Divided by Europe: affective polarisation in the context of European elections

Abstract As European integration has become politicised over the last several decades, scholars have paid keen attention to the role of identity in shaping political conflicts and contestation in Europe. This article investigates the microfoundation of the political divide over European integration by building on and extending theories of social identity and recent studies on affective polarisation. Large-scale survey experiment data from 25 European Union member states are analysed to explore the extent of divides over European integration. This analysis shows a statistically significant and substantively large magnitude of the divide based on European identity. Moreover, results show that this divide over European integration deserves particular attention as it is largely driven by out-group animosity, rather than in-group favouritism. Lastly, results show that electoral context, such as electoral salience, involvement in elections, elite polarisation, and the strength of Eurosceptic parties, influences the intensity of the European divide.

scholars who argue that the 'giant' of European politics is still 'sleeping' (van der Eijk and Franklin 2004;Green-Pedersen 2012), or those who claim that 'European integration has remained a project by political elites' (Koopmans 2007: 205;Statham and Trenz 2013). However, there have been relatively few empirical studies on whether and the extent to which a collective identity arising from European integration shapes political conflicts and contestation in contemporary Europe.
Building on and extending the theories on social identity and recent studies on affective polarisation, we seek to offer a new perspective and approach to address the longstanding debate around conflicts and contestation over European integration. Conceptualising affective polarisation as the tendency of people to like or trust members of their own group and dislike and distrust those from the other group, scholars have paid keen attention to increasing hostility and distrust across party lines across democracies today (see for example, Iyengar et al. 2012;Iyengar and Westwood 2015;Mason 2015Mason , 2016Westwood et al. 2018;Gidron et al. 2020;Reiljan 2020). We expand the scope of this existing literature to evaluate the extent and nature of conflicts and contestation over European integration by utilising a new research design in a broad cross-national context. Specifically, we design a large-scale survey experiment to investigate the extent of political conflict over European integration. By using dictator games, we measure and analyse bias generated by European identity as the difference between financial allocations to in-groups and out-groups. For each round, we ask respondents to specify the number of tokens they would like to allocate to another player we randomly display by manipulating the information about his or her European identity while varying other characteristics. We embed this experiment in large-scale online surveys fielded in 25 EU member states. This experiment allows us to systematically assess the impact of European identity at the individual level and to compare the extent of political conflicts across countries. We find a statistically significant and substantively large magnitude of the divide based on European identity. Our results also show that this divide over European integration deserves particular attention as it is largely driven by out-group animosity, rather than in-group favouritism.
As cleavages are rooted in the durable structure of political conflicts, political contestation based on these cleavages is expected to change only as a punctuated process that arises from external shocks that constitute critical junctures. However, existing literature also stresses the process in which specific conflicts win a dominant position in public debates and political space (Schattschneider 1960). Elections and campaigns play a significant role in this process. Political parties and other political actors contribute to activating political identities and conflicts by providing information and mobilising citizens (Michelitch and Utych 2018;Singh and Thornton 2019), in particular in high-stake elections (Hernandez et al. 2020). Existing studies also stress the importance of elite polarisation (Rogowski and Sutherland 2016;Carlin and Love 2018;Gidron et al. 2020) and the role of Eurosceptic parties in politicising issues of European integration (Kriesi 2007;Kriesi et al. 2008;de Vries and Edwards 2009;de Vries and Hobolt 2012;Hobolt and de Vries 2015). Using hierarchical models, we find that the divide based on European identity is intensified by the electoral context. In addition to electoral salience, involvement in election, elite polarisation and the strength of Eurosceptic parties play a significant role in shaping the divide over European identity. Taken as a whole, our study contributes to further understanding of the microfoundation of the political divide over European integration by building on and extending theories of social identity and recent studies on affective polarisation.

Conflict, cleavage and affective polarisation: the social identity approach
Since Schattschneider (1960) highlighted that political conflict is a key ingredient of democratic politics, scholars have sought to understand when political conflicts arise and how they shape the way democracies work (Lipset 1959;Dahl 1967;Rabushka and Shepsle 1972). In particular, Lipset and Rokkan (1967) focus on the concept of cleavage, a specific structure of political conflict that shapes political space and party competition. This so-called cleavage theory stresses how the configurations of contemporary European party systems are linked to much earlier social conflicts. According to this theory, traditional divisions formed around social class, religion, and region had been 'frozen' and European party systems still reflected the structure of social divisions that had existed in the early 20th century.
