Through the ideology of the beholder: how ideology shapes perceptions of partisan groups

Abstract

Growing attitudinal and affective differences across party lines and increasing social polarization are often attributed to the strengthening of partisanship as a social identity. Scholars have paid less attention to personal preferences as a contributor to these phenomena. Our focus is on how citizens’ policy beliefs—their operational ideologies—are associated with their views of partisan groups. We examine our perspective with two studies. In the first, we find that the attribution of ideologically extreme political views to an individual's peer significantly reduces interest in interpersonal interaction but find limited evidence that partisan group membership alone induces social polarization. In the second, we show that citizens’ policy views are strongly associated with their perceptions of their own partisan group as well as their counterpartisans. Together, our results have important implications for understanding the consequences of increased polarization and partisan antipathy in contemporary politics.

Partisans in the United States are increasingly divided not just by their views of political candidates but also in how they view each other. Over the last several decades, Democrats have registered increasingly negative views of Republicans just as Republicans have expressed increasingly negative views of Democrats (e.g., Iyengar et al., 2012; Abramowitz and Webster, 2016). By and large, these trends have been characterized as affective in nature, driven by the “power of partisanship as a social identity” (Iyengar et al., 2019, 130) and where voters base their partisan loyalties more on fear and loathing of the out-party than in-group affinity (Abramowitz and Webster, 2016, 2018).

In this paper, we study how Americans view each other across party lines and consider the connection between citizens’ policy views and their views of partisan groups. Though partisan identity remains important, our perspective suggests that existing scholarship generally fails to appreciate the extent of ideology's role in shaping mass attitudes and behaviors. While studies at the elite level show that increased ideological extremism in Congress is associated with more partisan rhetoric (e.g., Gentzkow et al., 2016), less intraparty collaboration (e.g., Desmarais et al., 2015), and more distant social ties (e.g., Alduncin et al., 2017), at the mass level researchers more commonly emphasize how group attachments and identities (e.g. Iyengar et al., 2012; Iyengar and Westwood, 2015), psychological traits (e.g., Simas et al., 2020), and partisan-ideological sorting (e.g. Levendusky, 2009; Mason, 2015) produce and reinforce political and social polarization.

We argue that evaluations of partisan out-group members are driven primarily by the combination of partisanship and ideological considerations. We develop two primary theoretical expectations. First, we argue that social evaluations reflect differences in ideological positions rather than reflexive responses on the basis of partisanship. Our account suggests that partisanship alone is insufficient to generate patterns of partisan teamsmanship and social distance, a finding generally associated with social identity theory accounts of affective polarization. Second, we posit that ideological beliefs are associated with the (mis)attribution of ideological extremity to out-party members. Accordingly, we argue that individuals with stronger ideological commitments hold increasingly exaggerated perceptions of partisan out-groups and apply these perceptions when evaluating out-group members. Thus, our perspective indicates that ideological extremity magnifies perceptual differences across party lines (see also Lelkes, 2021; Fowler, 2020).

We support our argument with two analyses on the linkages between partisanship, ideology, and polarization. In the first analysis, we use an original survey experiment with a nationally representative sample of Americans to distinguish the effects of partisanship and ideology on social evaluations. We find that the attribution of ideologically extreme political views to an individual's peer significantly reduces interest in interpersonal interaction. We find limited evidence, however, that partisan group membership alone induces social polarization. In the second analysis, we build on these findings by showing that ideological extremity is strongly associated with perceptions of both partisan in-groups and out-groups. It is the most conservative Republicans and the most liberal Democrats who are most likely to misperceive the beliefs held by out-partisans. Together, our studies suggest that social polarization is largely a product of ideological disagreement rather than group conflict on the basis of partisan identity.

1 Partisanship and group evaluations

Political parties are the most salient groups for how individuals experience political phenomena (Green et al., 2002; Iyengar et al., 2012). Citizens use partisanship as a heuristic for evaluating political candidates (Conover and Feldman, 1989), forming economic assessments (Ang et al., forthcoming), and attributing responsibility for political outcomes (Malhotra and Kuo, 2008). Partisanship is also associated with nonpolitical judgments and behaviors (e.g., Iyengar and Westwood, 2015; Mason, 2015). Rather than reflecting retrospective calculations (Fiorina, 1981), the evidence collectively suggests that partisanship separates individuals into partisan teams and influences how they perceive the world (e.g., Campbell et al., 1960; Green et al., 2002; Huddy et al., 2015; Mason, 2015).

Recent research documents increasingly negative relationships between party membership and evaluations of out-party members. Iyengar et al. (2012) show that social distance between Democrats and Republicans has increased over the last several decades, as partisans increasingly dislike out-party members and ascribe negative traits to them. Hostile feelings toward out-party members can subsequently influence partisans’ willingness to exhibit behavioral discrimination against them (Iyengar and Westwood, 2015). Likewise, other scholarship finds that partisans are less likely to engage in commercial activity with (McConnell et al., 2018; Engelhardt and Utych, 2020), work with (Panagopoulos et al., 2020), and date (Huber and Malhotra, 2017) people who do not share their partisan identity.

