Divided We Unite: The Nature of Partyism and the Role of Coalition Partnership in Europe

Abstract

Highlighting the strength of “partyism” in many democracies, recent scholarship pays keen attention to increasing hostility and distrust among citizens across party lines, known as affective polarization. By combining a conjoint analysis with decision-making games such as dictator and trust games, we design a novel survey experiment to systematically estimate and compare the strength of the partisan divide relative to other social divides across 25 European democracies. This design also allows us to investigate how the two components of affective polarization, in-group favoritism and out-group derogation, are moderated by the way parties interact with each other. We first find dominance of the partisan divide compared to other social divides that constitute traditional cleavages such as social class and religion. Second, we show that affective polarization in Europe is not primarily driven by out-group animus. Finally, we demonstrate that coalition partnership lessens affective polarization by reducing both in-group and out-group biases.

INTRODUCTION

In recent years, many scholars and commentators have paid keen attention to increasing hostility and distrust across party lines in democracies. In particular, there is strong evidence that this mass-level animosity across party lines has surged substantially over the last decades in the United States (see, for example, Hetherington and Rudolph 2015; Iyengar, Sood, and Lelkes 2012; Mason 2015; 2016; 2018). Moreover, a recent study (Iyengar and Westwood 2015) suggests that partisan animosity exceeds hostility based on race, arguably the most salient social divide in American society (Myrdal 1944; Pager and Shepherd 2008; Schuman et al. 1997) to an extent that some scholars call this dominance of partisanship “partyism” (Sunstein 2015; Westwood et al. 2018). Conceptualized as affective polarization, the tendency of partisans to like or trust members of their own group and dislike and distrust those from the other party/ies has been understood as one of the major challenges impeding democratic and effective governance, by weakening political trust (Hetherington and Rudolph 2015), promoting a willingness to sacrifice democratic principles (Graham and Svolik 2020; Kingzette et al. 2021), and even inducing discriminatory behavior toward opposing partisans in nonpolitical contexts (Huber and Malhotra 2017; Iyengar and Westwood 2015; McConnell et al. 2018).

There have been recent attempts to put this literature in a comparative perspective. These studies have shown that affective polarization also exists in other democracies and that the United States is not a country with a particularly high level of affective polarization among citizens (see Gidron, Adams, and Horne 2019; Reiljan 2020; Wagner 2021; see also Boxell, Gentzkow, and Shapiro 2020). We attempt to advance this literature by addressing the following challenges in investigating the extent of the partisan divide and its nature in a systematic manner. Although the well-known argument for the primacy of partyism presumes the strength of partisan divide relative to other social divides across different contexts, a systematic estimation and comparison of the multiple political divides has been challenging, in particular, in a cross-national setting. Most cross-national studies largely rely on analysis of observational data such as the Comparative Study of Electoral Systems (CSES) data (Gidron, Adams, and Horne 2019; 2020; Harteveld 2021; Hernandez, Anduiza, and Rico 2021; Reiljan 2020; Wagner 2021). However, partisanship can proxy other social identities (e.g., Achen and Bartels 2016; Berelson, Lazarsfeld, and McPhee 1954; Campbell et al. 1960; Green, Palmquist, and Schickler 2002) or convey information about issue positions (e.g., Abramowitz and Saunders 2006; Niemi and Jennings 1991). Thus, a conventional observational approach cannot establish and disentangle the causal effect of partisanship itself. Moreover, the most frequently used survey item measures respondents’ evaluations of parties, rather than fellow citizens with partisanship, which can bias the estimation of affective polarization among citizens (Druckman and Levendusky 2019; Harteveld 2021; Kingzette 2021). While some recent studies use experimental designs and control for such bias (e.g., Bassan-Nygate and Weiss 2022; Carlin and Love 2018; Westwood et al. 2018), they cover only a limited number of countries, often lacking a cross-nationally valid estimation scheme to evaluate and compare the extent of the partisan divide relative to other social divides across countries. This calls for a research design to enhance the internal and external validity of the existing findings and to study the role of contextual factors that influence affective polarization.

Highlighting the importance of the context of party competition, existing literature suggests that the way political parties interact with one another can play a key role in shaping the extent and nature of affective polarization. According to the social identity approach, affective polarization is a result of classifying co-partisans as members of an in-group and opposing partisans as members of an out-group (Iyengar et al. 2019). Although in-group favoritism does not necessarily lead to out-group hostility, the reciprocal relationship between the two sentiments tends to be strengthened in a political context in which parties and elites mobilize group identities and compete to secure or maintain power (Brewer 1999), as originally suggested by realistic conflict theory (Sherif and Sherif 1953). On the other hand, the existence of superordinate goals is widely believed to provide the conditions necessary for intergroup cooperation and conflict reduction (see, for example, Sherif 1966). Thus, the characteristics and conditions of interparty competition are expected to play a significant role in shaping affective polarization. In particular, in many democracies, multiple political parties do not only compete but also often cooperate to govern in coalition. Experiences of governing together create shared goals and reward structures between competing parties, which can motivate partisans to recategorize their social identities (Brewer 2000; Gaertner et al. 1999) and to perceive coalition partners’ positions as more similar than non-partners (Fortunato and Stevenson 2013). Despite the central role of coalition partnership in shaping interparty competition, and consequently the partisan divide, existing analyses of this relationship have mostly been limited to either observational data analyses (Gidron, Adams, and Horne 2020) or single-country experimental studies (Bassan-Nygate and Weiss 2022).

To address these theoretical and empirical challenges, we design a novel cross-national experiment by combining a conjoint analysis with decision-making games. Using dictator and trust games, we measure and analyze the partisan divide among citizens as the difference in the allocation of tokens to co-partisans versus opposing partisans (also see Carlin and Love 2013; 2018; Iyengar and Westwood 2015; Westwood et al. 2018; Whitt et al. 2021). For each round, we randomly display another player for the respondent to interact with by experimentally manipulating both their party affiliation and other characteristics across multiple dimensions that constitute major social cleavages such as social class, religion, and nationality. We embed this experiment in large-scale online surveys fielded in 25 democracies in Europe. This research design allows us not only to assess the impact of partisanship relative to other social divides and the nature of affective polarization across countries, but also to evaluate how party-level factors such as coalition partnership influence the partisan divide and its configuration.