Following this literature, scholars have debated whether traditional cleavages have declined and whether a new cleavage has emerged in Europe over the last decades (see for example, Bartolini and Mair 1990;Franklin 1992;Inglehart 1977Inglehart , 1985Inglehart , 1990Kriesi 1998Kriesi , 2006Kriesi , 2012Evans 1998;Hooghe and Marks 2009). According to Hooghe and Marks (2009), after decades of a 'permissive consensus' , European integration has become the object of intensified conflicts in party competition, elections, and public debates. A wealth of studies has followed to explore the nature of conflict structures that underlie the political space brought by European integration.
Our study aims to investigate whether and to what extent this new cleavage, which scholars call transnational (Hooghe and Marks 2018), integration-demarcation (Kriesi et al. 2008), or nomadic vs. standing (Bartolini 2005), 1 is 'socially embedded' in the everyday life of ordinary citizens. 2 It is Bartolini and Mair (1990) who stress the importance of a collective identity as an essential component to constitute a cleavage, along with a socio-structural basis and a particular political organisation that articulates the common interest of the group (see also Kriesi 1998). 3 All cleavage politics is, at least in part, identity politics, as cleavages depend on groups with a stable and shared sense of identity that are organised into politics by parties (Ford and Jennings 2020).
A social identity approach provides an important theoretical foundation to study the nature of political conflicts and cleavages. By identifying with a certain group, people divide the world into in-group and out-group. This identification leads to in-group favouritism and bias (Tajfel 1981;Tajfel and Turner 1979), the tendency to view in-group members positively, and out-group members negatively. Thus, political divides are rooted in the very nature of how people identify with political groups (Iyengar et al. 2019). However, people hold multiple identities that vary in their relevance and significance across different situations. Accordingly, this identity-based understanding of political cleavages directs our attention to how in-group and out-group relationships form along the multiple dimensions of the conflict structure. While recent scholarship has focussed on the political divide over political parties (see, for example, Iyengar et al. 2012;Iyengar and Westwood 2015;Mason 2015Mason , 2018Westwood et al. 2018;Reiljan 2020;Gidron et al. 2020), 4 there have been relatively few systematic studies that expand the scope of this research to explore other important dimensions of political conflicts and different contexts. One exception is the study by Hobolt et al. (2021). They suggest that affective polarisation is not unique to partisanship and propose so-called 'opinion-based' affective polarisation in the context of Brexit by analysing the specific case of the UK. 5 Compared to this study, we focus on the divide over European integration that encompasses the political space across different EU member states. Moreover, consistent with the concept of identity as a deep attachment to an entity, more than just approving current EU or national policies (see, for example, Dalton 2021), we evaluate and compare the extent of divide over European identity and other divides that constitute political cleavages in the EU member states. 6 The cross-national analysis that covers most of the EU member states enables us not only to evaluate the robustness of the findings but also to investigate the role of contextual factors.
Drawing upon the existing literature on a new cleavage and political conflict around European integration, we hypothesise that there exists a statistically significant and substantively large magnitude of divide based on European identity (Hypothesis 1). In particular, existing literature emphasises that European integration has become a highly politicised issue for party competition, elections, and public discourse in Europe, to an extent to which this transformation can be viewed as a critical juncture that has given rise to a new cleavage (Evans 1998;Kriesi 1998;Kriesi et al. 2006Kriesi et al. , 2012Bartolini 2005;Marks 2009, 2018). Marks (2009, 2018) claim that whether individuals see themselves as European is at the heart of this divide (Kuhn 2019). 7 However, political divides around a new cleavage may originate in multiple sources that are closely linked together: whether they are from a conflict between winners and losers of globalisation (Kriesi et al. 2008(Kriesi et al. , 2012, from the emergence of new values (Inglehart 1977(Inglehart , 1990Kriesi et al. 2006), or from conflicts around territorial identities Marks 2009, 2018). Debates about the mechanisms underlying a new cleavage are difficult to address from prior research based on observational data. Our research design based on randomised experiments allows us to disentangle the magnitude of the divide around European integration, estimated by the unique effect of European identity. Also, compared to exiting studies, we evaluate the extent of this European divide relative to other political divides including traditional cleavages (such as social class and religion) and the partisan divide that recent studies have highlighted.
