated as (marginal) within-group time coefficient in one model. Effects presented in the last two rows are coefficients from an interactive model.
Here and in all models to follow, time is normalized to a scale from zero (the start of the analyzed period) to one (the end of the analyzed period). The first row simply documents changes in the racial demographics of population as a whole: the white share of the American adult citizenry has declined by more than 26 percentage points, while the share that is African American has grown by 4 percentage points between 1952 and 2016. The second row reveals that whites' proportion of the Democratic Party coalition has declined by a whopping 43 percentage points over the last 64 years. The share of Democratic identifiers that are African American, in turn, has increased by almost 18 percentage points. The third row shows that racial demographic changes in the Republican Party have been smaller but still significant, with about 11 percentage points fewer whites and 3.5 percentage points fewer blacks identifying with the party. The demographic changes in both parties significantly differed from trends in the population as a whole, as shown in the 4th and 5th rows of Table 1. Whites, for example, declined as a share of the Democratic Party nearly 17 percentage points more than their decline in the population at large. African Americans' share of the Democratic Party increased 13 percentage points more than their change in the Note. Entries in the top three rows are regression coefficients for the changes in the proportions of the category on the left that falls into each racial category in the column. The coefficients in the last two rows represent differences in how the racial proportions within each party are changing compared to the population as a whole. Standard errors in parentheses. Sample size is the number of analyzed election years. *p < .05. **p < .01. ***p < .001.
share of all eligible voters. In sum, a large gap in relative racial compositions of the two partisan coalitions has emerged over time. Almost nonexistent in late 1950s and early 1960s, the racial sorting of nonwhites into the Democratic Party and whites into the Republican Party grew substantially by the early 2000s. A central question for this paper is how consistently and accurately these realworld shifts have become cemented in the minds of average citizens. The changes have been slow, and individuals may or may not be aware of their extent. The ANES cannot answer this question directly, since respondents are not asked their perceptions about racial group coalitions in each party. We do, however, have indirect evidence. If citizens view the parties in racialized ways, there should be a growing association between affect about racial groups and affect about parties.
To test this conjecture, we estimate the relationship between respondents' feeling thermometer scores for blacks and whites on the one hand and the feeling thermometer scores toward the Democratic Party and the Republican Party on the other. The associations are estimated as simple OLS regression coefficients by year, from 1980 to the present, in Figure 2. A line of best fit for each group is also included. The pattern is obviously quite stark. The associations between group and party feeling thermometers were rather small and similar to each other at the beginning of the time series. In 1980, feelings about both blacks and whites were positively related to feelings about the Democratic Party. Over time, these associations diverged, so that by 2016 feelings about the Democratic Party and feelings about African Americans were highly positively correlated, while liking whites was highly negatively associated with liking Democrats. A nearly identical pattern, in mirror opposite, appears for the association of racial group feelings and −.5 0 .5 1 9 8 0 1 9 8 4 1 9 8 8 1 9 9 2 1 9 9 6 2 0 0 0 2 0 0 4 2 0 0 8 2 0 1 2 2 0 1 6 1 9 8 0 1 9 8 4 1 9 8 8 1 9 9 2 1 9 9 6 2 0 0 0 2 0 0 4 2 0 0 8 2 0 1 2 2 0 1 6 Democratic FT Republican FT Black FT effect: Year estimate Trend White FT effect: Year estimate Trend Estimated effect Year Figure 2. Relationships between racial affect and partisan affect by year, 1980-2016. Simple OLS coefficient estimates with 95% confidence intervals evaluations of the Republican Party. Importantly, the overall pattern is one of consistent, secular change. Neither Obama's election, nor Trump's, caused a significant inflection in these trends. Since the patterns in the bivariate results over time approximate a linear trend, Table 2 presents OLS models that estimate the rate of change with the two thermometers, time, and interactions of the thermometers with time as the only predictors of Democratic and Republican affect. The coefficients in the first and third rows represent the impact of the black and white thermometers, respectively, in 1980, since that was the first year of the time series. Interaction terms, presented in the second and fourth rows, capture the rate of change in associations over the period of 36 years. By adding the baseline effect in 1980 and the interaction effect, one can estimate the relationship between racial and partisan affect in 2016. The coefficient for time represents estimated change in party affect for a respondent who scores zero on both black and white feeling thermometers.