The context of contemporary Europe provides an important opportunity to evaluate both the extent and nature of the partisan divide from a broad comparative perspective and the validity of the key mechanism to explain affective polarization in different settings. Contrary to the United States, there has been a long debate about the influence of partisanship in Europe (see, for example, Dalton and Wattenberg 2000; Holmberg 2007; Thomassen 1976; Thomassen and Rosema 2009). On the other hand, the classic account of European party systems posits the salience and persistence of traditional socioeconomic cleavages (Lipset and Rokkan 1967). Thus, European party systems that reflect long-standing social, cultural, and religious cleavages can offer insights into intense conflicts across party lines (Richardson 1991). Moreover, scholars searching for potential causes of and remedies for affective polarization have discussed the role played by the institutional characteristics of European political systems (see, for example, Drutman 2019; Lelkes and Westwood 2017). Our cross-national study seeks to contribute to a further understanding of the nature of affective polarization and the role of contextual factors by broadening its theoretical and empirical scope.

We find that partisanship dominates other social divides based on traditional cleavages in almost all democracies in our sample. The magnitude of the partisan divide is indeed substantively large and much stronger than other social divides based on social class, religion and even nationality, confirming Iyengar and his colleagues’ hypothesis on the primacy of partyism in a much broader cross-national context. Notably, we find that this partisan divide in Europe is not mainly driven by out-group derogation, in contrast to recent findings in the United States (Carlin and Love 2018; Iyengar and Westwood 2015; Westwood et al. 2018). In Europe, out-group animus is not more consequential than in-group favoritism. Lastly, we show that the extent of affective polarization varies with the way political parties interact with one another, but not necessarily with the relatively static institutional conditions that shape these interactions. In particular, coalition partnership significantly lessens affective polarization by reducing both in-group and out-group biases. Taken together, our study does not only provide strong evidence to support the primacy of partisanship but also offers a more dynamic understanding of affective polarization.

Affective Polarization and Coalition Partnership

Affective polarization comes from an individual’s identification with a political party. When people identify with a political party, they divide the world into an in-group and an out-group, which can lead to in-group favoritism and bias (Tajfel 1981; Tajfel and Turner 1979). Although in-group love might not necessarily lead to out-group hatred (Allport 1954), scholars suggest that the contemporary political environment including partisan-ideological sorting (Levendusky 2009) or social sorting (Mason 2015; 2018), ideological polarization of political elites (Rogowski and Sutherland 2016; Webster and Abramowitz 2017), negative political campaigns (Iyengar, Sood, and Lelkes 2012; Sood and Iyengar 2016), and the rise of partisan news and new media (Boxell, Gentzkow, and Shapiro 2020; Druckman, Levendusky, and McLain 2018; Lelkes, Sood, and Iyengar 2017; Levendusky 2013) can intensify partisan hostility toward out-groups.

Brewer (1999) notes that in-group favoritism, even in the absence of overt antagonism toward out-groups, can still be pernicious. Many forms of discrimination and bias can develop not because out-groups are hated, but because positive emotions are reserved for the in-group and withheld from out-groups. She claims that “the very factors that make in-group attachment and allegiance important to individuals also provide a fertile ground for antagonism and distrust” of those outside the in-group boundaries (Brewer 1999, 442). Thus, affective polarization is a “natural offshoot” of partisan identification (Iyengar et al. 2019, 130; see also Druckman and Levendusky 2019; Iyengar and Westwood 2015; Levendusky 2018). Accordingly, scholars have studied affective polarization from a social identity perspective.

However, in politics, people hold multiple political identities that vary in their relevance and significance across situations. According to the social identity view of partisanship, citizens tend to think about parties in terms of other, long-standing social groups—such as class, religion, and race/ethnicity—and their attitudes toward the groups that constitute the parties drive their partisan attachments (Ahler and Sood 2018; Claassen et al. 2021; Iyengar, Sood, and Lelkes 2012; Robison and Moskowitz 2019; see also Achen and Bartels 2016; Berelson, Lazarsfeld, and McPhee 1954; Campbell et al. 1960; Green, Palmquist, and Schickler 2002). ¹ Also, drawing upon studies by Brewer and her colleagues on the psychological effects of holding multiple social identities (Brewer 1999; Brewer and Pierce 2005; Roccas and Brewer 2002), recent studies claim that individuals whose group identities are nonaligned or cross-cutting tend to be more tolerant, less biased, and feel more positively toward out-groups (Mason 2015; 2016; 2018; Mason and Wronski 2018). Conversely, when group identities are aligned to the extent that they are seen as a single identity, people tend to be less tolerant, more biased, and feel more negatively toward out-groups. ² This identity-based understanding of affective polarization directs our attention to the multidimensionality of political divides, particularly the way partisanship is related to other social groups and identities. We build upon and contribute to the existing literature by presenting a rigorous estimation that disentangles the unique effect of partisanship from the effects of other social groups and directly compares the partisan divide to other social divides that constitute enduring social cleavages.