Scholars have explained these phenomena by conceptualizing partisanship as an expressive social identity that generates affective reactions toward out-party members (e.g., Iyengar et al., 2012; Huddy et al., 2015; Iyengar and Westwood, 2015; Mason, 2015). Applying social identity theory to party membership, this perspective conceives of partisanship as “a social and psychological attachment” (Mason, 2015, 129). As Huddy et al. (2015) elaborate, partisanship provides “a subjective sense of belonging to a group” in which party members seek to positively differentiate their party from the other. Because “the mere act of identifying with a political party is sufficient to trigger negative evaluations of the opposition” (Iyengar et al., 2012, 407), this perspective provides an identity-based account for affective party polarization documented in the research described above.

1.1 Distinguishing the effect of partisanship on group evaluations

Tests of identity-based accounts of affective polarization confront important inferential challenges. The primary challenge is that the nonrandom assignment of partisanship to individuals¹ makes it difficult to distinguish the effects of partisanship from other political characteristics that may also affect evaluations. For instance, membership in a political party is generally correlated with higher levels of political interest and activity (Campbell et al., 1960; Klar, 2014). Moreover, Klar et al. (2018) demonstrate that measures of antipathy toward out-party members conflates negative partisan affect with dislike of partisan politics more generally. Partisanship is also associated with other group-based characteristics such as race, religion, and class. Evaluations of partisans could also (at least partially) reflect attitudes toward these groups (Abramowitz and Webster, 2018; Ahler and Sood, 2018).

In contemporary American politics, partisanship is highly correlated with issue preferences, but with two prominent exceptions most studies of affective polarization and social distance do not distinguish the effect of party membership from the effects of ideological differences. Orr and Huber (2020) address this issue with survey experiments that ask respondents to evaluate a hypothetical individual who is randomly attributed with some combination of partisan cues, social cues, and policy views. While they conclude that “a great deal of measured partisan animosity reflects disagreement about contentious issues” rather raw partisan identity (Orr and Huber, 2020, 584), the design of the experiments does not permit them to estimate the effect of partisanship and issue positions on social evaluations relative to the absence of political information.² In an experiment involving a hypothetical candidate, Lelkes (2021) varied the presence of partisan cues and summary ideological information before eliciting respondents’ evaluation of the candidate. The results showed that evaluations were considerably more responsive to ideological information than they were to partisan cues. However, survey respondents harbor more animus toward party elites than they do toward rank-and-file party members (Druckman and Levendusky, 2019), and thus it is not clear whether the results from the candidate experiment apply to evaluations of ordinary voters.

We argue that inferences about the effect of party membership on animosity toward partisan out-groups outpace the available evidence. Instead, persuasive evidence of partisan animosity should demonstrate that individuals provide systematically worse evaluations of members of partisan out-groups than they do in the absence of information about members’ partisan affiliations. Our perspective is analogous to research on the role of political parties in legislative voting behavior. In this literature, scholars have argued for evaluating party effects based on how legislative outcomes would differ in the absence of political parties (see, e.g., Krehbiel, 1993) and by studying the extent to which voters support co-partisans without ideological congruence on policy issues (see, e.g., Fowler, 2020). While recent research makes important contributions to understanding how partisan cues and policy positions flow through one another in producing affective polarization (Dias and Lelkes, forthcoming), this scholarship does not show how party membership and ideological beliefs affect interpersonal evaluations relative to the absence of either political cue.

1.2 How ideology shapes interpersonal evaluations

We address the limitations of existing scholarship noted above and argue that perceptions of partisan outgroups are driven largely by ideological differences rather than by group membership alone. We contend that perceptions of partisan groups are rooted in perceived or actual differences in ideological beliefs. That is, individuals reach more positive assessments of their peers on the basis of their political agreement: we feel more warmly about individuals with whom we agree than with whom we disagree. Social polarization can thus result when individuals meet others with differing political views or perceive that others have differing political views. Because partisan membership is associated with policy views (e.g., Green et al., 2002; Gerber et al., 2010; Homola, 2021), we expect that individuals provide more negative interpersonal assessments of individuals who belong to partisan outgroups, as prior research on partisan social identity argues (e.g., Huddy et al., 2015; Mason, 2015).

We further expect that evaluations of out-party members are particularly negative for outgroup members who hold political views that are increasingly ideologically opposed to one's own. Prior work has shown that partisans perceive greater attitudinal extremity among members of the opposite party (Judd and Park, 1993; Levendusky and Malhotra, 2016), and other scholarship shows that higher levels of ideological extremity among the out-party are associated with affective polarization (Rogowski and Sutherland, 2016; Webster and Abramowitz, 2017). While information about ideological extremism may trigger negative interpersonal evaluations in the absence of explicit information about party membership, as Orr and Huber (2020) show, we argue that individuals view out-party members especially negatively when those individuals are attributed with extreme ideological beliefs. Our account posits that individuals view out-party members negatively when they hold, or are perceived to hold, views with which they disagree rather than simply on the basis of their party membership. This is consistent with evidence presented by Druckman et al. (forthcoming), who show that most Americans view members of the other party with indifference rather than hostility, and that hostility is expressed mostly by politically engaged members of a party when asked to evaluate ideologically extreme members of the out-party.