Moreover, our social identity approach enables us to explore the nature of the European divide by evaluating the relative contribution of in-group favouritism and out-group derogation. 8 Compared to other democratic countries with majoritarian institutions, European institutions that encourage multiparty systems tend to foster political compromise and cooperation across party lines along the traditional left-right dimension. Thus, partisan divisions in Europe may be prompted more by in-group favouritism. However, this may not be the case in the dimension where mainstream parties are challenged by Eurosceptic parties of both the radical left and radical right (e.g. Taggart 1998;Hix 1999;Hooghe et al. 2002). Moreover, even when the question of European integration motivates people to divide the world into in-group and out-group, the in-group is likely to remain heterogeneous and divided by other lines of conflict. Hence, the common dislike of the out-group may be a greater motivator of affective polarisation than the orientation towards a (heterogeneous) in-group (Brewer 1999;Roccas and Brewer 2002). In the context of contemporary Europe, we therefore expect the division over European integration to be primarily driven by out-group derogation rather than in-group favouritism (Hypothesis 2).
Recent empirical evidence suggests that the politicisation of European integration is far from straightforward, as Hutter et al. (2016) characterise it by 'a patchwork of politicising moments' rather than a uniform trend (Kriesi 2016). Recent findings of substantial variability across space and time cannot be explained solely by 'static' individual-level and country-level determinants, or by variables of critical junctures. We argue that reported cross-country or temporal variations in contestation over European integration may be explained by electoral context where political actors including politicians and political parties activate political identities and increase the intensity of political conflicts (Hypothesis 3). Indeed, both the so-called postfunctional theory of European integration and the general approach taken by cleavage theory argue that structural tensions or conflicts become politically salient to the extent that they are consciously perceived by the groups involved and are organised into politics by political elites and parties (Hooghe and Marks 2009;Kriesi 2009). Elections are 'the climax of intense group competition over state resources and policy control that promote both in-group cohesion and out-group threat' (Michelitch 2015: 44; see also Schattschneider 1960).
Existing scholarship shows that elections and campaigns play a significant role in strengthening political identities that are useful for electoral mobilisation, and further in exacerbating tensions around those identities (e.g. Eifert et al. 2010;Dinas 2014;Iyengar and Simon 2000;Iyengar et al. 2012;Michelitch 2015;Michelitch and Utych 2018;Singh and Thornton 2019). Here we focus on two major processes at the level of both voters and political elites. At the level of political elites, scholars have emphasised the role of Eurosceptic parties (Kriesi 2007;Kriesi et al. 2008Kriesi et al. , 2012Hooghe and Marks 2009;de Vries and Edwards 2009;Hobolt and de Vries 2015;Hutter et al. 2016) in politicising European integration. As Kriesi (2016) phrases, it takes political entrepreneurs who are capable and willing to mobilise the latent structural potentials for the dimension of European integration to become politically and electorally relevant. And Eurosceptic parties have been shown to play this role in seeking to achieve electoral benefits and advance their strategic positions by highlighting political conflicts over Europe and challenging the pro-European consensus of mainstream parties. This consideration leads us to hypothesise that the European divide is likely to be intensified as the strength of Eurosceptic parties increases (Hypothesis 3a). Moreover, scholars suggest that elite polarisation can increase the level of affective polarisation among the public by leading people to perceive out-group threats and increased stakes in political competition (Carlin and Love 2018;Rogowski and Sutherland 2016;Gidron et al. 2020). In the context of European elections, polarisation among party elites over European integration is expected to drive people to like or trust members of their own group and dislike and distrust those from the other groups (Hypothesis 3b).