Moreover, we note that relatively little attention has been paid yet to the dynamic nature of affective polarization, which is potentially shaped by the conditions and characteristics of interparty competition. According to the social identity approach, a political context involving struggles for power is likely to generate conditions that intensify the partisan divide (Brewer 1999). Political parties and elites mobilize partisan identities and use fears and distrust of out-groups to gain or maintain power (Iyengar, Sood, and Lelkes 2012; Michelitch 2015; Michelitch and Utych 2018; Singh and Thornton 2019). However, political parties also cooperate in many democracies with multiparty systems. Cooperative relationships can take many forms (see Casal Bértoa and Enyedi 2021). Party mergers, pre-electoral coalitions, and the withdrawal of candidates in another party’s favor are common strategies for political parties. Political parties and their members often benefit by sharing resources such as patronage, campaign infrastructure, donations, electorally relevant information, and media outlets (see also Golder 2006; Ibenskas 2016; Katz and Mair 1995). They also cooperate in formulating rules about electoral campaigns and legislative procedures (e.g., Benoit 2004).

The logic of interparty cooperation takes the most pronounced form in forming and running coalition governments. Once political parties form coalitions to secure political power, the presence of coalitions governing together is likely to foster superordinate political identities (Bassan-Nygate and Weiss 2022; González et al. 2009; Hagevi 2015) and create shared incentives for policy accommodation and compromise (Martin 2004; Martin and Vanberg 2011; see also Austen-Smith and Banks 1988; Baron and Diermeier 2001; Straffin and Grofman 1984). According to the common ingroup identity model (Gaertner et al. 1989; Gaertner and Dovidio 2000), the creation of a common category such as a coalition redefines the in-group and provides the basis for in-group favoritism toward an inclusive common group that subsumes the original categories of in-groups and out-groups. The superordinate identity with the coalition can not only lead to positive affect toward coalition members, but also neutralize the negative effects of an identity threat in party members’ attitudes toward allies (González et al. 2009). ³ Moreover, a party’s status as either a member of the governing coalition or a member of the opposition can serve as a cheap, but generally informative guide to citizens. Using this simple heuristic, they perceive coalition partners as more similar than non-partners (Fortunato and Stevenson 2013; see also Lupu 2013). ⁴ One’s perceived similarity to the prototypic group members relative to out-groups plays a key role in the development and function of social identity (Huddy 2001; see also Turner et al. 1987). The smaller party members perceive interparty differences to be, the more likely they are to like and trust out-partisans (Ahler and Sood 2018; see also Bougher 2017; Dias and Lelkes 2022; Webster and Abramowitz 2017).

Drawing upon this literature, we explore the extent and nature of the partisan divide relative to other social divides and also evaluate the implications of coalition partnership for affective polarization in Europe. The context of contemporary Europe offers an opportunity to advance our understanding of partisan divides and affective polarization in a broad comparative perspective. According to Iyengar and Westwood (2015), the extent of the partisan divide is larger than that of other social divides because it lacks social norms that discourage discrimination and hostility. Also, political parties articulate and aggregate different interests along multiple dimensions of social divides, and thus partisan ties reflect certain constellations of multiple social identities, such as social class and religion, which constitute social cleavages (Bartolini 2005; Lipset and Rokkan 1967). Consistent with recent studies that highlight the role of the group composition of parties (Ahler and Sood 2018; Claassen et al. 2021; Iyengar, Sood, and Lelkes 2012; Orr and Huber 2020; Robison and Moskowitz 2019), Richardson (1991) suggests that European partisans whose parties are based on long-standing cleavages tend to show more stable and intense partisan affect than others. ⁵ Thus, in the context of Europe, whose party systems represent long-standing social cleavages, we expect the primacy of partyism (Hypothesis 1).

To assess this partisan divide, we present a systematic estimation of the causal effect of partisanship by using a novel experimental design that combines a conjoint analysis with decision-making games such as dictator games and trust games. The standard “feeling thermometer” questions measuring respondents’ evaluations of parties, on a scale from 0 (“cold and negative”), through 50 (“neither warm nor cold”), to 100 (“positive and warm”) can lead to biased estimations of affective polarization among ordinary partisans. Studies by Druckman and his colleagues show that individuals tend to think about elites rather than fellow citizens when answering questions about the other party and that they harbor more animus toward the other party’s elites than they do toward fellow citizens in the other party/ies (Druckman and Levendusky 2019; see also Harteveld 2021; Kingzette 2021). They thus argue that findings based on these measures can be both incorrect and illusionary. Moreover, given that partisanship can proxy for other social identities (Ahler and Sood 2018; Claassen et al. 2021; Iyengar, Sood, and Lelkes 2012; Robison and Moskowitz 2019; see also Achen and Bartels 2016; Berelson, Lazarsfeld, and McPhee 1954; Campbell et al. 1960; Green, Palmquist, and Schickler 2002) or convey information about issue positions (e.g., Abramowitz and Saunders 2006; Niemi and Jennings 1991; Orr and Huber 2020), it is questionable whether an observational approach can precisely identify or compare the extent of partisan divide relative to other social divides.

Combining dictator and trust games with a conjoint analysis, we measure the extent to which respondents like or trust fellow citizens of their parties (or dislike or distrust those from the other parties) compared to other social groups that constitute traditional social cleavages. ⁶ The simplicity and comparability of these games enable us to analyze systematic differences in biases across different contexts, whereas conventional survey questions are not always easily comparable across individuals who might differently understand questions involving abstract or complicated concepts. Our experimental design allows us not only to estimate and compare the relative effects of multiple identity attributes, but also to address social desirability bias and systematic survey misreporting on sensitive topics, which can generate critical problems with internal validity in studies of political identities or social divides (Horiuchi, Markovich, and Yamamoto 2021). Compared with the few existing experimental studies based on a limited number of countries, we employ a cross-national design covering 25 European democracies to evaluate the robustness of the findings about affective polarization and to strengthen their external validity.