To the extent that party membership is associated with interpersonal evaluations, we expect that these patterns are driven largely by perceptions of ideological disagreement. We argue that partisans’ exaggeration of the opposing party's extreme views should be most pronounced among those who are more ideologically extreme. Ahler and Sood (2018) show that citizens tend to view out-party members as more extreme than they are, and strong ideologues are likely to assume that others’ beliefs are similarly extreme and thus project their own beliefs onto others (see, e.g., Van Boven et al., 2012). In a context where conservatives have sorted into the Republican Party and liberals have sorted into the Democratic Party (Levendusky, 2009; Mason, 2015), therefore, ideological extremism magnifies perceptions of partisan group differences.

We report evidence from two studies to evaluate our account. The first study reports evidence to distinguish the effect of party membership and ideology on interpersonal evaluations. In contrast with expectations offered from previous studies (e.g., Iyengar et al., 2012; Huddy et al., 2015; Mason, 2015; Dias and Lelkes, forthcoming), we test the hypothesis that party group membership affects interpersonal evaluations when it is accompanied by ideologically extreme beliefs but on its own has minimal effects on social polarization. The second study evaluates how an individual's ideological beliefs are associated with perceptions of partisan outgroups. Extending previous work (Ahler and Sood, 2018), we test the hypothesis that more ideologically extreme partisans hold more exaggerated perceptions of members of the opposite party. Across both, evidence in favor of these hypotheses would support our argument that ideological beliefs, instead of and/or in conjuction with party membership, contribute more to social polarization than previous research has recognized.

2 Experimental evidence on social polarization

We test our argument and distinguish the effect of partisanship and ideology on social polarization with a survey experiment. Our data come from the December 2016 wave of The American Panel Survey (TAPS), a monthly online panel survey from a national probability sample.³ Similar to the design used in related studies (e.g., Orr and Huber, 2020), respondents were randomly assigned to receive one of three vignettes which described the characteristics of a hypothetical new neighbor. (Complete vignette wording is shown in Table A.2.) These descriptors contained information about the person's age and education, and indicated that the person was from a suburb outside a larger Midwestern city. We chose this information so that it would not provide explicit cues about partisan affiliation.

The control group (N = 491) received additional non-political information about the new neighbor regarding pets and hobbies. They were also told that the neighbor frequently volunteers for local candidates’ campaigns, indicating a degree of political activism. By including this information, we ensure that comparisons between the control and treatment groups do not conflate dislike for politics or political activism in general (see, e.g., Druckman et al., forthcoming; Klar et al., 2018) with dislike for specific partisan groups.

Respondents in the partisan treatment condition (N = 497) received the same background information about the neighbor. They were also informed that the neighbor was registered as a voter for the opposing party as their own.⁴ The comparison between the control and partisan conditions allows us to compare evaluations of a neighbor who is described as affiliating with the opposite party to the counterfactual that respondents have no partisan information about the same neighbor. If out-party membership alone increases social polarization, we expect to observe more negative evaluations of the neighbor among both Republican and Democratic respondents relative to the control condition.

Respondents in the partisanship + ideology condition (N = 499) received the same information as respondents in the partisan condition. They were also told that the potential neighbor subscribes to ideological viewpoints that are stereotypically associated with the neighbor's party. These ideological perspectives were conveyed through specific policy preferences and more general worldviews. Democratic neighbors were characterized as believing that everyone should drive an electric car, that marijuana should be legal in all states, and that a nationalized healthcare system would improve the health of all citizens. Republican neighbors were described as believing that humans and dinosaurs walked the earth at the same time, that elementary school students should be required to recite the pledge of allegiance every morning, and that a fence should be built between the US and Mexico. If individuals automatically attribute extreme issue positions to members of the out-party upon learning their partisanship, we would expect that the treatment effects of the partisanship + ideology condition are similar in magnitude to the treatment effects of the partisan condition. On the other hand, if the attribution of ideological views is the primary mechanism through which social polarization is generated, we would expect the treatment effects of the partisan condition to be considerably smaller in magnitude than the treatment effects of the partisanship + ideology condition.

Our outcome variables measure respondents’ social evaluations of the hypothetical neighbor and elicit respondents’ interest in ostensibly non-political interactions. After receiving the vignettes, we asked respondents whether they would be interested in befriending, hiring, dating, inviting the person over for a meal, and allowing their children to play together.⁵ Each question was measured on a five-point scale range from “strongly disagree” to “strongly agree.” For simplicity, we recoded each to a binary measure that indicated whether respondents “agreed” or “strongly agreed” that they would be interested in interacting with the neighbor. Finally, all analyses use weights benchmarked to national population parameters based on the Current Population Survey (CPS) at the time of the interviews.

