Who Dislikes Whom? Affective Polarization between Pairs of Parties in Western Democracies

Abstract

While dislike of opposing parties, that is, affective polarization, is a defining feature of contemporary politics, research on this topic largely centers on the United States. We introduce an approach that analyzes affective polarization between pairs of parties, bridging the US two-party system and multiparty systems in other democracies. Analyzing survey data from twenty Western democracies since the mid-1990s, first, we show that partisans' dislike of out-parties is linked to elite policy disagreements on economic issues and, increasingly over time, also to cultural issues. Secondly, we argue and empirically demonstrate that governing coalition partners in parliamentary democracies display much warmer feelings toward each other than we would expect based on elite policy (dis)agreements. Third, we show that radical right parties are disliked much more intensely than we would expect based on policy disputes and coalition arrangements. These findings highlight the policy-based and institutional underpinnings of affective polarization.

Concerns over citizens' contempt for partisan opponents have attracted increased attention in Western democracies. US partisans are increasingly likely to avoid interactions with members of the other party (Lelkes 2016). In the Netherlands, people express greater dislike toward out-partisan than other social outgroups (Harteveld 2021a; Harteveld 2021b). In Sweden, the rise of the radical right has been followed by increased interparty resentment, which now exceeds the intensity of partisan animosity in the United States (Reiljan and Ryan 2021).

Research about hostility across party lines, that is, affective polarization, has until recently focused on the United States (Druckman and Levendusky 2019; Iyengar et al. 2019). Yet, a growing comparative affective polarization literature has emerged (Adams et al. 2022; Bassan-Nygate and Weiss 2022; Boxell, Genzkow, and Shapiro 2020; Gidron, Adams, and Horne 2020; Harteveld 2021a, Harteveld 2021b; Hernandez, Anduiza, and Rico 2021; Hobolt, Leeper, and Tilley 2020; Horne, Adams, and Gidron 2022; Lauka, McCoy, and Firat 2018; Reiljan 2020; Wagner 2021; Westwood et al. 2018). This research primarily analyzes cross-national variations in affective polarization. Yet, this country-level focus may conceal interesting between-party variations in the context of the multiparty systems found outside the United States. For instance, in a country characterized by lower overall levels of affective polarization, certain parties may still be intensely disliked by some voters (Helbling and Jungkunz 2020; Reiljan 2020; Reiljan and Ryan 2021).

To advance our understanding of affective polarization across Western democracies and move beyond cross-national variations, we address two questions: “Which partisans dislike which opposing parties?”; and “What explains variations in affective evaluations between pairs of parties?” We report three key findings, each supported by theoretical arguments. First, we show that affective polarization is linked with elite policy disputes, and its association with elite disagreements on cultural issues, such as immigration, national identity and multiculturalism, has increased over time relative to economic disputes—a finding that resonates with scholarship on the growing electoral impact of cultural disagreements (Kriesi et al. 2006; Kriesi et al. 2008; Norris and Inglehart 2019). Secondly, in line with research on coalition heuristics (Fortunato and Stevenson 2013; Fortunato et al. 2021), we show that supporters of parties that jointly govern express warmer feelings toward each other than we would predict based on their policy positions. Third, extending work on changing societal cleavages following the rise of radical parties (Bornschier et al. 2021; Helbling and Jungkunz 2020), we show that the radical right is intensely disliked by mainstream society, even when controlling for policy disputes and for coalition arrangements. We also show that this intense dislike has been stable over the last two decades, despite the radical right's “normalization” (Mudde 2019).

We explore these issues by analyzing survey data from the Comparative Study of Electoral Systems (CSES) across twenty Western polities, using eighty-one election surveys between 1996 and 2017. We analyze affective polarization between pairs of parties at both the party-dyad level and the individual level. This allows us to bridge gaps between US and comparative affective polarization research (Iyengar et al. 2019), and to compare predictors of affective polarization in the US Democratic–Republican party dyad with predictors across the multiple party pairs found in other Western systems. By analyzing factors that vary across party pairs within the same country, we derive additional leverage for testing hypotheses about the predictors of affective polarization and extend the standard comparative approach that examines country-level variations.

While much affective polarization research analyzes psychological individual-level mechanisms, our findings highlight the policy and institutional underpinnings of out-party dislike. With respect to policy, we find that the rise of cultural issues in Western politics matters not only for partisan strategies and vote choices (Abou-Chadi and Wagner 2019; Kitschelt 1994; Kriesi et al. 2008), but also for interparty hostility. With respect to institutions, our findings support the argument that proportional electoral systems produce “kinder, gentler” politics than do majoritarian democracies (Drutman 2019; Gidron, Adams, and Horne 2020; Horne, Adams, and Gidron 2022; Lijphart 2010) by encouraging the formation of coalition governments that prompt positive affective evaluations between their members. At the same time, proportional systems may intensify affective polarization by facilitating the rise of intensely disliked radical right parties.

Analyzing Affective Polarization across Countries

The United States' intensifying affective polarization is attracting growing attention due to its negative economic, social, and political consequences. Affective polarization in US society is linked with lower trust in government (Hetherington and Rudolph 2015) and the erosion of democratic norms (Levitsky and Ziblatt 2018). In comparative research, more intense affective polarization is linked to lower satisfaction with democracy (Wagner 2021), higher perceptions of ideological polarization (Ward and Tavits 2019), and decreased interpersonal trust (Westwood et al. 2018).