Studies utilizing similar dictator and trust games present largely consistent evidence that out-group derogation, rather than in-group favoritism, contributes to affective polarization in the United States (see Carlin and Love 2018; Iyengar and Westwood 2015; Westwood et al. 2018). ⁷ However, the social psychology literature has generally concluded that intergroup discrimination is mainly the result of in-group favoritism rather than that of out-group derogation (see, for example, Allport 1954; Brewer 1979; Dovidio and Gaertner 2010; Hinkle and Brown 1990; see also Balliet, Wu, and De Dreu 2014). Moreover, unlike the majoritarian two-party system that lacks strong incentives for interparty cooperation, the proportional electoral institutions that characterize most European democracies tend to foster multiple political parties and encourage political compromise. This could generate different dynamics that change the relative contributions of in- and out-group sentiment to affective polarization. Multiparty systems encourage compromise and cooperation across party lines, what Lijphart (1999) characterizes as “kinder, gentler” politics, to form and maintain governing coalitions (Bassan-Nygate and Weiss 2022; Drutman 2019; Gidron, Adams, and Horne 2020). ⁸ These theoretical considerations lead us to hypothesize that affective polarization in Europe might not be mainly driven by out-party animus (relative to in-party favoritism) (Hypothesis 2). Our experimental design allows us not only to differentiate in-group favoritism from out-group derogation but also to evaluate this configuration in Europe.

While the conditions for party competition are largely constrained by the institutional environment, particularly the electoral system (e.g., Cox 1997; Duverger 1954; Katz 1979), its specific characteristics and dynamics are significantly influenced by the strategic interactions among political parties and their elites who pursue policy or seek office (König et al. 2022). We focus on this dynamic nature of interparty relationships. As discussed above, the creation and presence of coalitions are expected to serve as superordinate identities that redefine in-groups and out-groups and thus produce positive affect toward out-groups belonging to the same coalition (González et al. 2009; Hagevi 2015; see also Gaertner and Dovidio 2000). Moreover, as the information of coalition partnership helps citizens to perceive out-party members belonging to the same coalition as more similar, affective polarization is likely to decrease (Fortunato and Stevenson 2013; Lupu 2013). This theoretical expectation leads us to focus on party-level explanations of affective polarization. Accordingly, the partisan divide is likely to vary not only across different democracies, but also between different partisans. Taken together, we hypothesize that political parties’ coalition partnerships play a significant role in bridging the partisan divide (Hypothesis 3). Our experiment is designed to estimate the role of this interparty relationship, particularly coalition partnership, by experimentally manipulating the party affiliation of another player with whom the respondents interact. Moreover, our cross-national study covering 25 European democracies allows us to evaluate the role of contextual factors that shape the nature of party competition and thus affective polarization. Overall, we seek to offer a comprehensive and dynamic understanding of the nature of partisan divide and its relationship with coalition partnership.

RESEARCH DESIGN

Building upon existing experimental work (Carlin and Love 2013; 2018; Iyengar and Westwood 2015; Westwood et al. 2018; Whitt et al. 2021; see also Fowler and Kam 2007), we utilize two decision-making games, the dictator and trust games, to estimate the extent of affective polarization by measuring the difference between financial allocations made to co-partisans and out-partisans. Dictator and trust games have been widely used to address various research questions in many different fields including behavioral economics, social psychology, and political science (see Camerer 1997; Roth 1995). Although these two games are originally thought to tap other-regarding preferences (i.e., altruistic behavior or fairness concerns) and reciprocity (i.e., trust or trustworthiness), respectively, they have been used to capture the effects of various factors on “tastes for discrimination” or bias, particularly by measuring the amount of transfer made by subjects under a wide range of experimental conditions. We depart from prior work by combining these decision-making games with a conjoint analysis in a cross-national design covering 25 democracies.

We fielded the survey experiment including these games from the end of May to mid-August 2019, based on a sample of around 1,200 respondents per country, totaling 29,827 respondents in 25 European countries. Respondents were recruited by Dynata (formerly SSI), an independent survey company and the sample was roughly representative of the underlying national populations in terms of key demographic variables such as age and gender. Tables A1 and A2 in Section A of the Supplementary Material provide an overview of the representativeness of our cross-national sample.

The dictator game presents a simple one-shot setup in which respondents (considered as player 1) receive a certain number of tokens (10 tokens in this experiment) and are asked how many they would like to allocate to another person (introduced as player 2) described in a short profile. The trust game extends this basic setup, adding a stage in which respondents are told that the researchers will triple the number of tokens allocated to player 2, and that player 2 will have a chance to transfer some, none, or all of those tokens back to the respondent, player 1. Participants are asked to play three rounds of the dictator game and three rounds of the trust game. ⁹ For each round of the game, respondents are provided with a randomly drawn profile for player 2 that includes the information about that player’s group affiliations. After displaying each profile, we ask respondents how many of their 10 tokens they are willing to allocate to player 2, and how many they would like to keep for themselves. ¹⁰ While established studies using these games often focus on a single dimension of identity such as partisanship, or additionally include only a few dimensions to study their relative effects (e.g., Carlin and Love 2013; Iyengar and Westwood 2015; Westwood et al. 2018), we use a conjoint analysis design to randomly vary the profiles of player 2 along six dimensions simultaneously: gender, age, class, religion, nationality, and partisanship. Section B of the Supplementary Material describes our experimental design in detail.

This research design has several important benefits. First, the simplicity and comparability of the decision-making games enable us to analyze systematic differences in biases across different contexts, whereas conventional survey questions are not always easily comparable across individuals who might understand differently the wording of questions involving abstract or complicated concepts. A cross-nationally valid measure of affective polarization is necessary to ascertain the robustness and external validity of arguments about the primacy of partisanship. Nevertheless, existing studies using dictator and trust games have rarely exploited this potential benefit for cross-national analyses as they have covered only a limited number of countries. Our cross-national design also provides a valuable opportunity to explore the role of contextual factors in explaining notable variations in affective polarization across different democracies.