At the level of individual constituents, elections and campaigns are expected to serve to heighten the intensity of the European divide by several potential mechanisms, of which we examine the following: individuals' attention, electoral engagement/participation, and perceived salience. Exposure to the information provided by the electoral context and political engagement during this period are both well founded to engender and reinforce political identities including partisanship (Iyengar and Simon 2000;Iyengar et al. 2012;Dinas 2014;Michelitch and Utych 2018;Singh and Thornton 2019). 9 According to Iyengar and his colleagues (2012), given that individuals hold multiple social identities, the extent of in-group/out-group bias depends on the salience of the group identity, which is influenced by the characteristics of the information environment. European elections increase the volume of political information and mobilising frames to which voters are exposed, in particular with respect to conflicts over European integration, and thus greater attention to political campaigns are likely to strengthen their in-group favouritism and/or out-group hostility (Hypothesis 3c). Moreover, voters who engage in political conversation, campaign activities, and turn out to vote are more likely not only to get exposed to this political information, but also in turn to take sides and increase identification with their in-group (Hypothesis 3d). Lastly, existing studies suggest that the perceived stakes of European elections can intensify the conflict over Europe. According to so-called realistic group conflict theory (Blumer 1958;Bobo 1999;Sherif et al. 1961), electoral competition alone may induce individuals to strengthen their collective identity for or against Europe. This literature suggests that group membership is politically consequential to the extent that tangible group gains and losses drive group members' political decisions (Huddy 2013). Thus, we hypothesise that this contextual effect of elections is likely to be more pronounced when citizens perceive the electoral competition to have significant consequences on their lives (Hypothesis 3e). Taken together, we hypothesise the electoral context, such as electoral salience, involvement in election, elite polarisation, and the strength of Eurosceptic parties, should influence the intensity of the European divide.

Research design
We embed our experiment to study the divide over European identity into large-scale cross-national surveys designed to explore the relative effects of multiple dimensions of identity structure in Europe (Hahm et al. 2022). In the context of the European Parliament election 2019, we fielded an online survey based on a sample of around 1200 respondents per country, totalling to 29,827 respondents in 25 EU member states, including all but the three smallest countries: Cyprus, Luxembourg, and Malta. 10 The field time started just before the election in the end of May and lasted until the end of July 2019. The core of the survey questionnaire was a survey experiment featuring dictator games, where respondents (who are assumed to be 'Player 1′) were told that they hold 10 tokens and asked to specify the number of tokens they would like to allocate to a randomly generated 'Player 2′, and how much they would like to keep for themselves. This dictator game experiment measures group dislike or discrimination, which constitutes a social divide, as the amount an individual respondent (Player 1) sends to another player (Player 2) and the variation in this allocation is attributed to characteristics of Player 1 or/and those of Player 2 (e.g. Fershtman and Gneezy 2001;Fong and Luttmer 2011). 11 In particular, by manipulating the profile information of Player 2, this approach has been expected to serve to control for competing explanations of group bias but also to increase the generalisability of the findings across different contexts (see an overview by Iyengar et al. 2019).
We utilise this decision-making game experiment that allows for estimating the unique treatment effect of European identity. Compared to existing scholarship that relies on traditional survey items, which typically utilise self-reported evaluations of their attitudes, our research design enables us to better estimate the impact of European identity and to quantify the extent of divide over Europe. In each round, we asked people to allocate tokens to Player 2, whose characteristics we manipulate the information about Player 2's nationality (as a co-national, an EU national or a non-EU national) while varying other attributers including social class, religion and party affiliation. However, European identity concerns a deep attachment to the EU, more than just the information about the nationality, consistent with the existing literature (see, for example, Kuhn and Nicoli 2020;Dalton 2021). Thus, importantly for this study, in a subset of the dictator games where we display Player 2 as an EU national, we randomly display one piece of additional information about Player 2's attachment to the EU (whether 'Player 2 feels that he/she is a citizen of the European Union' or 'feels that he/she is not a citizen of the European Union') as well as the control condition that does not provide any information about his/her EU attachment. The inclusion of a control group lacking any information about Player 2's EU attachment can provide a baseline from which we assess bias for in-group and out-group members. We match this information with the respondents' attachment to the EU, which we measured on a five-point scale. Before the experiment, respondents were asked about their attachment to the EU and they could indicate their level of attachment on a scale between 'Very attached' (4) and 'Not at all attached' (0). We interpret the presence of a minimum of EU attachment ('Not very attached' (1) or higher) to mean that a respondent harbours some subjective feeling as a citizen of the EU, correspondingly treating Player 2 as in-group if he or she feels to be a citizen of the EU and as out-group if he or she 'feels that he/ she is not a citizen of the EU' . Respondents who indicate no EU attachment 'at all' are treated conversely. 12 The dimensions we use to describe Player 2 correspond the information which we asked respondents to provide before the survey experiment, which is expected to increase the realism of our survey design. Respondents were asked to play three rounds of the dictator game.