Regression estimates show that moving from lowest to highest on the white feeling thermometer resulted in a 10-degree increase in warmth toward the Democrats in 1980. By 2016, the effect sign flips: the same shift in 2016 led to a 10-degree decrease in the Democratic thermometer. For the Republican Party, in 1980 a move from low to high on the white feeling thermometer resulted in a 15-degree increase in warmth. By the end of the analyzed period, the effect almost doubles: same shift on the white thermometer led to a 28-degree increase in warmth for the Republicans. The corresponding effects of the black feeling thermometer show the opposite pattern. Warm feelings toward African Americans were positively and significantly linked to feelings toward Democrats in 1980, and this effect slightly (but not significantly) increased by 2016. The magnitude of the change over time is even larger for the relationship between feelings toward African Americans and the Republican Party: from almost zero effect in 1980 in changed to significantly negative in 2016. These results are replicated in Table 3 using a different measure of racial attitude, the racial resentment scale, and controlling for ideology. Estimates are from OLS regressions that include racial resentment, ideology, time, and interactions of racial resentment and ideology with time as predictors of Democratic and Republican affect. Both racial resentment and ideology here are recoded to have the same 0-100 scale as the feeling thermometers. Entries should be interpreted similarly to Table 2: direct effects are estimated effects in 1980, whereas time interactions are estimated changes in effects from 1980 to 2016. Time effect is the estimated change in party affect for someone who is least racially resentful and most ideologically liberal.
We find that even in the 1980s, racially resentful respondents felt more warmly toward Republicans and cooler toward Democrats. By 2010s, these effects approximately doubled for both Democrats and Republicans. Note that these models include controls for ideology and its interaction with time, so these observed changes in the association between racial resentment and partisan affect are probably not simply derivative of changes in party ideologies.

Discussion
Study 1 has documented the quite profound changes in racial demographic compositions of the U.S. partisan coalitions from 1952 to 2016. The Democratic Party has been steadily becoming more diverse, outpacing racial diversification of the American adult population overall. The Republican Party, on the other hand, remains largely a party of whites, objectively and relative to the population. The racial gap in partisanship was greatest in 2012, when whites constituted 90.8% of self-identified Republicans and only 51.4% of Democrats. These data also demonstrate that affect toward racial groups has become much more tightly linked to partisan attitudes over time. On average, the magnitude of the association between racial attitudes and partisan feelings almost doubled between the 1980s and 2010s. This pattern is consistent with the cognitive mechanism we suspect is at play-an increasingly strong race-party schema in the minds of many in the American electorate. While these results are consistent with our general theory, the ANES does not include questions that would allow us to directly test the causal mechanism of interest: variation in perceptions about the racialization of each party at the individual level. To overcome this limitation, we carry out two original survey studies where we measure the degree to which respondents associate race and U.S. parties.

Study 2: the implicit racialization of party images
To directly measure variation in the perceived racialization of U.S. parties, we developed a sorting task based on the technique used in the implicit association test (IAT; Greenwald, McGhee, and Schwartz 1998). The IAT is a computer-based task that uses objective response times to assess the direction and strength of automatic memory associations between categories and attributes. Respondents quickly sort stimuli into category pairs constructed either pro-stereotypically (flower/pleasant, insect/unpleasant) or counter-stereotypically (flower/unpleasant, insect/pleasant). Differences in response times are used to measure the strength of a pro-stereotypical association relative to the counter-stereotypical one.
The original IAT design was used to measure implicit affect toward blacks and whites in the United Sates. However, the IAT architecture can help tap implicit group schemas in politics: automatic cognitive linkages between social groups and political attitude objects, such as parties or policy target populations. Our modification uses stimuli for social groups (African American and Caucasian faces) and political categories (recognizable party symbols) but relies on the same IAT response metrics. See detailed description of the IAT procedure in Supplementary Material.