Second, our conjoint analysis allows us to estimate the effect of partisanship while controlling for other social groups that affect the dependent variable and might otherwise distort the estimate of the partisan divide. For example, respondents might make inferences about other characteristics such as social class or religion on the basis of party affiliations. Indeed, recent studies, such as Ahler and Sood (2018), show that these inferences about the group compositions of the parties affect partisan animosity (see also Claassen et al. 2021; Orr and Huber 2020; Robison and Moskowitz 2019), consistent with the so-called group account of partisanship (Achen and Bartels 2016; Berelson, Lazarsfeld, and McPhee 1954; Campbell et al. 1960; Green, Palmquist, and Schickler 2002). By explicitly providing randomized information about these social groups, some of which constitute traditional social cleavages, we decompose composite treatment effects and identify the net effect of partisanship by addressing the potential issue of “aliasing” (Hainmueller, Hopkins, and Yamamoto 2014) or “masking” (Bansak et al. 2021). To isolate the effect of party from other attributes that are perceived to be correlated with party, it is critical to include information on the full set of relevant attributes (Bansak et al. 2021; see also Dafoe, Zhang, and Caughey 2018).

Third, our design allows us to identify and compare the causal effects of multiple identity attributes simultaneously. Because the resulting estimates represent effects on the same outcome, they can be compared on the same scale to evaluate the estimated partisan divide relative to other important social divides or cleavages. By focusing on the same set of social identity dimensions (gender, age, class, religion, nationality, and partisanship), we seek to systematically assess and compare the main lines of conflict across 25 European countries. ¹¹

Fourth, our conjoint analysis also is a useful tool for addressing social desirability bias, or systematic survey misreporting on sensitive topics, which can generate critical problems in studies of political identities or social divides. In particular, it is important to consider that group-related attitudes and behaviors are constrained by social norms or pressures with respect to many social divides, such as gender, social class, and religion. ¹² A standard conjoint design does not directly ask respondents to state their attitudes about controversial topics. Instead, it seeks to estimate their attitudes indirectly through their evaluations of various profiles that randomly vary in multiple attributes. Accordingly, it is widely believed that attitudes estimated from conjoint analysis are less susceptible to social desirability bias than those obtained using other designs (Horiuchi, Markovich, and Yamamoto 2021).

Before conducting our main experiment of dictator and trust games, we asked a series of questions to elicit some demographic and identity-related information from the respondents. This structure provides greater experimental realism by allowing us to display player 2 profiles using the same items that respondents saw just before the experiment. Although the player 2 profiles the respondents get to see are fully randomized, we exclude certain implausible configurations (see Hainmueller, Hopkins, and Yamamoto 2014). For example, player 2’s partisan affiliation could not be displayed if player 2 was not a co-national because it would be implausible to assign player 2 an affiliation to a national party for which they were ineligible to vote. ¹³ Figure 1 presents an example of a randomly drawn profile for player 2 that could have been displayed to our respondents. Section C of the Supplementary Material shows the full set of attributes from which the player 2 characteristics are drawn, and Table C1 in Section C of the Supplementary Material provides their distribution. ¹⁴

Figure 1. Player 2 Profile Example

EMPIRICAL ANALYSIS

Primacy of Partisan Divide in Europe

Figure 2 displays our main results, which are the estimates from a hierarchical linear model with random intercepts at the respondent and the country level. ¹⁵ The coefficient estimates indicated by white dots represent the estimated difference in the number of tokens allocated for the out-group relative to the in-group (indicated by black dots) on each dimension, together with their 95% confidence intervals. Comparing the results from the dictator and trust games side by side, the findings are remarkably similar, consistent with the conceptualization of affective polarization as the tendency to like or trust members of one’s own group and to dislike and distrust the out-group.

Figure 2. Effects of Identity Attributes on Allocations to Player 2

Note: Pooled analysis. J = 29,827; N = 178,936. The figure illustrates estimates of the effects of the randomly assigned identity attributes on the tokens allocated to player 2. The bars capture 95% confidence intervals. The black dots without horizontal bars refer to reference categories (in-groups).

Overall, we find that partisanship has the largest effect, with the out-group (out-partisans) receiving approximately one out of 10 tokens less (dictator game: −0.99 tokens, trust game: −1.06 tokens) than the in-group (co-partisans). This is a sizable effect, corresponding to about 25% of the average number of tokens allocated to player 2 who belongs to the in-group on all the identity dimensions (dictator: 3.92 tokens, trust game: 4.06 tokens). This impact of partisanship also stands out in comparison to other identity attributes: the second-biggest effect is that of religion, with the out-group receiving roughly half a token less than the in-group (dictator game: −0.43 tokens, trust game: −0.51 tokens). These results confirm Hypothesis 1. Nationality and class have discernible effects: In the dictator games, foreigners receive 0.12 tokens less than co-nationals (trust game: −0.14). Also, people with a different socioeconomic status receive 0.12 tokens fewer tokens than members of the same class as the respondent (trust game: −0.13 tokens). Finally, age and gender produce relatively smaller differences. Specifically, discrimination by gender does not reach statistical significance, compared to other attributes (for full results, see Table D1 in Section D of the Supplementary Material). ¹⁶

Figure 3 shows that our finding about the primacy of partisanship is consistent across all 25 democracies. Although the primacy of partisanship is widely accepted, surprisingly little evidence has shown its strength relative to other social divides or the generalizability of its primacy across different countries. Despite some exceptions (see Bornschier et al. 2021; Carlin and Love 2018; Westwood et al. 2018), they focus on only a few group attributes in a limited number of countries. We present clear evidence that the impact of partisanship dominates other group attributes that constitute social cleavages based on individual-level experimental data in a broad cross-national sample. ¹⁷ On the other hand, the findings also show some cross-national variation in the level of discrimination based on partisan affiliations, which we will further examine below. In the dictator games, the difference between in- and out-group is estimated to range from 0.57 tokens in Ireland to 1.50 tokens in Hungary. In the trust games, the estimated difference between in- and out-group ranges from 0.52 tokens in Portugal to 1.46 tokens in Hungary. Additionally, although religion turns out to constitute an important social divide in Europe, the findings vary significantly across countries.