The design of our experiment differs in several key respects from experiments used in related scholarship. First, following guidance from Druckman and Levendusky (2019), we measure interpersonal evaluations with questions about respondents’ willingness to interact with ordinary people rather than feeling thermometer ratings (Iyengar et al., 2012) which may capture views toward elites. Second, our primary interest is in identifying whether out-party membership is sufficient to induce social polarization, or whether affective reactions depend on the combination of out-party membership and ideological differences. Because it is not our goal to distinguish the effect of ideological views absent partisanship (contra Dias and Lelkes, forthcoming and Orr and Huber 2020), we do not include a condition that provides ideological positions without information about party membership. Third, because we are interested in how partisanship on its own contributes to affective polarization, we include a pure control condition while Dias and Lelkes (forthcoming) and Orr and Huber (2020) do not. Finally, the partisanship + ideology condition attributes relatively extreme ideological beliefs to the neighbor. This choice contrasts with other research that has signaled ideology through more standard issue positions or more abstract ideological summaries (Dias and Lelkes, forthcoming; Orr and Huber, 2020; Lelkes, 2021). We believe the choice is justified given that partisans tend to exaggerate the extremism of out-party members (Ahler and Sood, 2018, study 2 below). If party membership is sufficient to activate these exaggerated perceptions, we would expect similar interpersonal evaluations for respondents in the partisan and partisanship + ideology conditions. Yet if out-party membership alone does not prime respondents to attribute extreme positions to the neighbor in the vignette, as our account posits, we would expect to observe more negative evaluations from respondents in the latter condition.

2.1 Results

Figure 1 shows results for each dependent variable.⁶ The plotted points indicate the difference between the control condition compared with the partisan (shown with the triangle) and partisanship + ideology (shown with the circle) conditions for the outcome variables shown along the y-axis. The x-axis represents the difference in the mean share of respondents who expressed interest in interacting with the potential neighbor that, by design, is from the opposing party of the respondent. The vertical line at zero indicates the null hypothesis of no treatment effect; positive values along the x-axis indicate that larger proportions of treatment group respondents than control respondents expressed interest in interacting with the potential neighbor, while negative values indicate that smaller proportions of treatment group respondents than control respondents expressed interest in interacting with the potential neighbor.

Fig. 1. Effect of partisanship and ideology on interpersonal evaluations. Note: Values along the x-axis indicate the difference in proportions when comparing each of the treatment groups to the control group. The vertical line at zero indicates the null hypothesis of no treatment effect. The horizontal lines show the 95 percent confidence intervals.

As the figure shows, we find mixed evidence that the partisan condition significantly increased social polarization. For two of the dependent variables—hiring the neighbor as an employee and being friends with them—the effects are negative and statistically significant, indicating a six or seven percentage point decrease in social activity. The estimates for the three other dependent variables are not statistically distinguishable from zero. Notably, the effect on interest in dating—one of the most common indicators previous scholarship has used to assess social distance (e.g., Huber and Malhotra, 2017)—is estimated as 0.00.

Figure 1 also shows statistically significant and substantively large effects of the partisanship + ideology condition on interpersonal interaction. Compared with the control group, respondents in this treatment group were substantially less interested in interacting with the hypothetical potential neighbor in all five scenarios. These differences range from − 9 percentage points (allowing children to play together) to − 22 percentage points (consider hiring as employee). Importantly, the effects of the partisanship + ideology condition are also considerably larger than and statistically distinguishable from the effects of the partisan condition for each outcome variable. To the extent that partisan differences contribute to social polarization, they do so mostly when paired with information about an individual's ideological worldview.

We conducted several additional analyses which we report in the Appendix. First, the patterns in Figure 1 generally apply to both Democratic and Republican respondents. The treatment effects of the partisanship + ideology condition were negative and statistically significant for all five dependent variables among Democratic respondents and for four of the five dependent variables among Republican respondents. We did find, however, that the negative treatment effects are somewhat smaller in magnitude among Republicans than Democrats. Second, the effects of the partisan condition were moderated somewhat by respondents’ ideological extremity and partisan identity strength (see Figures A.1–A.3 and Tables A.5–A.6). Ideologically extreme respondents and respondents with stronger partisan identities reacted more negatively to the partisan condition than moderate respondents, for whom all the results are null.⁷

The experimental findings support our claim that differences in ideology, rather than party membership alone, are the primary driver of out-party hostility. We examine this proposed mechanism and study how respondents used the information in the vignettes to make inferences about the neighbor's ideological position. We asked respondents to place the neighbor on a five-point ideological scale that ranged from very liberal (1) to very conservative (5). We expect respondents in the partisan and the partisanship + ideology conditions to provide more ideologically extreme assessments of the neighbor relative to respondents in the control condition. For Republican respondents, this would manifest in a more liberal rating of the neighbor, and for Democratic respondents this would manifest in a more conservative rating of the neighbor. Given our account, we further expect that respondents in the partisanship + ideology condition would evaluate their neighbor as more extreme than either the control or partisan conditions.