Divided by Europe
We evaluate the division over European integration by examining the extent to which identity attributes underlying cleavages lead to meaningful discrimination of out-groups relative to in-groups, measured by the allocation of tokens. We juxtapose our main result based on the EU attachment experiment with the previous findings by Hahm et al. (2022). Figure 1 represents the extent of these divisions over different identity attributes. The height of the bars represents the estimated difference in tokens between out-group and in-group, together with confidence intervals. Higher bars indicate stronger divides, as indicated by fewer tokens for the out-group compared to the in-group on each dimension.
Before we discuss the main results about EU attachment, some of the findings about nationality deserve attention (refer to the right panel in Figure 1). Specifically, respondents allocated about 0.28 tokens (out of 10 tokens) less to non-EU nationals than to co-nationals. The divide between co-nationals and non-EU nationals, measured by the difference in tokens allocated to out-group (non-EU nationals) relative to in-group (co-nationals), is comparable to the divide over religion. However, we find no statistical significance regarding the divide between co-nationals and EU nationals. This may suggest the presence of a common identity that reduces the bias towards fellow EU citizens even if they are not co-nationals. Additionally, class, one of the major traditional cleavages along with religion, turns out to have a statistically significant but relatively moderate effect.
The dark bar on the left panel in Figure 1 shows the size of the divide based on European identity, suggesting the potential mechanism generating the puzzling findings about nationality. According to the results, there was a significant penalty of 0.39 tokens for opposing groups with respect to EU attachment. Whether respondents share the same EU attachment has a significant influence over how people perceive others (Hypothesis 1).
The direct comparison between nationality and EU attachment shows that the divide based on EU attachment is comparable in size to the divide between co-nationals and Non-EU nationals. This comparison suggests that whether people share a sense of European identity, rather than nationality per se, determines bias against EU nationals.
Conducting the same analysis for each country individually, we see notable consistency of this pattern ( Figure A2 in the online appendix): As to nationality, the aggregate pattern shown in Figure 1 is supported in a large majority of countries. The only exceptions to the general pattern are Greece, Ireland and UK, where Brexit was scheduled to take place at the end of 2019, the year the survey was fielded. In these three countries, bias against EU nationals is estimated to be slightly higher than bias against non-EU nationals, even though these differences are indistinguishable statistically. In the remaining 22 countries, the pattern is consistent with the pooled analysis, where the bias towards non-EU nationals is at least as large as the bias towards EU nationals. Importantly, the cross-national comparison demonstrates that EU attachment is a meaningful source of bias in a broad set of countries, and thus is not  Hahm et al. 2022. the figure illustrates estimates of the effects of the randomly assigned identity attributes on the tokens allocated to player 2. the bars capture the estimated extent of divides, measured in the number of tokens that are withheld from the out-group relative to the in-group on each identity attribute. For nationality, the in-group is co-national. longer bars indicate larger gaps between in-group and out-group. lines indicate 95 percent confidence intervals.