Figure 3. Effects of Identity Attributes by Country

Note: The figure illustrates estimates of the effects of the randomly assigned identity attributes on the tokens allocated to player 2. The bars capture 95% confidence intervals. The black dots without horizontal bars refer to reference categories (in-groups). Empty dots indicate the coefficient estimates for the out-group (relative to the in-group), based on the dictator games. Empty rectangles indicate the same estimates based on the trust games.

In-Group Love or Out-Group Hate?

As an attempt to further our understanding on the nature of affective polarization, analyses of the two underlying components of partisan discrimination, in-group favoritism and out-group derogation, have mostly been conducted in the context of the United States (Iyengar and Westwood 2015; Lelkes and Westwood 2017; McConnell et al. 2018). Whereas studies in social psychology have generally concluded that intergroup discrimination can be attributed to in-group favoritism rather than out-group derogation, many studies on affective polarization have focused on out-party animus rather than in-party favoritism as the key driver (e.g., Iyengar, Sood, and Lelkes 2012; Iyengar and Westwood 2015; see also Westwood et al. 2018). However, other evidence suggests that partisan bias is largely motivated by in-group favoritism in some situations (see, for example, Lelkes and Westwood 2017; McConnell et al. 2018).

Our experimental design allows us to identify the relative contributions of in-party favoritism and out-party aversion to affective polarization in Europe. For this purpose, we include a control group that does not receive information about the partisanship of player 2 but only other group attributes. From this neutral baseline, we can assess the relative contributions of in- and out-party sentiment to affective polarization. Using a similar strategy, Iyengar and his colleagues find that affective polarization in the United States is mainly driven by out-party derogation (Iyengar and Westwood 2015; see also Carlin and Love 2018; Westwood et al. 2018). We also control for the information about group attributes that respondents might use to speculate about the political outlook and partisan leanings of player 2 (Bansak et al. 2021; Hainmueller, Hopkins, and Yamamoto 2014).

Figure 4 presents the results for the pooled sample. ¹⁸ On the left, we present findings for the dictator games, with those for the trust games on the right. The first row presents the estimates for in-party favoritism and out-party derogation relative to the control group that does not have any information about player 2’s partisanship. The second row compares these effects according to their absolute size, allowing for an assessment of relative contributions of in-group favoritism and out-group animus among citizens to the overall amount of affective polarization. We first find that both in-party favoritism and out-party derogation contribute significantly to the discrimination patterns that we observe for partisanship. In comparison to the control group, respondents allocate significantly more tokens after learning that player 2 is affiliated with the party they identify with (dictator game: +0.52 tokens; trust game: +0.60 tokens). Similarly, after learning that player 2 is affiliated with an out-party, respondents on average allocate significantly less tokens (dictator game: −0.47 tokens, trust game: −0.45 tokens) (for full results, see Table D2 in Section D of the Supplementary Material). The results suggest that the effect of in-party favoritism is at least as large as the effect of out-party derogation. In the trust game, in-party favoritism is not only on a par with out-party derogation, but significantly larger (+0.15 [95% CI: 0.02, 0.27]).

Figure 4. In-Group Favoritism and Out-Group Derogation

Note: Pooled analysis. J = 18,925; N = 77,456. The first row illustrates estimates of the effect of the randomly assigned partisan attributes on the tokens allocated to player 2, along with 95% confidence intervals. The second row shows the absolute effect sizes in comparison, along with 95% confidence intervals.

Breaking down the results by country shows both this notable pattern and cross-national variation, as shown in Figure 5 (for the coefficient estimates of in-group favoritism and out-group derogation and their comparison in each country, see Table D3 in Section D of the Supplementary Material). For the dictator games, in-group favoritism differs significantly from zero in 20 of the 25 democracies (trust game: 22 out of 25), but out-group derogation is statistically significant in only 17 democracies (trust game: 17 out of 25). In terms of absolute size, the point estimates for in-group favoritism and out-group derogation are statistically indistinguishable in 23 of the 25 democracies in the dictator games. In 13 countries, in-group discrimination has a larger effect than out-group discrimination, although this difference is statistically significant only in one country (the Netherlands). We find similar results in the trust games, in which the extent of in-group favoritism and that of out-group derogation are not statistically distinguishable in 21 of the 25 democracies. In-group favoritism even trumps out-group derogation in three countries (France, Latvia, and Slovakia). Confirming Hypothesis 2, out-group derogation is not more consequential than in-group favoritism in Europe. ¹⁹ To account for this pattern and cross-national variation, we follow our identity-based understanding of partisanship and pay further attention to the conditions and characteristics of interparty competition. Political parties compete and cooperate by forming and running coalition governments in many of European democracies. Specifically, whether an out-party is or has been a coalition partner can influence partisans’ motivations to moderate their in-group favoritism or out-group derogation. This dynamic and relational understanding of affective polarization is what we turn to next.

Figure 5. In-Group Favoritism and Out-Group Derogation in 25 European Democracies

Note: The figure shows estimates for the effects that the randomly assigned partisan attributes have on the tokens allocated to player 2, along with 95% confidence intervals. In particular, it shows the distribution of in-group favoritism (circles) and out-group derogation (triangles) across countries. Countries are listed in order of the level of affective polarization. The gray vertical lines show the average allocation of tokens to in-group (solid lines) and out-group (dashed lines).

Partisan Divide and Coalition Partnership

From an identity-based perspective coalition partnerships can influence the way partisans categorize others into in- and out-groups. Consequently, whether political parties cooperate and compete by forming and running coalition governments can shape partisans’ motivations to develop in-group favoritism and out-group derogation, contributing to the extent and nature of affective polarization. To evaluate this expectation, we estimate multilevel models that estimate the effect of coalition partnership, along with a range of other individual-, party-, and institutional-level variables. To capture the effects of coalition partnership, we code two indicators of current coalition partnership using the Parlgov dataset (Döring and Manow 2021). For the equation predicting the number of tokens for co-partisans, we code an indicator for whether the party the respondent identifies with is currently in a coalition government with another party. For the equation predicting the tokens for out-partisans, we code an indicator for whether player 2’s randomly displayed party is currently a coalition partner of the party the respondent identifies with.