Using a similar IAT-based task, we measured the automatic association between racial images and party symbols. To signify race in the sorting task, we used African American and Caucasian faces from the standard "race IAT" (Nosek et al. 2007). To represent the parties, we collected publicly available images representing both official and unofficial symbols of the Democratic Party and the Republican Party such as logos, campaign buttons, and posters. The IAT was administered using Inquisit software. The survey also included questions on party identification, partisan feeling thermometers, and racial resentment. See Supplementary Material for the IAT stimuli and exact question wordings We collected a sample of respondents using Amazon Mechanical Turk (MTurk) in May-June 2015. MTurk samples are not representative of the U.S. population, but they are more diverse than many other convenience samples, such as college undergraduates (Berinsky, Huber, and Lenz 2012). The analyzed sample included only respondents who answered the survey from the United States and had unique IP addresses as well as low error rates in the IAT task. This left us with 396 valid cases out of 523 completed interviews. In the final sample, 80.8% of respondents were white, 6.3% were African American, and 12.9% identified with some other ethnic or racial group. Among respondents who identified as neither African American nor white, approximately equal shares were Hispanic (42.0% of the "other" group) and Asian (51.0%). The sample was balanced in terms of gender (49.8% female), but overrepresented the highly educated, with nearly half holding Bachelor's degrees or higher (45.2%). Mean age was 37.3 years. The sample was heavy on Democrats, at 47.0%, while 18.4% identified as Republicans, and 34.6% as independents.

Results
We first examine differences in automatic race-party schemas by race and partisanship. Race-party schemas are measured using IAT D-scores: negative scores indicate an automatic association of Republicans with blacks and Democrats with whites. Positive scores indicate an automatic cognitive association of Democrats with blacks and Republicans with whites.
Mean race-party schemas (as indicated by IAT D-scores) by party are presented in Table 4 separately for three groups: whites, blacks, and all others. This analytic strategy is informed by our theoretical expectations. The race-schematic model of party affect and polarization proposed in this paper should work for members of all groups, not just whites. Recall that the IAT D-scores measure the race-party schemas unidirectionally: positive scores mean the respondent is faster in pro- Note. Entries are mean D-scores by partisanship and race. D-scores measure the relative speed of associating the two parties and the two racial categories. Positive D-score correspond to faster associations of the Democratic Party with blacks and the Republican Party with whites than vice versa. Negative D-scores correspond to faster associations of the Democratic Party with whites and the Republican Party with blacks than vice versa. Standard errors in parentheses. *p < .05. **p < .01. ***p < .001.
stereotypical pairings (associating the Democratic Party with blacks and the Republican Party with whites) than in counter-stereotypical pairings (associating the Democratic Party with whites and the Republican Party with blacks). Negative scores, respectively, mean the respondent is relatively faster in counterstereotypical pairings (associating the Democratic Party with whites and the Republican Party with blacks). For the purpose of this comparison, those who "lean" toward one party are treated as partisans whereas pure independents are excluded. On average, white Democrats and white Republicans are faster to associate their in-party with whites and the out-party with blacks than vice versa, and the mean difference in implicit race-party schemas between white partisans is statistically significant.
Still, the negative D-score for white Democrats seems puzzling. Do white Democrats really believe Republicans are black? We do not think this is what the results indicate. First, one must remember that the D-score is a relatively measure: a negative score simply means an individual is faster, on average, at identifying the white-Democratic and black-Republican pairing than at the black-Democratic and white-Republican pairs. This could happen when a respondent holds a strong mental association between the Democratic Party and whites (rather than blacks) while not having strong racial schemas about the Republican Party. 1 We suspect this is happening with white Democrats: they are more likely to associate their racial ingroup with their party without necessarily strongly associating the Republican Party with African Americans. This is, therefore, a form of identity projection: respondents tend to associate the in-party with the racial ingroup more strongly than they hold other mental pairings, even stereotype-consistent ones. Indeed, black Republicans also seem to demonstrate a similar projection effectassociating the in-party with their racial group-but it is not significant, most likely due to sample size. The main takeaway from Table 4 is that there exists substantial variation in the race-party schemas in the population. Some see the Democratic Party as a party of African Americans, but others (mostly white Democrats) hold less racially stereotypic images of the party coalition.