While scholars increasingly apply the concept of affective polarization to diverse political contexts, no consensus has emerged regarding its measurement in multiparty systems. In the two-party US system, affective polarization is defined in terms of differences in affective evaluations between one's preferred party and the out-party (Iyengar et al. 2019); yet, how should it be measured in countries with more than two major parties? The standard approach is to compute country-level scores by averaging partisans' affective evaluations toward all out-parties in the system, weighted by party vote shares (for a review, see Wagner 2021). This approach has proved useful in identifying country-level predictors of affective polarization, such as political corruption and income inequality (Gidron, Adams, and Horne 2020; Reiljan 2020); yet, it poses two challenges.

First, we have data from only about twenty Western countries that differ along dozens of dimensions that may influence affective polarization. Thus, when analyzing predictors of affective polarization across countries, scholars quickly exhaust their degrees of freedom. Secondly, partisans in multiparty systems may dislike some parties while expressing positive feelings toward other parties (Bergman 2021). Thus, a low country-level aggregate affective polarization score may conceal intense dislike toward specific parties. For instance, the Netherlands stands out in cross-national research for its muted affective polarization, yet mainstream Dutch partisans express intense hostility toward the radical right (Harteveld 2021b; Reiljan 2020).

There is a path forward, in that some variables linked to affective polarization vary not only between countries, but also between pairs of parties within the same country. This applies to elite policy differences. Moreover, parliamentary democracies with coalition governments feature pairs of parties that govern together, pairs of opposition parties, and pairs consisting of one governing and one opposition party. We can also compare affective relationships between different types of parties, such as mainstream parties versus the radical right and left. This pairwise approach provides added statistical leverage by increasing variation in the independent variable, as well as by greatly increasing the number of cases to be analyzed, for instance, a country with six major parties (such as Germany following the 2017 elections) represents a single case when analyzed at the party-system level but features fifteen distinct party pairs. The party-pairs approach also bridges analyses of affective polarization in the US two-party system (with one party pair) and in multiparty systems (with multiple pairs).

While our dyadic approach cannot illuminate the impact of factors that are constant across pairs of parties in the same system—such as electoral laws, economic inequality, and legacies of democratization—it may illuminate factors that vary between party pairs at the same time point. On this basis, we proceed to theorize predictors of affective polarization across pairs of parties.

Who Dislikes Whom? Hypothesizing Affective Polarization across Pairs of Parties

The Role of Elite Ideological and Issue-Based Disagreements

Scholars of US polarization have extensively analyzed the relationship between elite-level ideological polarization and mass-level affective polarization. Rogowski and Sutherland (2015) show that greater ideological differences between US candidates intensify affective polarization (see also Lelkes 2021), while Abramowitz and Webster (2017) find that partisans' affective evaluations respond to perceived ideological distance. Comparatively, Westwood et al. (2018) demonstrate that partisans discriminate more against out-partisans that are farther away from them ideologically. There is a growing agreement that affective polarization is rooted, at least partially, in policy disagreements (Orr and Huber 2020).

Yet, an agreement does not yet imply full consensus, as alternative accounts de-emphasize ideological disputes. From the perspective of social identity theory, affective polarization reflects partisans' psychological needs to enhance their group's status (Huddy, Bankert, and Davies 2018). Shifting the emphasis from ideological disagreements to group identities, this suggests that US affective polarization is primarily driven by the overlap of religious, racial, and partisan identities, rather than ideology (Mason 2018). Levendusky and Stecula (2021, 1, emphasis in the original) summarize this view by noting that “even when the parties agree on the issues, they still dislike and distrust one another—this is not simply a difference of opinion, but is rather something deeper.” We evaluate this relationship across pairs of parties:

The Left–Right ideology hypothesis (H1): Greater elite ideological distance between pairs of parties predicts greater affective polarization between these parties' supporters.

Issue-based elite polarization: economics and culture

While comparative affective polarization research emphasizes elite disagreements on Left–Right ideology (see, for example, Reiljan 2020; Wagner 2021), there is a rationale for distinguishing between two cross-cutting dimensions that largely structure Western politics: economics and culture (Hooghe and Marks 2017; Kitschelt 1994; Kriesi et al. 2006; Kriesi et al. 2008). Disagreements on such issues as national identity might generate especially intense partisan hostility, as it may be harder to compromise on cultural questions that are intertwined with deeply held identities (Goren and Chapp 2017; Tavits 2007). Levitsky and Ziblatt (2018, 171) contend that “the two [US] parties are now divided over race and religion—two deeply polarizing issues that tend to generate greater intolerance and hostility than traditional policy issues such as taxes and government.” Indeed, Hetherington, Long, and Rudolph (2016) show that mass polarization in the United States is linked to moral and racial issues. In this regard, Sides, Tesler, and Vavreck (2018, 10, emphasis added), succinctly summarize this point: “Issues like immigration, racial discrimination, and the integration of Muslims boil down to competing visions of American identity and inclusiveness. To have politics oriented around this debate—as opposed to more prosaic issues like, say, entitlement reform—makes politics ‘feel’ angrier.”

There is no reason to expect this strong relationship between cultural disagreements and a negative partisan affect to be uniquely American. In cross-national research, Norris and Inglehart (2019, 54) observe that “economic issues are characteristically incremental, allowing left and right-wing parties to bargain…. By contrast, cultural issues, and the politicization of social identities, tend to divide into ‘Us-versus-Them’ tribes.” This suggests that political disputes over issues of national identities and core cultural values should generate intense negative emotions across Western democracies.