We also explore a set of institutional factors that could structure party competition and cooperation. First, prior studies have hypothesized that electoral institutions are an important factor influencing the level of affective polarization (Gidron, Adams, and Horne 2020). To evaluate the role of electoral systems, we include the (logged) average district magnitude in each country, using data from Bormann and Golder (2013). Following the prior empirical analysis, lower average district magnitude is expected to represent more majoritarian electoral systems, while higher magnitude more proportional ones. Second, our model includes the effective number of parties to evaluate the role of electoral competition (Laakso and Taagepera 1979) and popular expectations about its role in moderating affective polarization (Drutman 2019). Third, we consider and measure the overall level of polarization at the elite level by using the index proposed by Dalton (2008), in which polarization is aggregated from the vote-share weighted ideological distances between the parties. Fourth, we include a set of additional control variables. On the individual level, we control for standard demographic variables, age, gender, and education from our survey questionnaire. Education is measured as the age at which respondents stopped their full-time education. For respondents still in school, we use their current age. Education is an important factor that influences a respondent’s awareness of coalition partnerships and political knowledge about their institutional environment (Luskin 1990). We control for some standard factors on the country level, too, including population size and GDP per capita as a measure of wealth. Taking into account our experimental design, our models also include and control for the group attributes of player 2 (age, gender, social class, religion, and nationality) (for descriptive information about dependent and independent variables, see Table C3 in Section C of the Supplementary Material).

Table 1 presents the main findings from multilevel models with random intercepts for countries and respondents. Columns 1 and 2 present the findings for the dictator games, and columns 3 and 4 present the findings for the trust games, with each model separately focusing on co-partisans and out-partisans. Column 1 shows that being in a coalition government reduces the number of tokens allocated to co-partisans, which indicates reduced in-group favoritism. Column 2 shows that for out-partisans, conversely, coalition partnership leads to an increase in the number of tokens, reducing out-group derogation as well. Overall, we find that coalition partnership reduces affective polarization in the dictator games by reducing both in-group favoritism and out-group derogation. This finding is further supported by the consistent results from the trust games: being in coalition reduces in-group favoritism and out-group derogation, even though the latter effect fails to reach the conventional threshold for statistical significance.

Table 1. Results of Hierarchical Linear Models: Current Coalition Partnership

Note: Treatment components refer to controls for player 2’s social group attributes. SD indicates the estimate of the standard deviation of the random intercept. For full results, see Table D4 in Section D of the Supplementary Material. ⁺p < 0.1, *p < 0.05,**p < 0.01.

Compared to coalition partnership, which varies across rounds of the games, country-level characteristics do not substantially shape the allocation of tokens. District magnitude, the effective number of parties, elite polarization, wealth, and population size all have substantively small and statistically insignificant coefficients. Respondent-level characteristics such as gender and age, by contrast play a certain role, with women and older respondents generally allocating relatively less tokens.

To evaluate the substantive implications of these findings, we present the predicted number of tokens for each case in Figure 6. Depending on coalition status and the difference between allocations we simulate the allocation of tokens while holding all remaining covariates at their median values. Corresponding to the results in Table 1, the first column shows that the allocation of tokens declines from 4.67 for co-partisans when a party is not in a coalition government to 4.48 for co-partisans when a party is in coalition government (−0.20 tokens [95% CI: −0.29, −0.10]). The second column shows that the number of tokens for out-partisans increases from 3.43 when the party is not a coalition partner to 3.72 when it is a coalition partner (+0.28 [0.11, 0.46]). The comparison across columns suggests that while co-partisans and out-partisans are always treated differently, coalition partnership leads to a significant reduction in partisan discrimination. For the trust games, columns 3 and 4 show similar findings. Overall, these figures illustrate that the effect of coalition partnership on in-group favoritism and out-group derogation is substantively meaningful.

Figure 6. Effect of Coalition Partnership on In-Group Favoritism and Out-Group Derogation

Note: Visualization of the coalition effects reported in Table 1, showing the predicted allocation of tokens (dots) and 95% confidence intervals (lines). Predictions for a fixed profile (30-year-old, female, middle class, no religion, first round), with respondent- and country-characteristics held at median values.

Returning to Table 1, compared to the findings at the party-level, institutional factors at the country-level hardly matter for in-group favoritism and out-group derogation. Average district magnitude, the effective number of parties, and elite-level polarization exert substantively small effects that are indistinguishable from zero. This result supports our party-level explanation of affective polarization in Europe. ²⁰ Coalition partnership reduces partisan in-group favoritism and out-group derogation, contributing to a reduction in affective polarization.

One drawback of this dummy indicator for coalition partnership is that it does not capture the history of coalition partnerships, which could also influence partisans’ recategorization of in-groups and out-groups. To address this concern, we explore an alternative, continuous operationalization of coalition experience, based on the notion that groups build collaborative relationships over time. To reflect this dynamic aspect, we code the number of years of experience that each party participated in a coalition government, again using the ParlGov dataset (Döring and Manow 2021). Covering the time frame since 2000, we capture two aspects of coalition experience. ²¹ For the models about co-partisans, we count the number of years each respondent’s party spent in coalition governments. For the models about out-partisans, we count how many years each respondent’s party spent in coalition governments with the party randomly displayed for player 2 (for more details, see Figure C1 in Section C of the Supplementary Material). We expect these considerations to shape the level of in-group favoritism and out-group derogation.