Figure 2 shows the results. Negative numbers along the x-axis indicate that respondents provided more liberal evaluations relative to the control group and positive numbers indicate that respondents provided more conservative evaluations relative to the control group. The vertical line at zero again indicates the null hypothesis of no difference in evaluations relative to the control group. The results for Republican respondents (evaluating Democratic neighbors) are shown with the dark circles and the results for Democratic respondents (evaluating Republican neighbors) are shown with dark triangles.

Fig. 2. Effect of partisanship and ideology on evaluations of ideology. Note: Values along the x-axis indicate the treatment effects of the conditions shown on the y-axis relative to the control group. Negative numbers indicate more liberal placements and positive numbers indicate more conservative placements. The horizontal lines show the 95 percent confidence intervals.

Republican and Democratic respondents in the control condition provided generally similar ideological evaluations of the potential neighbor. On the five-point scale, Democrats placed the neighbor at 3.1 and Republicans placed the neighbor at 3.4. Absent specific partisan or ideological information about the neighbor, respondents therefore placed the individual referenced in the vignettes around the center of the ideological scale.

The top panel shows the results for the partisan condition. Overall, Republican respondents placed the Democratic neighbor almost a full point in the more liberal direction. The results for placements by Democratic respondents were about one-third the magnitude, indicating that the partisan condition increased their evaluation of the neighbor's conservatism by 0.31 points on the five-point scale. Absent any policy or other ideological information, therefore, we find that partisan affiliation polarizes how partisan respondents evaluated the ideological orientation of the neighbor referenced in our vignettes.

The bottom panel shows the results for the partisanship + ideology condition. For both Republican and Democratic respondents, we find that this treatment significantly increased the perceived extremity of the potential neighbor. And for both parties, the magnitude of the effect is roughly 50 percent larger relative to the effect of the partisan condition. Among both Republican and Democratic respondents, therefore, partisanship, particularly when coupled with ideological information, significantly affected their evaluation of the extremism of the potential neighbor's ideology.

Our experimental findings provide new evidence about how partisanship and ideology affect interpersonal evaluations. In contrast to claims about the effect of party membership on affective polarization, partisanship has relatively small, and mostly statistically insignificant, effects on social polarization. In other words, absent the presence of ideological cues, partisanship itself has a relatively limited effect on social interaction (contra Orr and Huber, 2020). Consistent with our argument, the experimental results indicate that the invocation of ideological characteristics significantly increases social distance, at least when these views are paired with information about party membership. Overall, these findings suggest that previous research might have overstated the contributions of partisanship to social polarization and indicate that differences in ideological beliefs are more substantial contributors to interpersonal evaluations.

3 How partisans view the other

The experiment in the preceding section shows how the invocation of issue positions causes individuals to create social distance between themselves and out-party members. In a second analysis, we study the sources of perceptions of partisan outgroups. In particular, we focus on how an individual's own ideological beliefs are associated with perceptions of ideological extremism among supporters of the opposing party. To the extent some individuals hold systematically more extreme perceptions of members of the opposing party, the predictors of these perceptions rather than partisanship per se may help explain partisan affective polarization documented in previous survey research.

We used data from the March 2014 and May 2014 waves on TAPS, on which we fielded a battery of questions to measure perceptions of partisan groups. Our sample includes 1301 panelists who completed both waves. This subset of panelists, which forms our sample for this analysis, is demographically representative of the US population. Summary statistics are shown in Tables B.1 and B.2.

We asked ten questions to measure partisan beliefs in the March 2014 wave. For each party, our questions were intended to tap into perceptions of lifestyles and cultural values in addition to policy attitudes. Five statements corresponded to perspectives that may be typically associated with Republicans and included topics such as creationism, gun ownership, and nationalism (e.g., “We should build a fence between the United States and Mexico.”). The other five statements aimed to capture perspectives more likely to be associated with Democrats including topics such as paternalism, tax increases, and eco-friendliness (e.g., “This country would be better if every citizen drove an electric car.”). The complete question wordings are in Table B.3.

Though these statements are not exhaustive, they were inspired by messages from political actors that attributed specific qualities to their opponents and media descriptions of prominent partisans. For example, Donald Trump made the construction of a wall between the US and Mexico a prominent component of his 2016 presidential campaign. Similarly, while president, Barack Obama not only made it his goal to put one million all-electric cars on the road by 2015 but also promised to buy one himself after his presidency.⁸ We used responses to these statements, which we term first-order beliefs, to calculate the proportions of Democrats, Republicans, and Independents that agreed with each.