Yet, again, there is an alternative perspective. Abramowitz and Webster (2017, 633) and Iyengar, Sood, and Lelkes (2012, 422) find that Americans' negative affect toward out-parties is more strongly associated with welfare attitudes than lesbian, gay, bisexual, trans and queer (LGBTQ) rights. However, attitudes toward welfare may not be a “pure” economic issue, particularly in the United States, where welfare attitudes are highly racialized and hence intertwined with notions of multiculturalism (Gilens 1999). On this basis, we evaluate whether elite polarization between pairs of parties on cultural issues is significantly associated with affective polarization, independently of elite economic polarization:

The culture and economics hypothesis (H2a): Elite polarization on cultural issues predicts partisans' affective evaluations of out-parties, independently of polarization on economic issues.

We might also expect the link between party elites' cultural positions and affective polarization to have intensified over time. Since the 1980s, parties have increasingly emphasized cultural issues, as described by Norris and Inglehart (2019, 50):

[T]oday the most heated political issues in Western societies are cultural, dealing with the integration of ethnic minorities, immigration and border control, Islamic-related terrorism, same-sex marriage and LGBTQ rights, divisions over the importance of national sovereignty versus international cooperation, the provision of development aid, the deployment of nuclear weapons, and issues of environmental protection and climate change.

Kriesi et al. (2008) similarly show that cultural issues have gained increased salience in Western electoral politics. Analyzing data from the Comparative Manifesto Project (CMP), Hall (2020) shows that since the late 1990s, cultural issues have become more salient in partisan platforms than economic issues. This development is reflected in mass politics, where citizens increasingly define their political identities based on cultural worldviews (de Vries, Hakhverdian, and Lancee 2013). Within the US context, Drutman (2019) links the rise of partisan resentment to the shift from the (economic) politics of “Who gets what?” to the (cultural) politics of “Who are we?” This suggests that elite cultural disagreements have become a stronger predictor of partisan dislike over time:

The cultural intensification hypothesis (H2b): Over time, the impact of elite cultural disputes on affective polarization has increased relative to the impact of elite economic disputes.

The Role of Power-Sharing: Interparty Cooperation in Coalition Governments

While US politics scholars emphasize the role of elite ideology, additional factors are plausibly associated with partisans' affective evaluations in multiparty systems. Coalition governments, common in multiparty systems, are one such factor, for two reasons. First, Fortunato and Stevenson (2013) show that citizens employ a coalition heuristic to infer that co-governing parties share more similar ideologies than is implied by their manifestos (see also Fortunato et al. 2021). This coalition heuristic may enhance partisans' affect toward co-governing out-parties by prompting them to perceive these parties as sharing the partisans' own policy views, beyond what we would expect based on objective measures of party position.

Secondly, partisans may warmly evaluate coalition partners because they form positive impressions of their character-based qualities, such as competence (Green and Jennings 2017). This is because governing elites tend to eschew public character-based criticisms of coalition partners; conversely, opposition elites relentlessly attack governing parties' integrity, leadership, and competence, while directing fewer attacks at co-opposition parties (Weschle 2018). Given that partisans take cues from preferred parties with respect to positions (Lenz 2012; Steenbergen, Edwards, and de Vries 2007) and values (Goren, Federico, and Kittilson 2009), they may also take cues about rival parties' character. Focusing on Israel, Bassan-Nygate and Weiss (2022) experimentally demonstrate that co-governing partisans provide each other with an affective bonus.

The preceding considerations suggest that governing parties will evaluate their coalition partners more warmly than we would expect based on these parties' objective policy positions:

The co-governance hypothesis (H3a): Controlling for objective policy differences, governing parties' partisans express warmer affective orientations toward their coalition partners.

Similarly, opposition party elites' tendencies to publicly attack the government, not co-opposition parties, may prompt opposition parties' supporters to evaluate other opposition parties more warmly than governing parties, all else equal:

The co-opposition hypothesis (H3b): Controlling for policy differences, opposition parties' supporters express warmer affective evaluations toward other opposition parties.

Radical Right Exceptionalism

Finally, we consider the possibility that radical right partisans despise other parties and are reciprocally despised, beyond what we might expect based on elite policy differences and coalition arrangements. Supporters of the radical right may intensely dislike mainstream parties because the radical right's discourse often generates negative emotions. Marx (2020) shows that radical politicians' populist rhetoric angers supporters by identifying scapegoats who are blamed for people's hardship. This relationship is more pronounced among supporters of the radical right than the radical left (Nguyen 2019, 3). As noted by Betz and Oswald (2021, 122, 134), “anger, rage resentment, and indignation together with anxiety and fear” play a key role in the discourse of populist radical right parties, whose success “is to a large extent the result of their ability to exploit a range of negative emotions.” These emotions may spill into out-party affect:

The radical right dislike of other parties hypothesis (H4a): Radical right parties' supporters dislike other parties more than is implied by elite policy distances and coalition arrangements.

While research on populism predicts radical right supporters' negative affect toward mainstream parties, we might also expect radical right parties to prompt reciprocal hostility from mainstream partisans (Meléndez and Kaltwasser 2021; Mudde and Kaltwasser 2018). Indeed, previous work has already revealed the intense dislike of mainstream partisans toward the radical right in Germany and Austria (Helbling and Jungkunz 2020), as well as Sweden (Reiljan and Ryan 2021). As noted by Reiljan (2020, 392):

Partisan like–dislike matrices demonstrate that right-populist parties clearly stand out in otherwise affectively rather moderate NWE [Northwestern European] countries. Be it PVV [Party for Freedom] in the Netherlands, SD [Sweden Democrats] in Sweden, SVP [Swiss People's Party] in Switzerland and so on: all of these parties are most intensely disliked by the supporters of many other parties. It is very likely that this is due to the rigid stances the right-populist parties exhibit on issues like immigration, European integration and Islam.