Table 2 presents the results, which are consistent with the current indicator of coalition partnership reported above. In particular, coalition experience reduces the number of tokens that co-partisans receive, whereas it increases the number of tokens for out-partisans. Consistent with the previous findings, therefore, coalition experience reduces both in-group favoritism and out-group derogation, thereby diminishing affective polarization. Overall, consistent with Hypothesis 3, the results confirm our theoretical expectation more clearly and consistently than the results considering current coalition partnership.

Table 2. Results of Hierarchical Linear Models: Coalition Experience

Note: Treatment components refer to controls for player 2’s social group attributes. SD indicates the estimate of the standard deviation of the random intercept. For full results, see Table D5 in Section D of the Supplementary Material. ⁺p < 0.1, *p < 0.05, **p < 0.01.

Again, we visualize the substantive importance of these effects in Figure 7, showing the predicted allocation of tokens and 95% confidence intervals. For the dictator games, coalition experience reduces the allocation of tokens for co-partisans from 4.69 to 4.33 (−0.35 [95% CI: −0.52, −0.19]), whereas it increases the allocation for out-partisans from 3.39 to 4.02 tokens (+0.63 [+0.38, +0.88]). We find similar results from the trust games. Taken together, these findings indicate that accumulated coalition experience over time reduces the extent to which partisans like and trust co-partisans as well as the extent to which they dislike and distrust out-partisans. In this way, coalition experience reduces affective polarization by being an important manifestation of interparty cooperation.

Figure 7. Effect of Coalition Experience on In-Group Favoritism and Out-Group Derogation

Note: Visualization of the coalition effects reported in Table 2, showing the predicted allocation of tokens (lines) and their 95% confidence intervals. Predictions for a fixed profile (30-year-old, female, middle class, no religion, first round), with respondent- and country-characteristics held at median values.

ROBUSTNESS CHECKS

We perform a series of robustness checks to check the sensitivity of our results to alternative model specifications that control for some competing explanations. All results can be found in Section F of the Supplementary Material.

First, ideological proximity between parties might also be an important factor for party competition and cooperation. On the one hand, ideological proximity is one of key factors to predict coalition formation (Martin and Stevenson 2001; see also Axelrod 1970). On the other hand, it is also often hypothesized to influence affective polarization (e.g., Gidron, Adams, and Horne 2020). ²² We use data from the Comparative Manifesto Project, focusing on the general left–right dimension, which traditionally structures party competition in Europe. We calculate the proximity between the respondent’s party and the party displayed in the experiment. Section F.1 of the Supplementary Material presents the corresponding results tables. For both current coalition partnership and coalition experience (Table F.1 in Section F.1 of the Supplementary Material), the results remain very similar.

Second, partisan sorting is considered to be an important driver of affective polarization (Harteveld 2021; Levendusky 2009; Mason 2015; 2018; Reiljan 2020). We include a Partisan-Ideological Alignment score that subtracts the respondent’s ideological distance to the in-party (ranging from 0 to 10) from their average ideological distance to out-parties (ranging from 0 to 10). This measure produces a scale that ranges from −10 to +10, with −10 indicating that the distance to one’s own party is much larger than the average distance to the other parties (weak alignment or sorting) and +10 indicating that the average distance to the other parties is much larger than the distance to one’s own party (strong alignment). Values close to zero indicate intermediate levels of alignment. The analyses in Section F.2 of the Supplementary Material show substantively identical results.

Third, we check an alternative measure for country-level polarization, adopting a measure introduced by Maoz and Somer-Topcu (2010) to capture bipolarization in multiparty democracies. This measure of bipolarization follows the idea that affective polarization in multiparty democracies often involves camp-building (Wagner 2021). As Section F.3 of the Supplementary Material shows, this alternative operationalization does not alter our main findings.

Fourth, we examine the possibility that some contextual factors can influence the configurations of political divides and affective polarization. We consider the idea that economic inequality can play a role. We test this in our data by including the Gini-coefficient of inequality. Also, considering the recent migration crisis in Europe and its implications for political divides, we also include country-level information about immigration in 2018 from the Eurostat data, which is the most recent available at the time of our survey. As Sections F.4 and F.5 of the Supplementary Material show, however, these considerations do not alter our findings about the importance of coalition partnership for the level of affective polarization.

CONCLUDING REMARKS

Our study estimates and compares the extent of partisan divide relative to other social divides and evaluates the nature of affective polarization by exploring how coalition partnership shapes in-group favoritism and out-group derogation. We combine decision-making games such as dictator and trust games with a conjoint analysis to increase the validity of our cross-national experiment that manipulates the partisan affiliations of citizens (rather than elites or parties) along with other group attributes in 25 European democracies.

We find that the magnitude of partisan divide is substantively larger than any other key social divide in almost all democracies in our sample. Moreover, according to our analysis, affective polarization in Europe is not mainly attributable to out-group derogation compared to in-group favoritism. And the variations we find in the relative contributions of in-group favoritism and out-group animosity across countries can be explained by the coalition partnerships which characterize the conditions of interparty competition in Europe. Thus, our findings contribute to a more dynamic understanding of affective polarization compared to alternative accounts that focus on relatively fixed individual-level traits or institutional conditions. They can also offer further insight to the ongoing discussion in searching for potential remedies of affective polarization.

Some caveats to our findings need to be considered. Although we focus on a common set of group attributes that constitute social cleavages, specific contexts, and group attributes that are more relevant to individual countries should be taken into account for a more nuanced understanding of certain cases. For example, ethnicity might be another potentially important group attributes in some European democracies. However, the classification of ethnicity is more complex than that of other groups such as age, gender, and social class, and thus it was difficult to include it in our cross-national analysis. Moreover, future research can advance our understanding of affective polarization by further exploring the possible interactions among various contextual and individual factors that shape the nature of partisan divide. Further work can be conducted, for example, to investigate how other characteristics of interparty relationship and identity attributes can influence affective polarization and its underlying mechanisms. A more thorough exposition that takes into account these considerations is beyond the scope of this paper, but our research design could be useful for investigating other unresolved questions that remain for future research.