Table 1 presents respondents’ first-order beliefs and shows the percentages of Democrats, Independents, and Republicans that agreed with each item. The top panel shows rates of agreement with the items intended to assess typical perceptions of Republicans and the bottom panel shows rates of agreement with the items expected to elicit commonly-held Democratic perceptions. Each item generally performed as expected, with greater rates of agreement with the Republican items among self-identified Republicans. For example, 42.5 percent of all Republican respondents agreed with the statement that “this country would be safer if every law-abiding citizen possessed a firearm,” while agreement was much lower among Independents (25.1 percent) and Democrats (14.2 percent). Similarly, for the five Democratic items we observe the highest levels of agreement among self-identified Democrats. For example, 38.7 percent of Democrats agreed that “this country would be better if every citizen drove an electric car,” while smaller percentages of Independents (22 percent) and Republicans (11.4 percent) agreed. Furthermore, for nine of the ten questions Independents agreed with each statement at rates in between those for Democrats and Republicans.

Table 1. Descriptive statistics: first-order beliefs

Note: The agreement columns report the percentage of Democrats, Independents, and Republicans who indicated agreement with the respective item. The second to last column reports the absolute difference in mean level of agreement between Democrats and Republicans. Bold differences indicate statistical significance at the 95 percent confidence level.

Columns 4 and 5 of Table 1 present results of statistical tests of differences in agreement between Democrats and Republicans. Column 4 presents the absolute difference in percentage points, and column 5 reports the accompanying t-statistics for differences in means tests. The differences in agreement between Democrats and Republicans are statistically significant for nine of the ten items. The only statement that does not exhibit significant differences is: “The federal government should impose a ban on a sale of soda.”⁹ Moreover, for all of our items, the differences are of substantively important magnitudes. Excluding the soda ban item, the differences range from 9.3 percentage points on the question regarding humans and dinosaurs to a striking 47 percentage points when asked about nationalized health care. Given the variation in rates of agreement among partisans and the differences between parties, the first-order beliefs reported in Table 1 provide good leverage for studying perceptions of partisan groups.

The May 2014 wave asked panelists to indicate the percentages of Democrats/Republicans they perceived to agree with each of the statements introduced in the March wave. Methodologically, the two-month lag between asking respondents about their own beliefs and their perceptions of partisan groups’ beliefs helps to minimize potential biases from anchoring or reference effects. Respondents represented their perceptions of partisan groups’ beliefs by choosing from five response categories that corresponded to their perceived rates of agreement with each statement: 0–20, 21–40, 41–60, 61–80, and 81–100 percent. We coded these responses on a five-point scale scale, where larger numbers represent a respondent's perception that greater proportions of the partisan group share that belief or attitude. For example, when Democratic panelists were asked to indicate what percentage of Republicans they believe agreed with the statement “Elementary students should be required to recite the pledge of allegiance every day,” a larger number indicates that they perceived substantial agreement with that statement among Republicans. This measure of second-order beliefs characterizes how strongly respondents attributed these perceptions to partisan groups.¹⁰ These second-order beliefs are our primary dependent variable.

A potential limitation is that biased perceptions of the outgroup could be caused by a desire to express ideological extremity of the opposite party. Additionally, our strategy cannot directly address the causes of variation in the independent variable, and a key threat to inference is whether there is some third factor that causes both own (apparent) extremity and (apparent) views about the extremity of the other party. Yet there are reasons to believe the survey responses are sincere and not endogenous (see also Berinsky, 2018). By asking about multiple extreme beliefs, we observe a tremendous amount of variation as, for example, ideologically extreme respondents report holding some, but not all, of the partisan misperceptions. This variation suggests sincerity on the part of the respondents; homogeneity in extremity would suggest more affective (perhaps signaling) behavior.

4 Ideology and partisan perceptions

Do Republicans [Democrats] overestimate the prevalence of Democrats [Republicans] who hold extreme beliefs? Our data indicate that outpartisans overestimate the beliefs of their counterparts with high frequency. For all but one item (the legalization of marijuana), the modal Republican overestimated the true percentage of Democrats who agree with the issue. This overestimation is most pronounced when considering the attitudes about taxes. Only 19 percent of Democrats agreed with raising everyone's taxes, while more than 80 percent of Republicans believed the percentage to be higher than that figure. In fact, the modal Republican responded that over 80 percent of Democrats agreed with the statement. Democrats followed a similar pattern: the modal Democrat overestimated the “true” percentage for four of the five items.¹¹ Thus, we see that both Republicans and Democrats overestimate the true percentage of their counterpartisans who hold an extreme belief.¹²

4.1 Partisan (mis)perceptions and ideological extremity

We now test our prediction that an individual's own ideological beliefs are associated with increased exaggerations in perceptions of the opposite party. Specifically, we expect that those who are themselves ideologically extreme will be most likely to exaggerate the extremity of out-party supporters. We characterize ideological extremity with a measure of ideology based on respondents’ agreement with a series of 13 policy statements known to load highly on a single ideological dimension (Claassen et al., 2015).¹³ The scores have a mean of zero (SD = 1); because higher values of this measure represent more conservative preferences and lower values indicate liberal preferences, we refer to this measure as Conservatism.