Yet, it might be that radical right parties are disliked not just because of their issue positions on these cultural issues. Radical right parties are harshly criticized for promoting intolerance by both political opponents (Akkerman and Rooduijn 2015; Hjorth 2018) and the media (van Heerden and van der Brug 2017). Helbling and Jungkunz (2020) suggest that this hostility toward the radical right is driven not only by policy differences, but also by the threat radical right parties pose to mainstream voters' social norms and perceptions of decency (for a discussion of the Swedish case, see Reiljan and Ryan 2021). If this is the case, then mainstream partisans' hostility toward radical right parties should exceed what we can explain based on these parties' policy positions:

The radical right ostracism hypothesis (H4b): Radical right parties are more strongly disliked by other parties' partisans than is implied by elite policy distances and coalition arrangements.

Data and Measurement

We analyze survey data from twenty polities and eighty-one elections between 1996 and 2017, with 506 pairs of parties, taken from the CSES. We analyze every Western democracy for which at least two CSES election surveys are available: Australia, Austria, Canada, Denmark, Finland, France, Germany, Great Britain, Greece, Iceland, Ireland, Israel, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, and the United States. We limit our sample to Western democracies for comparability. While previous comparative affective polarization research has included non-Western democracies (for example, Hernandez, Anduiza, and Rico 2021; Ward and Tavits 2019), our theoretical arguments center on features of Western polities, such as the structure of the policy space and the rise of the radical right. Section S1 in the Online Supplementary Material lists the elections and parties in our dataset.

The CSES includes a 0–10 feeling thermometer asking respondents to rate parties in their country.¹ While this survey question has been used by scholars of comparative politics to capture the likelihood of supporting a party (van der Eijk et al. 2006, footnote 17), it is, in fact, the most common measure of out-party dislike in affective polarization research (Iyengar et al. 2019). The feeling thermometer is closely—albeit not perfectly—correlated with other affective polarization measures (Druckman and Levendusky 2019). We recode this variable so that higher values denote more negative out-party evaluations. The CSES surveys also include a question about party identification,² which we use to classify respondents as party supporters. We follow previous affective polarization research and focus on party identifiers (for a discussion, see Reiljan 2020, 388). To measure affective evaluations toward out-parties, we must first classify voters based on their “in-party” identity.

Initial Patterns: Ideological Distance and Radical Right Exceptionalism

Figure 1 previews our findings and displays partisans' average thermometer ratings of out-parties. Parties were grouped into families based on the CMP classifications. To clarify the computations, the top-left panel displays the mean thermometer scores that radical left parties' partisans assigned to out-parties from different families. The bars denote mean thermometer ratings, where higher values denote more intense dislike. The labels below each bar display the party family that radical left partisans were evaluating.

Fig. 1. Affective distance between party families in twenty Western democracies.

Notes: The figure displays the mean thermometer ratings that partisans assigned to out-parties, with out-parties grouped into families (listed along the horizontal axis). Thus, the upper-left panel displays the average thermometer ratings (the vertical axis) that radical left parties' partisans assigned to parties from the other party families (green, social democratic, liberal, Christian democratic, conservative, and radical right). We have reversed the thermometer ratings so that 10 denotes maximum dislike and 0 maximum liking. The family assignments are based on the CMP classifications. Section S1 in the Online Supplementary Material lists the countries, elections, and parties in our dataset.

Figure 1 highlights the links between ideology, radical right exceptionalism, and affective polarization. First, the patterns support the Left–Right ideology hypothesis (H1) that greater Left–Right distance between pairs of parties predicts greater affective polarization between these parties' supporters. The panels in Figure 1 display a general pattern whereby supporters of leftist parties (from the radical left, green, and social-democratic party families) tend to assign other leftist parties neutral ratings, on average, but express greater hostility toward parties on the right. Similarly, partisans of mainstream right conservative and Christian-democratic parties report neutral feelings toward each other, on average, but express hostility toward mainstream leftist parties and intense hostility toward the radical left.

Secondly, we see evidence of radical right exceptionalism: the mainstream right conservative and Christian-democratic partisans express more hostility toward radical right parties than toward mainstream left parties, on average. By contrast, green and social-democratic party supporters do not express comparable hostility toward the radical left. These patterns suggest that it is not radical parties in general, but radical right parties in particular, that incite hostility—in line with the radical right ostracism hypothesis (H4b) that radical right parties are disliked more intensely than is implied by policy distances.

Multivariate Analyses

To further evaluate our hypotheses, we report multivariate analyses. Our dependent variable, Party i's supporters' evaluations of out-party j (t), is the average thermometer score that party i's partisans assigned to party j in the election survey administered at time t. We focus on out-party affective evaluations, rather than the difference between in-party and out-party evaluations, because our theoretical expectations are centered on feelings toward out-parties. For instance, we expect that people will evaluate parties that co-govern with their own party, but we do not have any expectations for how co-governance would affect in-party evaluations. Moreover, to the extent that coalition governance does impact in-party evaluations, this warmth would not be dyad specific, but spill over into the in-party component for all of a party's paired relationships, even ones lacking co-governance. This decision follows Levendusky and Stecula (2021, 27), in their work on the effects of cross-party conversations on affective evaluations, who write: “Our theoretical argument in Section 1 focuses on how cross-party discussion reduces animosity toward the other party. Given this, we measure affective polarization by looking at ratings of the other party, rather than by examining the difference between same-party and out-party ratings, as some scholars do.” We follow their approach and use out-party affective evaluations as our dependent variable.