primarily driven by a small set of influential cases. We find a significant divide based on European identity in all regions of Europe. 13 So far, we showed that the question of EU attachment constitutes a meaningful divide for people across Europe, motivating the division of the world into in-group and out-group as well as significant bias towards the out-group. To further explore the significance of this European divide, we examine how it is relevant to political divides along partisan lines. Our earlier analysis (Hahm et al. 2022) showing the relative effects of identity attributes confirms existing studies documenting the primacy of partisanship (Iyengar et al. 2012;Iyengar and Westwood 2015;Westwood et al. 2018). Here, we investigate whether and to what extent bias towards out-partisans is conditioned on the European divide. As presented in Figure 2, while out-partisans generally receive fewer tokens (indicated by a significant penalty for out-groups), out-partisans that share the respondent's EU attachment do not experience as much bias (a significant penalty of 1.04 tokens) as out-partisans that do not share the respondent's EU attachment, either (a significant penalty of 1.35 tokens). Again, this pattern can be confirmed in the vast majority of countries ( Figure A3 in the online appendix). 14 Figure 2. partisan divide conditional on eu attachment. Notes: pooled analysis. J = 14,020; n = 20,629. the figure illustrates estimates of the effects of partisanship on the tokens allocated to player 2. the bars capture the estimated extent of the partisan divide, measured in the number of tokens that are withheld from the out-partisans relative to co-partisans. longer bars indicate larger gaps between in-group and out-group. lines indicate 95 percent confidence intervals.
Our experimental design enables us to explore the nature of the political divide over European integration, by identifying the relative contribution of in-group favouritism and out-group derogation. We identify these two sources of bias by exploiting the randomisation scheme, where some respondents receive no information as to Player 2's EU attachment. Relative to this control group, we can separately estimate whether in-group status results in allocation of more tokens; whether out-group status results in the allocation of less tokens; and compare the relative size of these two effects. Figure 3 shows that out-group derogation in general dominates in-group favouritism with respect to the European divide (Hypothesis 2): the divide materialises in that respondents allocate roughly 0.10 tokens more to Player 2 when they learn that he or she shares the respondent's EU attachment. Correspondingly, respondents allocate 0.29 tokens less to Player 2 when they learn that he or she falls into the out-group. Also, while the size in these differences varies across countries, this aggregate pattern can be shown to hold in the vast majority in our sample of 25 EU countries ( Figure A4 in the online appendix). Notes: pooled analysis. J = 29,822; n = 83,232. the figure illustrates the decomposition of the divide based on european identity into separate estimates for in-group favouritism and out-group derogation. the bars capture the estimated extent of divide, measured by the difference in tokens allocated to in-group and out-group relative to a control group where we do not display whether player 2 feels that he or she is a citizen of the eu. lines indicate 95 percent confidence intervals.

Electoral context and the European divide
Lastly, we investigate how electoral factors influences political divides based on European identity by utilising the context of the European Parliament election 2019 where the survey experiments were implemented. We focus on both individual-level factors (such as an individual's perception of electoral salience, attention to election, electoral engagement, and electoral participation) and country-level factors (such as elite polarisation about European integration and the strength of Eurosceptic parties) in the electoral context, in particular. By using hierarchical linear models, we estimate whether and how the allocation of tokens is influenced by these individual-level and country-level factors. Specifically, we evaluate how these factors intensify or reduce the divide based on European identity. This conditioning effect is captured by the interaction effects between the experimental treatments and electoral variables of interest.
On the individual level, we construct measures to operationalise the salience of the European Parliament election for the respondents, as well as their attention to, engagement with, and participation in the election. For electoral salience, we use a five-point scale on which respondents answer the question 'How important is the outcome of the upcoming election to you personally?' , with answer categories 'Very important' (1) to 'Not important at all' (5). We turn this scale around such that higher values indicate higher levels of importance. The resulting measure ranges from low (0) to high (4) salience. We standardise this scale onto the unit interval (0 to 1) to permit easier comparisons across variables.
We construct a measure of an individual's attention to the election using our survey question about the sources they used to get information about the 2019 European election campaign. Respondents could indicate the type of media such as TV, newspaper, radio, internet sources, among others. They could also indicate that they 'have not heard anything about the election campaign from any of these sources' . We code electoral attention as 1 if respondents indicate one or more sources, and 0 if they indicate that they had not followed the election campaign.
In order to construct a measure of electoral engagement, we utilise the survey item where respondents could indicate whether they engaged in personal conversation during the European Parliament election campaigns. We create a dummy variable for electoral engagement, which takes the value of 1 if respondents had such conversations and takes the value of 0 otherwise.