We regress each respondent's second-order beliefs of out-party members on their ideological leanings using ordinary least squares. For respondents identifying as partisans, we focus on the five items that relate to the opposite party. The unit of analysis is respondent i's reported perception of out-party views on question j. For example, for the statement “this country would be safer if every law-abiding citizen possessed a firearm,”$\text {\emph {Perception}}_{ij}$ is included for every Democrat in our sample and takes on values between 1 and 5, where higher values indicate they believe a larger share of Republicans agreed with the statement. Our model is as follows:

$$\text{\emph{Perception}}_{ij} = \beta_1 + \beta_2 \text{\emph{Conservatism}}_{i} + {\bf \gamma} {\bf X}_{i} + \delta_j + \varepsilon_{ij},\;$$

where $\text {\emph {Conservatism}}_{i}$ is our measure of operational ideology for respondent i.¹⁴ To address potential question-specific variation in perceptions, we include indicator variables for each item, represented by δ_j. Additionally, γX_i captures the effect of a series of control variables, including political knowledge and political interest.¹⁵ Respondents with high political knowledge and interest may hold exaggerated second-order beliefs about members of the out-party because they are more familiar with the opinions and beliefs of out-party politicians and supporters due to their exposure to elite rhetoric. We also include a broad set of socio-demographic controls.¹⁶ We cluster standard errors by respondent.

The results are presented in Table 2, where column 1 shows Republicans’ views of Democrats and column 2 explains Democrats’ views of Republicans. Consistent with our expectations, the coefficient estimate for Conservatism is statistically significant in both models. The positive coefficient estimate in column 1 indicates that more conservative Republicans perceived higher rates of agreement among Democrats with the Democratic items, while the negative coefficient estimate in column 2 suggests that more liberal Democrats perceived that larger percentages of Republicans subscribed to the Republican items.

Table 2. Partisans’ views of out-party members

Note: Table entries are linear regression coefficients with standard errors in parentheses, clustered on individuals. The outcome variable is $\text {\emph {Perception}}_{ij}$, the degree to which a respondent i believes members of the out-party agreed with item j as described in the text. Additional socio-demographic controls (income, education, sex, race, age, Metropolitan Statistical Area, and US Census region) and question indicators are included but not reported. *p < 0.05.

The third column of Table 2 includes both Democrats’ and Republicans’ views of the out-party. Because Conservatism is standardized at 0, we use its absolute value as a measure of Ideological Extremity. High values of this variable indicate respondents that are very liberal or very conservative, while lower values indicate respondents with more ideologically moderate views. The positive and significant coefficient estimate shows that more ideologically extreme partisans perceived higher levels of agreement with items among their respective out-party.¹⁷

The estimated coefficients for the control variables are also of substantive interest. More politically interested and sophisticated respondents tended to believe that a larger share of members of the other party agreed with that party's items.¹⁸ Though the exact level of statistical significance varies across these estimates, the results generally suggest that greater attentiveness to politics and public affairs among partisans is associated with attributing these beliefs to members of the out-party at greater rates.¹⁹

Our results are robust to a number of different model specifications and measurement choices. We estimated models that included an indicator for whether a respondent is a “strong partisan” to explore whether our results are driven by strength of party identification rather than ideological extremity (Table B.9). We also substituted a traditional seven-point symbolic ideology variable for our measure of ideology (Tables B.10 and B.11). To disentangle ideological extremity from one's ideological identity, we controlled for both symbolic and operational ideology in the same models (Tables B.12 and B.13). We estimated models in which we accounted for the partisan's own position on each statement to explore the possibility that respondents may hold exaggerated perceptions in ways that varied systematically with their position on that statement (Table B.14). We also measured party identification in which leaners (people thinking of themselves as closer to one of the parties in a follow-up question) are classified as partisans rather than as Independents (Table B.15). Finally, we created an indicator for responses whose partisan identities were aligned with their ideologies to explore whether our results are driven by partisan-ideological sorting (Table B.16).²⁰ We also estimated models that included the different controls for strong partisanship, symbolic ideology, and partisan ideological sorting all at the same time (Table B.17). These models produce consistently strong evidence that ideological extremity is associated with exaggerated perceptions of partisan groups.

In addition, because our outcome measure is a five-category variable, we replicated all analyses using ordered logit models (Table B.18).²¹ And, while our analyses pool all perception items together, we estimated a series of models that focused on one individual statement at a time (Tables B.19 and B.20). Moreover, in December 2016 we also asked respondents to indicate their answers on a 100-point scale (Tables B.21 and B.22). These robustness checks also support the findings above.