We next calculate bivariate relationships between the implicit race-party schemas and several political outcomes of interest: partisanship (on the standard 7-point scale from 1 = Strong Democrat to 7 = Strong Republican), feeling thermometers toward Democrats and Republicans, and racial resentment. Remember that we need to examine these associations separately for each group: if our theory is correct, whites with positive D-scores should feel warmer toward Republicans and colder toward Democrats, because they believe Republicans represent their racial ingroup. For African Americans with positive D-scores, these associations should be negative-for the same reason. For members of other groups, mostly Asians and Latinos, there are no clear predictions. For example, if an Asian person believes the Democratic Party is dominated by blacks and the Republican Party by whites, it is not clear how feelings about their ingroup would predict feelings about the parties.
Our prediction is that variation in race-party schemas, at least for blacks and whites, would be associated with each of these outcomes. The results by racial group are presented in Figure 3 as coefficients from simple OLS regressions. For this 1 Results do not change when controlling for demographics. See Figure S1 in Supplementary Material analysis, all variables are normalized to a 0-1 scale. Among whites, all relationships are significant and in the predicted directions: white respondents who imagine the Democratic Party as black and the Republican Party as white are more likely to identify as Republican, feel cooler toward Democrats and warmer toward Republicans, and score more highly on racial resentment. The small sample size for African Americans renders these associations statistically insignificant due to low statistical power and high standard errors. At the same time, the point estimates suggest that the relationships between race-party schemas and political attitudes among blacks are opposite to those among whites and about the same size, as our theory would predict: African American respondents who imagine Democrats as black are less likely to identify as Republican, feel warmer toward Democrats and cooler toward Republicans, and score lower on racial resentment. Among respondents who belong to other ethnic groups we find a mixed pattern of results, as we expected we would, with none reaching statistical significance. The main takeaway is that while there is statistical uncertainty here for African Americans and others, the overall pattern is consistent with our expectations: whites who identify their in-party as white like that party more and the out-party less, while blacks who identify their in-party as black like that party more and the out-party less. Members of other groups show no consistent pattern.  Figure 3. Estimated relationships of implicit race-party schemas with partisanship, party affect, and racial resentment. Note. Implicit race-party schemas measured with IAT D-scores, the relative speed of associations between the two parties and the two racial categories. Greater D-score = faster association of the Democratic Party with blacks and the Republican Party with whites. Bivariate coefficients estimated independently for each covariate using simple linear regressions. All variables normalized to a range from 0 to 1 In Study 2, we have estimated the associations between implicit race-party schemas and several political outcomes of interest. The analyses have returned important results: first, we have found substantial variation in implicit race-party schemas across the population. Second, among white respondents, this variation in raceparty schemas is associated with partisanship, party affect, and racial resentment in the predicted direction. Samples of nonwhite respondents are not enough to establish any significant relationships, but point estimates of effects among African Americans are in line with our theory. In Study 3, we replicate and extend these results using a different (explicit) measure of race-party schemas and a broader range of dependent variables, such as specific issue positions-including not explicitly racialized issues like environmentalism.