To construct a measure of electoral participation, we use the respondent's stated intention, just before the election, to turn out to vote. In our question, respondents could indicate how likely they are to cast a vote ('Certain not to vote' , 'Not likely to vote' , 'Might vote' , 'Likely to vote' , 'Certain to vote') along with options to indicate non-eligibility or that they already have voted (via mail-in ballots). From this information, we create a dummy of election participation, indicating as 1 those respondents who said they are 'likely' or 'certain to vote' , or have already voted. Otherwise, respondents are coded as 0.
On the country level, we first construct the measure of elite polarisation, using the most recent party manifesto positions on European integration (Volkens et al. 2020). We code party positions by applying the logit transformation proposed by Lowe et al. (2011) on the relevant items in the CMP dataset (per108 and per110) in order to produce comparable estimates of party positions. Based on these positions, we compute party polarisation using the formula as proposed by Dalton (2008), which weights distances by vote shares of each party.
For the strength of Eurosceptic parties, we follow de Vries and Edwards (2009: 11) to define political parties as Eurosceptic if their position on the EU lies more than one standard deviation below the average in a given party system, utilising Lowe et al.'s (2011) positions on European integration. We subsequently sum up the vote share of parties that are classified as Eurosceptic by this standard. The resulting measure ranges between 0% (Croatia) and 49% (Hungary). 15 In addition to the contextual factors of interest, we also include a variable that measures the length of EU membership. The populations of old EU members may have developed a division between in-and out-group based on European identity to a greater extent than citizens of more recent EU members. Also, the Gini coefficient is a measure of economic inequality in the population which may worsen distributional conflicts among competing political groups (Gidron et al. 2020).
Lastly, we include a battery of controls. At the individual level, we control for standard variables used in electoral research: gender, age and education. For gender, we include an indicator for female respondents. For education, we use the age where respondents reported to have stopped full-time education. At the country level, besides population size, we control for GDP per capita as a measure of wealth and the unemployment rate. Table 1 provides an overview over the independent variables used.
The multilevel model predicts the allocation of tokens to Player 2 as a function of the distinction between in-and out-group based on European identity, individual/contextual-level variables that capture electoral context, and their interactions. Focussing on individual-level and country-level moderators of European divide, our goal of inference is to estimate coefficients for these interaction effects. The following model represents this setup: Here, r = {1,2,3} indexes the rounds of the dictator games, i indexes individual survey respondents, and c indexes countries. The coefficient β 1 estimates the effect of randomly assigning Player 2 to the out-group rather than in-group when it comes to the question of EU attachment on the allocation of tokens.
We hypothesise that this effect is moderated by individual-level factors M i and country-level factors M c . M i includes the individual-level variables such as electoral salience, attention to the election, electoral engagement, and participation. M c includes the contextual-level variables of interest, country-level measures of elite polarisation and the electoral strength of Eurosceptic parties as well as the length of EU membership and the Gini coefficient. Furthermore, Controls includes control variables at the level of experimental rounds (whether partisan information was shown), individuals and countries. We use a multilevel model using random intercepts given the structure of our data. 16 Table 2 presents the results from estimating this model. Core variables of interest at the individual and country level are standardised between 0 to 1, if they are not already binary indicators, and then centred on their mean values for easier comparison of the effects. 17 Figure 4 illustrate these interaction effects of electoral context, along with those of the length of EU membership and the Gini coefficient. The bars indicate the intensity of the European division at different levels of the individual-level and country-level moderators based on our analysis results. We find that EU attachment is a more relevant source of a divide among respondents to whom the 2019 European Parliament election was salient. Respondents who do not perceive the election as particularly salient withhold 0.26 tokens from people who do not share their EU attachment. However, respondents who do find it salient withhold 0.53 tokens. The difference of 0.27 tokens is significant at the 5% significance level (see Model 6). Also, we find electoral attention and engagement to make a meaningful difference, but these moderation effects are not statistically significant. On the other hand, we find that discrimination based on EU attachment is moderated significantly by political participation: respondents who do not turn out in the European Parliament election withhold 0.29 tokens from the out-group, whereas respondents who do turn out withhold 0.42 tokens. These findings suggest important variation among respondents in the relevance of a common EU identity. More precisely, this identity motivates the demarcation of, and bias against an out-group, and this motivation is amplified by factors related to electoral context: the salience of the 2019 European Parliament election for the respondent, as well as his or her level of electoral involvement and in particular, his or her electoral participation (Hypothesis 3c/e). This finding is supported by also taking into account the country-level characteristics of electoral context. We find that the elites' polarisation over European integration goes together with higher levels of the divide based on EU identity (Hypothesis 3a). Also, consistent with the existing literature, our results confirm the important role of Eurosceptic parties in intensifying the European divide, confirming our hypothesis (Hypothesis 3 b).