Figure 3 illustrates the substantive relationship between ideology and partisan perceptions and shows predicted levels of our outcome variable $\text {\emph {Perception}}_{ij}$ across the range of values of operational ideology. Based on Models 1 and 2 in Table 2, Figure 3 displays the differences in perceptions of members of the out-party.²² The y-axis corresponds to the categorical outcome measure, where a value of “2” indicates a belief that between 21 and 40 percent of out-partisans agree with a given item and “4” indicates a belief that between 61 and 80 percent of out-partisans agree with a given item. The dashed line is the predicted perception among Democrats toward Republicans and the solid line is the predicted perception among Republicans toward Democrats. The shaded regions represent the 95 percent confidence intervals. Figure 3 suggests that the most conservative Republicans and the most liberal Democrats perceive that members of the opposite party agree with a respective item a full category higher than their more moderate co-partisans, all else equal. That is, given two Republicans who are identical in all respects, with the exception of one being a far-right conservative and the other being an ideological moderate, our model predicts that the more conservative individual will believe about 20 percent more Democrats agree with a given Democratic item.

Fig. 3. Predicted second-order beliefs for partisans. Note: Values along the x-axis indicate respondents’ operational ideology, values along the y-axis the predicted values of second-order beliefs. The shaded areas show 95 percent confidence intervals.

We also examined whether respondents identified the correct proportion of out-party members who agreed with each survey item. These results are shown in the Appendix in Table C.1. For each item, we created an indicator for whether a respondent correctly identified the percentage category that contained the observed level of agreement with the partisan item.²³ We find that the likelihood of identifying the correct proportion of agreement among partisan out-groups is significantly related to the respondent's own ideology, where more ideologically extreme views are associated with less accurate perceptions of the other party.

The results from this analysis support our argument that perceptions of partisan out-groups are substantially influenced by ideology. Among both Democrats and Republicans, ideological extremity is associated with greater inaccuracy and increased exaggeration in perceptions of the out-group. Consistent with our theoretical perspective and the experimental results discussed above, we find that ideology plays an important role in explaining exaggerated perceptions.

5 Conclusion

Several generations ago, responsible party theorists emphasized the centrality of political parties to modern democracy (e.g., Schattschneider, 1942). Today, however, membership in and identification with political parties is often attributed with a host of social ills, including support for undemocratic behavior (Graham and Svolik, 2020) and increased antipathy toward members of the opposite party (e.g., Iyengar et al., 2012; Iyengar and Westwood, 2015; Mason, 2015; Huddy et al., 2015; Dias and Lelkes, forthcoming). On this latter score, our results suggest that previous research has overstated the consequences of partisanship for affective polarization. Instead, partisans are most likely to harbor negative views of out-party members when those individuals hold, or are believed to hold, ideological views that are out of step with their own. In contrast with research that argues affective polarization is caused primarily by partisan identity (Dias and Lelkes, forthcoming), our findings indicate that social distance is generated largely by individuals’ ideological commitments. Our analyses suggest that accounting for policy views in the contemporary electorate will allow researchers to more fully understand the dynamics that characterize mass politics.

Our findings contribute new insights about the potential consequences of contemporary partisan polarization for mass political behavior. In particular, our results suggest that increased partisan polarization may be associated with the use of increasingly exaggerated partisan caricatures. As the public forms perceptions of partisan groups at increased rates and with decreased accuracy, the “pictures in our heads” may in fact make it more difficult for citizens to relate to one another across the partisan aisle. These patterns of results suggest a potential explanation for why increased polarization at the elite level has led to increased social polarization among partisans in the mass public. They also suggest an important link between increased ideological extremity and decreased affect toward political out-groups.

By design, our research has some important limitations. First, our survey experiment asked respondents to evaluate the potential for interpersonal interaction based on the information they received in a hypothetical setting. We are less certain how the experimental results generalize into real-world settings in which, for instance, social norms could serve as countervailing influences on the application of partisan perceptions. Second, we considered a small number of items that may be salient for forming perceptions of partisans. Third, our TAPS data mostly represent a single snapshot in time and limit our ability to make stronger conclusions about the causal relationship between partisan polarization and perceptions. It is unclear whether the nature of partisan perceptions has intensified in the contemporary era relative to a generation or two ago. Fourth, the design of our studies does not permit us to conduct convincing mediation analyses to evaluate the mechanisms implicated by our theory. More systematic research, both experimental and observational, is necessary to more definitively answer these questions.

At the same time, our research raises several important questions about the nature of partisan perceptions and their implications. Perhaps most importantly, future research should interrogate the possibility of correcting exaggerated (mis)perceptions about partisans. For instance, Ahler (2014) shows that correcting exaggerated perceptions of ideological polarization reduced respondents’ self-reported levels of ideological extremism. To the degree that partisan perceptions inhibit social interaction across partisan lines, efforts to increase the accuracy of partisan perceptions may therefore prove fruitful for achieving greater levels of empathy, reduced affective polarization, and more consensual political outcomes. Finally, our research does not address temporal dynamics in the formation and deployment of partisan stereotypes, though we suspect that they emerge, evolve, and are displaced over time as the parties themselves change (e.g., Green, Palmquist, and Schickler 2002). These questions are all important for future research.