We begin with the analysis of affective evaluations between 2,232 party dyads. Later in the text, we present the results of individual-level regressions, analyzing 359,277 party evaluations made by 85,977 distinct individuals. The results of the individual-level analyses closely mirror those of the dyad-level analyses and are reported in the section on robustness checks.

Our independent variables are as follows. We rely on measures developed by the CMP to assess the relationship between affective polarization and Left–Right [RILE] ideological disagreements (H1), as well as the relationship between affective polarization and economic and cultural disputes (H2a and H2b) (Volkens, Lehmann, and Matthieß 2017). The variable Elite Right–Left polarization i, j (t) denotes the Right–Left distance between parties i and j in the election at time t. The variable Elite economic polarization i, j (t) denotes elite-level distance on economic issues, such as welfare state policies and economic regulation. The Elite cultural polarization i, j (t) variable denotes elite-level distance on cultural issues pertaining to national identity and traditional morality. All scales range between −100 and +100. In constructing our elite economic and cultural distance measures, we follow the coding scheme from the CMP.³ Section S2 in the Online Supplementary Material reviews the construction of these measures, including the list of issues included in each measure.

To evaluate the co-governance hypothesis (H3a) that co-governing status predicts warmer affective evaluations, we created the variable i, j are coalition partners (t), which equals 1 if parties i and j were coalition partners at the time of the election. To evaluate the co-opposition hypothesis (H3b), we created the dummy variable i, j are opposition partners (t). We rely on data from ParlGov on cabinet participation to construct these variables. Excluding the unique case of Switzerland,⁴ the average Left–Right distance between coalition partners is 18.7 units, whereas the average Left–Right difference between opposition pairs is 22.4 units. Interestingly, co-governing and co-opposition parties are nearly equally distant on economic issues (12.7 versus 11.7 units), but co-governing parties are much closer on cultural issues than are opposition dyads (14.4 versus 19.6 units). Our universe of cases includes coalition partners that are ideologically distant from one another, for instance, the right-wing conservative National Coalition and Left Alliance in Finland governed together during 1999–2003.

To evaluate the radical right ostracism hypothesis (H4b) that radical right parties are more disliked by other parties' partisans than is implied by elite policy distances and coalition arrangements, we create the dummy variable Out-party j is radical right, which equals 1 if j, the out-party that i's partisans are evaluating, is a radical right party. To evaluate the radical right dislike of out-parties hypothesis (H4a), we create the dummy variable In-party i is radical right, which equals 1 if i, the in-party whose partisans evaluate the out-party j, is a radical right party. We rely on the CMP classifications to assign parties to party families.

Table 1 presents descriptive statistics for our variables. It should be noted that each party pair i, j in each election survey enters our data set twice, as we analyze both the mean thermometer rating that party i's partisans assign to party j and the mean rating that j's partisans assign to i. Out-party ratings average about 6, where 10 denotes maximum dislike and 0 denotes maximum liking.⁵ The standard deviation is 1.53, indicating considerable variation in partisan constituencies' mean ratings of different out-parties. It is this variation we seek to explain. About 8 per cent of the party pairs in our study were coalition partners, about 45 per cent were both in opposition, and 46 per cent included one opposition and one governing party. Roughly 9 per cent of the pairs included a radical right in-party (hence, roughly 9 per cent included a radical right out-party).

Table 1. Descriptive statistics

Notes: N = 2,232. Table 1 presents descriptive statistics of the independent variables in our analyses of out-party dislike between pairs of parties. Section S1 in the Online Supplementary Material lists the countries, elections, and parties in our dataset.

Hypothesis Tests

We estimate the parameters of two multivariate models that control for all of these factors simultaneously. We first estimate a Left–Right full model that incorporates elites' Left–Right disagreements, power sharing arrangements, and radical right exceptionalism:

(1)$$\eqalign{& Party \,i^{\prime}s \,supporters^{\prime} \,evaluations \,of \,out\!\!-\!\!party\,j( t ) = b_1 + b_2[ {elite\,right-left\, polarization i, \;j( t ) } ] \cr & \quad + b_3[ {i, \;\,j\,are\,coalition\,partners( t ) } ] \cr & \quad + b_4[ {i, \;\,j\,are\,opposition\,partners( t ) } ] \cr & \quad + b_5[ {in {\hbox -} party \,i \,is\,radical\,right} ] \cr & \quad + b_6[ {out {\hbox -} party\,j\,is\,radical\,right} ] } $$

The Left–Right ideology hypothesis (H1) implies that the coefficient on the Elite Right-Left polarization i, j (t) variable will be positive, that is, that greater elite Left–Right distance between parties i and j intensifies party i's partisans' dislike of party j. The co-governance (H3a) and the co-opposition hypotheses (H3b) imply that the coefficients on the i, j are coalition partners (t) and the i, j are opposition partners (t) variables will be negative, denoting that shared coalition and shared opposition status defuse out-party dislike, compared to the residual category of a party pair consisting of one governing and one opposition party. Our radical right exceptionalism hypotheses imply that the coefficients on the In-party i is radical right and Out-party j is radical right variables will be positive, that is, that radical right partisans dislike other parties (H4a) and are reciprocally disliked (H4b), controlling for policy disputes and coalition arrangements.