Study 3: the explicit racialization of party images
We collected a second sample using the MTurk platform in February 2016. We removed those with duplicate IP addresses as well as those who completed the survey from outside of the United States, yielding the final sample of 466 respondents out of 520 submitted questionnaires. Demographics, again, deviated from national parameters in expected ways. Specifically, our respondents were predominantly males (57.3%), with college degrees (53.2%). The sample was also relatively young, with a mean age of 36.3 years. As much as 76.2% of the sample were white, 7.9% were African American, and 15.9% identify with some other racial group. Among respondents who identified as neither black nor white, the largest groups were Asians (56.8% of the "other" group) and Latinos (28.4%). In terms of partisanship, 49.1% of respondents were Democrats, 19.0% were Republicans, and 31.9% were independents. Explicit measures of the perceived racial compositions of the two major parties were constructed by asking respondents to identify the "the typical supporter" of both the Democratic Party and the Republican Party in terms of race. Specifically, respondents were asked to choose a single ethnic or racial category: white, African American, Hispanic, Asian, Native American, and Native Hawaiian/Pacific Islander. In addition to question related to schemas, we included partisan feeling thermometers and a set of political issues: size of government, defense spending, environmentalism, and abortion. See Supplementary Material for exact question wordings. We were concerned about potential social desirability bias in questions about race and party, but if this bias were strong enough, the test would return null results. 2 2 Unfortunately, we cannot test this explanation directly. Negative D-scores could of course be the result of strong mental associations between the Republican Party and blacks (rather than whites) while not having a strong schema about the Democratic Party. Substantively, that would be "negative projection," or imposing the racial outgroup identity onto the out-party. We suspect this explanation is less plausible, especially in the light of findings from Study 3 demonstrating that explicit schemas of the Republican Party as white are uniform in the sample.

Results
We first document the frequency of racialized partisan schemas in the sample. The percentages of respondents who choose each specific racial group as typical for Democrats and Republicans are presented in Table 5. Two thirds of the sample view the typical Democrat as white, which is of course correct. The typical Democrat is thus seen as nonwhite by approximately one third of respondents, whereas almost everyone in the sample thinks of a typical Republican as white. Overall, more racial and ethnic diversity is seen among Democrats compared to Republicans. It is also worth noting that African Americans are much more commonly associated with the Democratic Party than any other nonwhite group.
Next, we estimate the relationship between the explicit race-party schemas and the outcome variables of interest: partisan affect and issue positions. Since we have found almost no variance in racial schemas about the Republican Party, we focus exclusively on the impact of variation in racialized schemas of the Democratic Party in Figure 4. 1 There we present differences estimated using simple OLS regressions between respondents who see the Democratic Party as white versus those who perceive it as black, in terms of their partisan affect and issue positions. Variables are normalized to a scale from zero to one. The results show that white respondents who think of the typical member of the Democratic Party as black express relatively more negative feelings toward Democrats and more positive feelings toward Republicans. They also take more conservative positions on a host of issues: the size of government, defense spending, and environment-but not on abortion, for reasons we could only speculate about.
None of the estimated effects of race-party schemas among African Americans are not statistically significant-most likely, due to a small sample size and the resulting lack of statistical power. However, some of the coefficients carry the predicted sign: for instance, African American respondents who see the typical Democrat as black feel warmer toward the Democratic Party. Effects of the race-party schemas among respondents who identify with other ethnic and racial groups are in the same direction as ones among whites, and some of them are statistically significant: specifically, respondents who are neither white nor black and see the typical Democrat as an African American feel warmer toward the Republican Party and prefer higher defense spending. Note. Entries are percentages stating that each racial group on the left is the typical member of each party in the columns.

Discussion
Study 3 provides an additional piece of the puzzle that, overall, is favorable to our argument with regard to the role of race-party schemas in American politics. First, a non-trivial number of respondents perceive the Democratic Party as African American, so there is substantial variation in the racial schemas for that party. This measure returns relatively little variation in the imagined typical Republican: almost the entire sample perceives that person to be white. Second, variation in explicit racial schemas among whites is related to partisan affect and issue positions in the expected direction. Specifically, white respondents who perceive the Democratic Party as African American are less favorable toward Democrats, more favorable toward Republicans, and take more conservative positions on political issues.

Conclusion
In this paper, we have offered a novel perspective on the pervasiveness of race and racial attitudes in U.S. partisan politics that seems to be only growing with time.