Next, we estimate a model in which we substitute the elite economic and cultural polarization variables for the elite Left–Right polarization variable:

(2)$$\eqalign{& Party\,i^{\prime}s \,supporters^{\prime}\,evaluations\,of\,out {\hbox -} party \,j( t ) = b_1 + b_2[ {elite \,economic\,polarization \,i, \;\,j( t ) } ] \cr & \quad + b_3[ {elite\,cultural \,polarization \,i, \;\,j( t ) } ] _ + b_4[ {i, \;j\,are\,coalition\,partners( t ) } ] \cr & \quad + b_5[ {i, \;j\,are\,opposition \,partners( t ) } ] + b_6[ {in {\hbox -} party\,i \,is \,radical \,right} ] \cr & \quad + b_7[ {out {\hbox -} party\,j \,is \,radical \,right} ] } $$

The culture and economics hypothesis (H2a) implies that the coefficients on the Elite cultural polarization i, j (t) and Elite economic polarization i, j (t) variables will both be positive.

We estimate the models using ordinary least squares (OLS) with robust standard errors clustered by election, as there is reason to expect that error terms are correlated within elections. We standardize the ideological and issue distance variables so that the coefficient estimates denote the variation in the dependent variable values associated with one-standard-deviation changes in the independent variable values. Our models include country and year fixed effects to capture unmeasured factors that are associated with specific countries and time periods, such as national economic conditions, political scandals, and so on. Thus, our parameter estimates reflect within-country and within-year differences in partisans' ratings of different out-parties. We report a set of reduced models that separately examine each of our predictors in the Online Supplementary Material (see also Lenz and Sahn 2021).

Results

The parameter estimates reported in Table 2 support our hypotheses that out-party dislike is linked with elite ideological and issue differences, with coalition arrangements, and with radical right exceptionalism. With respect to ideology, the estimates support our Left–Right ideology hypothesis (H1). In the Left–Right model (Column 1), a one-standard-deviation increase in Left–Right distance between parties i and j (about 17 units on the −100 to +100 RILE scale) increases predicted out-party dislike by 0.54 units on the 0–10 thermometer scale, or about one-third of this variable's standard deviation (which is 1.56 units), when controlling for coalition arrangements and for radical right exceptionalism.

Table 2. The predictors of mean out-party dislike

Notes: N = 2,232. The dependent variable, Party i's supporters' evaluations of out-party j (t), is the average thermometer rating on a 0–10 scale that party i's partisans assigned to party j in the election survey administered at time t. We reversed the thermometer ratings so that 10 denotes maximum dislike. The dependent variables are defined in the text. The OLS regression models were estimated with standard errors clustered on elections, with country and year fixed effects, which are reported in Section S3 of the Online Supplementary Material. Section S2 of the Online Supplementary Material reports reduced models, while Section S1 lists the countries, elections, and parties in our dataset. ** p ≤ 0.01; * p ≤ 0.05 (two-tailed tests).

We also find support for the expectation that elite polarization on cultural issues is associated with partisans' out-party dislike, independently of polarization on economic issues (H2a). The coefficients on the elite economic and cultural distance model (Column 2), +0.37 and +0.40, respectively (p < 0.01), imply that elite economic and cultural disagreements were equally associated with out-party evaluations across the 1996–2017 period of our study, with one-standard-deviation increases in elite economic and cultural differences each increasing predicted out-party dislike by about 0.4 thermometer units, all else equal. In the following, we show that—consistent with our cultural intensification hypothesis (H2b)—the association between elite cultural polarization and affective evaluations has increased across this time period.

Next, our estimates support our expectation that governing parties' partisans express warmer affective evaluations toward their coalition partners (H3a). In our Left–Right ideology full model (Column 1), the coefficient on the i, j are coalition partners (t) variable, −0.87 (p < 0.01), denotes that when party i is in government, i's partisans' predicted dislike of the out-party j is 0.87 units less on the 0–10 thermometer scale when j is a coalition partner than when j is in opposition, holding elite Left–Right differences and radical right status constant. Moreover, our negative estimates on the i, j are opposition partners (t) variable also support the co-opposition hypothesis (H3b). We estimate a co-opposition “affective bonus” of about 0.3 thermometer units. This relationship is weaker than the one for co-governance, which makes sense, for while the norm of collective responsibility largely precludes co-governing parties from publicly attacking each other (Martin and Vanberg 2011), there is no such norm for co-opposition parties. Nevertheless, the conventional wisdom is that opposition party elites typically direct their attacks toward the government, not each other. Our estimates support this expectation.

Finally, we detect evidence of right-wing exceptionalism. We estimate large positive coefficients on the Out-party j is radical right variable, which supports the radical right ostracism hypothesis (H4b) that radical right parties are more strongly disliked than is implied by elite policy distances and coalition arrangements. For the economics and culture model (Column 2), the coefficient on the Out-party j is radical right variable, +1.47 (p < 0.01), implies that party i's partisans' predicted dislike for a radical right out-party j is 1.47 thermometer units more intense than when j is not radical right, all else equal. This “affective penalty” is enormous: it equals a nearly one-standard-deviation change of the out-party dislike variable. It should also be noted that this is the estimated effect when controlling for elite disagreements on the cultural issues that define radical right parties' core identities. This implies that while radical right parties may be despised, in part, for their cultural issue stances (Reiljan 2020), they are also despised for reasons that go beyond these cultural debates.

We also estimate positive coefficients on the In-party i is radical right variable, which supports the radical right dislike of other parties hypothesis (H4a). The estimates imply that radical right supporters penalize other parties by 0.3 to 0.5 thermometer units, beyond what we would predict based on elite policy differences and coalition arrangements. This implies that radical right supporters disproportionately dislike other parties but do not fully return the intense hostility that others direct at them.