Drawing upon recent contributions, we have developed a theory linking objective changes in racial compositions of partisan coalitions, racialization of party images in the minds of voters, and political polarization. Using the group-schematic model of political cognition, we have argued that the growing racial gap between the  Figure 4. Estimated bivariate relationships of explicit race-party schemas with partisanship, party affect, and political issue positions. Note. Explicit race-party schemas measured with self-reported associations between the Democratic Party and the two racial categories. Score of 0 = respondent perceives the typical Democrat to be white. Score of 1 = respondent perceives the typical Democrat to be black. Bivariate coefficients estimated independently for each covariate using simple linear regressions. All variables normalized to a range from 0 to 1 Democratic and Republican support bases leads to formation of racialized stereotypes about the two parties. Specifically, a non-trivial share of American electorate currently views the Democratic Party as nonwhite and the Republican Party as white, though in reality whites continue to be a majority of both parties. We label these images race-party schemas and hypothesize that variation in the racialized images of the two major parties contributes to the general racialization of issues and partisan attitudes in contemporary U.S. politics.
Real changes in racial demographics of American society have surely contributed to the racialization of the party images. However, our findings also suggest that the imagined racial coalition of each party varies considerably across the population, and that variation matters. This finding carries profound implications for the ongoing discussion in the discipline about affective polarization in American politics: whites feel colder toward the Democratic Party when they imagine its coalition to be more heavily made up on nonwhites and feel warmer toward the Republican Party when they perceive it to be dominated by their racial group. 3 As a consequence, rather than a cause, they may then come to accept a more conservative issue package advocated by the modern Republican Party. This speculation that issue polarization is the downstream consequence of race-party schemas requires much more careful study.
In addition, our findings suggest that the emphasis on de-facto party leaders, including such transformative ones as Barack Obama and Donald Trump, in explaining the recent surge of racial politics might be premature. The schemas of the two parties have been racialized over time beginning at least in the 1980sand, may be, even earlier. Therefore, Obama and Trump could not have caused the observed racialization of parties, even though their electoral victories marked historic watersheds in the activation of the race-party schemas in American politics. In other words, Obama and Trump did not create the coalitions that helped them to win-instead, they were elected as seeming embodiments of the new Democratic and Republican voting blocs.
Our contribution also opens up possibilities for future research. First, it remains unclear which racial schemas are particularly important for the images of the U.S. parties, and the Democratic Party in particular. We focus mostly on African Americans as the stereotypically Democratic racial group in part due to data availability-feeling thermometer questions about blacks and whites have been asked in the ANES long enough for our time-series analysis. And our findings reported in Study 3 confirm that African Americans remain the racial group most often named by those who perceived Democrats as nonwhite. However, recent changes in the national political discourse, such as increasing emphasis on immigration, might have also increased the salience of Latinos as an important part of the Democratic coalition (Ostfeld 2019).
Second, even though our theory is formulated in general terms, we were unable to meaningfully explore the content and political consequences of race-party schemas among nonwhite Americans due to sample limitations. Specifically, due to small samples of nonwhite respondents-a common problem for crowdsourcing survey platforms-we do not have enough statistical power to make strong inferences. Future studies will need to recruit oversamples of nonwhite groups to be sure these patterns hold.
Finally, our findings have a limitation related to the fact that the sample in Study 2 underrepresents Republicans-that is, again, common for platforms like MTurk. As a result, we may miss some respondents with the most conservative views and, potentially, the most pronounced pro-stereotypical race-party schemas. If present, these sample imbalances should lead us to underestimate the variance of our key variable of interest as well as its covariates. Therefore, estimates reported in the paper likely reflect a lower bound of the corresponding population parameters.
In comparative politics, the ethnicization of party systems has been long associated with political polarization and democratic instability (Rabushka and Shepsle 1973). An influential interpretation for these effects is rooted in the social identity theory: as soon as ethnic parties start to compete for political power, winningrather than implementing a certain policy-becomes the goal in and of itself due to associated boost in group status and self-esteem of its members (Horowitz 1985). Moreover, comparative evidence suggests that U.S. plurality-based electoral system contributes to politicization of ethnic cleavages rather than mitigates them (Huber 2012). Therefore, the racialization of American parties is likely to continue, and the intensity of political conflict in the United States is likely to grow.