Panels A and B of Figure 2 illustrate the relationship between our independent variables and out-party dislike, based on the estimates for the Left–Right ideology model (see Column 1 in Table 2). Panel A of Figure 2 displays predicted values of party i's partisans' dislike of the out-party j (the vertical axis) as a function of Left–Right polarization between these parties' elites (the horizontal axis) when neither party is radical right. The figure plots predicted out-party dislike for values of the Elite Right–Left polarization i, j (t) variable, ranging from one standard deviation below the mean value in our dataset to one standard deviation above the mean (a range from about 5 to 40 units of interparty distance on the CMP RILE scale). The figure displays predicted out-party dislike both when parties i and j are coalition partners (the bottom, blue line) and when the i–j party pair comprises one governing and one opposition party (the top, red line).⁶ Consistent with the Left–Right ideology hypothesis (H1), out-party dislike increases with elite Left–Right distance. Consistent with our co-governance hypothesis (H3a), party i's partisans are predicted to evaluate the out-party j much more warmly when these parties co-govern than when one is governing and the other is in opposition. This co-governance “affective bonus” can therefore override substantial elite Left–Right differences.

Fig. 2. Elite Left–Right polarization, coalition status, and radical right exceptionalism.

Notes: Figure 2 displays predicted levels of out-party dislike along the vertical axis (where higher values denote more intense dislike) as a function of elite Left–Right polarization (the horizontal axis), based on the parameter estimates for the Left–Right model (see Column 1 in Table 2). Panel A shows ideology and coalition status; Panel B shows ideology and radical right status. The predictions are displayed with 95 per cent confidence intervals.

Panel B of Figure 2 illustrates the relationships between elite ideological polarization, radical right exceptionalism, and out-party dislike when in-party i is not radical right and both parties are in opposition. The figure displays predicted levels of out-party dislike when moving along the elite Left–Right polarization variable for two scenarios: one where the out-party j is radical right (top, red line); and one where the out-party j is not radical right (bottom, blue line). We see the large affective penalty exacted on the radical right: party i's partisans are predicted to dislike out-party j much more intensely when j is radical right and parties i and j espouse similar ideologies separated by only 5 units on the RILE scale (one standard deviation below the mean) than when j is not radical right and i and j espouse sharply different ideologies separated by 40 units on the RILE scale (one standard deviation above the mean).

How the Relative Impact of Culture versus Economics Has Changed over Time

The parameters reported in Table 2 were estimated over all the cases in our data and thus do not pertain to the cultural intensification hypothesis (H2b) that the impact of elite cultural disagreements on affective polarization has increased over time relative to the impact of economic disagreements. To evaluate this hypothesis, we divided our data into two time periods: 1996–2006 and 2007–2017.⁷ Table 3 reports our parameter estimates for a basic economic and cultural polarization model (see Columns 1–2) and a fully specified model (see Columns 3–4).

Table 3. The predictors of out-party dislike over time

Notes: The dependent variable, Party i's supporters’ evaluations of out-party j (t), is the average thermometer rating on a 0–10 scale that party i's partisans assigned to party j in the survey administered at time t. We reversed the thermometer ratings so that 10 denotes maximum dislike. The dependent variables are defined in the text. The OLS regression models were estimated with standard errors clustered on elections. All models include country and year fixed effects. **p ≤ 0.01; *p ≤ 0.05 (two-tailed tests).

These findings support the cultural intensification hypothesis (H2b). Without control variables (see Models 1–2), the predicted changes in out-party dislike associated with one-standard-deviation changes in elite economic and cultural polarization are nearly identical for the 1996–2006 period; however, for the more recent 2007–17 period, the estimated impact of cultural polarization on out-party dislike is more than double that for economic polarization. This difference is statistically significant (p < 0.01). When we look at the models with the control variables (Models 3–4), the results are somewhat weaker. In these models, we still see an increase in the relative predictive strength of cultural ideological polarization as opposed to economic polarization, but the over-time change in the difference between these coefficients is no longer statistically significant. Controlling for whether a party is a member of the radical right party family, which holds extreme positions on the cultural dimension, captures much of the over-time change and suggests that a major driver of the increasing importance of the cultural dimension is the rise of radical right parties.

By contrast, we detect no significant temporal changes in the predictive power of other variables. It is worth stressing that we find similar estimates of radical right exceptionalism for the earlier and later time periods. This is notable given that radical right parties in several European party systems have been “normalized” during more recent time periods, as they joined coalition governments (Akkerman and Rooduijn 2015; Mudde 2019).

Robustness Checks

The statistical analyses we reported earlier featured as the dependent variable the mean rating that all partisans of a focal party i assigned to a given out-party j. While this aggregate approach is consistent with previous comparative work that analyzes affective polarization at the country level (Gidron, Adams, and Horne 2020; Reiljan 2020), it does not control for possible differences across respondents in their interpretations of the 0–10 thermometer scale, that is, the possibility of differential item functioning.

To address this issue, we estimated individual-level models in which the dependent variable was each respondent's out-party thermometer rating, while including respondent-specific intercepts designed to account for variations in how different respondents centered the thermometer scale. Since our data on coalition history and parties' ideological and issue positions were constructed at the party-dyad level, for these analyses, we created a stacked dataset in which each row was a partisan survey respondent's rating of a given out-party j and the coalition history and ideological positioning of the parties in the dyad. This leaves us with 359,277 party evaluations made by 85,977 distinct individuals. These models do not include the in-party radical right variable because an individual's in-party does not vary. These analyses, reported in Table 4, support the same substantive conclusions reported earlier.

Table 4. The predictors of out-party dislike, party-dyad and individual-level analyses

Notes: The dependent variable is the respondent's thermometer rating on a 0–10 scale assigned to party j in the election survey administered at time t. We reversed the thermometer ratings so that 10 denotes maximum dislike. The dependent variables are defined in the text. The OLS regression models were estimated with standard errors clustered on individuals and include individual-level fixed effects. Section S1 in the Online Supplementary Material lists the countries, elections, and parties in our dataset. ** p ≤ 0.01; * p ≤ 0.05 (two-tailed tests).

We performed several additional robustness checks to substantiate our findings, as reported in Section S5 of the Online Supplementary Material. First, we acknowledge that while the CMP is a rich source for data regarding partisan positions, and one that converges with other sources of information about partisan positions (Adams et al. 2019), it is not without limitations. More specifically, it directly measures salience of issues, rather than positions, though the two dimensions of a partisan manifesto are most likely linked. We therefore re-estimated our models using alternative measures of elite polarization based on the Chapel Hill Expert Surveys, in which experts place European parties on various ideological and policy scales (Bakker et al. 2015). These analyses continue to support our substantive conclusions. Secondly, we analyzed models that examine change over time in the relationship between affective polarization and elite Left–Right polarization, and found that this relationship has been stable. Third, to substantiate that our findings on the intensification of cultural issues over time are not an artifact of countries moving in and out of our dataset across the different time periods, we re-estimated our models while including only those countries with CSES election surveys included in both the earlier (1996–2006) and later time periods (2007–17) in our study. These analyses continue to support the cultural intensification hypothesis. Fourth, we examined the cultural intensification hypothesis using individual-level data, and the findings are substantively similar. Fifth, to explore the possibility of a nonlinear relationship between elite disagreements on out-party dislike, we re-estimated our models while adding a squared elite ideological distance term to our specifications and found no evidence of nonlinearity. Sixth, we clustered the standard errors by country and the results are substantively similar. Lastly, we present partial residual plots for each country separately, with the raw data for the three continuous variables (elite Left–Right polarization, elite economic polarization, and elite cultural polarization) and the fitted lines. The patterns nicely mirror those presented in Figure 2.

Discussion and Conclusions

Animosity and resentment across party lines, that is, affective polarization, prompts concern in contemporary US politics and across Western democracies. We have introduced an empirical approach to this topic that emphasizes affective relationships between pairs of parties. Using this perspective, we have argued and empirically shown that: out-party dislike reflects elite policy disputes; governing parties' supporters grant coalition partners a large “affective bonus” (and co-opposition status prompts a smaller affective bonus); and radical right parties engender far more hostility than we can explain based on elite policy disputes and coalition arrangements, while radical right supporters' reciprocal dislike of opponents is less intense. We also showed that elite cultural disagreements increasingly predict out-party dislike, relative to the economic disputes that formerly dominated post-war Western politics.

Our party-pairs perspective confers two advantages over the standard, party-system-level study of affective polarization: first, it bridges the analysis of the US two-party system and the multiparty systems found in other Western democracies; and, secondly, it enhances our statistical leverage by vastly increasing the number of cases to be analyzed. At the same time, our approach provides no added leverage in estimating the effects of national-level factors that are constant across different party pairs in the same national election, such as countries' electoral laws, national economic conditions, democratic histories, and media regimes. Our approach, moreover, weighs all party pairs equally, whereas analysts may prioritize affective relationships between the larger, more influential parties. Thus, we see our party-pairs focus as a complement to, rather than a substitute for, country-level analyses of affective polarization.

Hostility toward the radical right is not merely a reflection of opposition to radical right parties' positions on cultural issues, such as immigration, multiculturalism, and national identity. In this regard, previous studies document that mainstream elites disparage the radical right for reasons that go beyond specific issue positions (Hjorth 2018). Our estimates suggest that mainstream parties' supporters may be cued off this elite rhetoric. Radical right partisans sense feelings of hostility and disdain from mainstream partisans (Gest 2016; Gidron and Hall 2017; Gidron and Hall 2020), and—whether this is normatively desirable or not—our estimates substantiate radical right supporters' perceptions.

The second implication of our study concerns the relationship between power-sharing arrangements and polarization. Lijphart (2010) advances the influential argument that proportional voting systems with coalition governments promote “kinder, gentler” politics compared to majoritarian democracies. Moreover, previous work does indeed show that mass-level affective polarization tends to be less intense in multiparty systems (Gidron, Adams, and Horne 2020; Horne, Adams, and Gidron 2022; McCoy and Somer 2019). Our finding that governing parties' supporters grant a large “affective bonus” to coalition partners uncovers one mechanism behind Lijphart's argument, as it implies that co-governance defuses interparty hostility. Future research might explore whether radical right parties receive a comparable affective bonus from their coalition partners' supporters. Previous work has examined the “legitimizing” effects of radical right governance (Akkerman and Rooduijn 2015), yet we do not know whether mainstream parties' refusal to cooperate with the radical right carries affective consequences. There are currently not enough cases of radical right governance in our dataset to evaluate this issue.

Since comparative affective polarization research is at an early stage, our findings suggest several follow-up research topics. First, while we have advanced theoretical arguments that predict our empirical findings, we have not parsed out the causal mechanisms underpinning the relationships we document. Future work should turn to panel data to examine, for instance, how parties' movements into and out of governing coalitions impact affective evaluations. Secondly, our study of affective polarization in the time period covered by the CSES (1996 to the present) might be extended back to the 1980s, at least in those countries with appropriate election surveys from earlier time periods (Boxell, Genzkow, and Shapiro 2020). These issues underline the long road ahead as we explore the factors that drive affective polarization across Western publics. We believe our study provides promising first steps toward this goal.