Additionally, we find that countries that became EU members more recently show lower levels of European divide. This is consistent with the argument that people in old EU member states have had more time to develop this division between in-group and out-group. Finally, the negative coefficient on the Gini coefficient suggests that more unequal countries show a greater divide based on European identity. However, these effects disappear when estimated jointly with country-level factors variables of the electoral context. 18 To summarise, at the individual level, involvement in the 2019 European Parliament election amplifies the European divide: people who perceive high electoral saliency and are involved in the election (in particular, through the path of electoral participation) discriminate others based on their European identity (Hypothesis 3c/e). At the country-level, elite polarisation on European integration and the presence of strong Eurosceptic parties are accompanied by higher levels of the European divide (Hypothesis 3a/b).  Note: *p < 0.1; **p < 0.05; ***p < 0.01.

Conclusion
In order to further understand the new conflict and political contestation around European integration, we combine the literature on cleavages with studies on social identity and affective polarisation. In particular, our approach allows us to investigate the microlevel basis of a new cleavage in contemporary Europe. By utilising decision-making games, we identify the unique effect of European identity and disentangle the extent of the division based on European integration from that of other mechanisms underlying the new multifaced cleavage that existing literature has discussed. Moreover, with this research design, we evaluate and compare the relative magnitudes of divisions over different identity attributes across countries. We find a statistically significant and substantively large magnitude of the divide based on European identity, which is comparable to the magnitude of the religious divide. While the impact of European identity does not compare with partisanship, it plays a significant role in narrowing or widening the partisan divide. Our results also suggest that this divide over European integration is attributed more to out-group hostility, rather than in-group favouritism. Lastly, using a large-scale cross-national analysis, we offer a new perspective to address the existing puzzle around the substantial variability in politicisation of European integration. Using hierarchical models, we show how the electoral context influences the intensity of the European divide. Compared to existing studies, we find that specific characteristics of electoral campaigns, such as electoral salience, electoral involvement, elite polarisation, and the strength of Eurosceptic parties, matter. Taken together, our study offers a new perspective to understand the structure of political conflicts in Europe.
On the other hand, several limitations of our study open up potential research questions for future research. First, while our experimental design enables us to estimate the size of the European divide, our results may fail to accurately predict the exact forms and levels of discrimination between actual groups or people. The transfer of tokens in the experiment may not necessarily translate into corresponding transfers of (financial) resources in the real world. Further work using realistic treatments can be done to ensure the practical relevance of our findings. Second, the allocation of tokens within our experiment might be affected by the level of abstraction at which the treatments are specified. In particular, out-national profiles of player 2 are not made more specific by reference to their specific nationality. Of course, some considerable heterogeneity may exist in how European respondents treat out-nationals, for example, between a 'Ukrainian' Player 2 and one from the Middle East or Sub-Saharan Africa. In the future, one might be interested in exploring this heterogeneity. Third, our measures to capture the specific characteristics of electoral context are observed rather than manipulated. Thus, to test the causal directions in how the European divide is related to the characteristics of electoral environment, our comparative findings can serve as a starting point for more focussed research designs that are capable of identifying causal effects of particular variables of the electoral context. Fourth, these relationships between the conflict over European integration and the features of elections and campaigns are likely to be moderated by varying characteristics of citizens, such as political interest and information. Future research could build upon our findings presented here and address the dynamic process underlying the European divide. Lastly, our study sets the stage for further research to understand better the development and the nature of this conflict structure in Europe by tracking its variability more systematically over a long period of time.
3. However, some seminal studies of electoral behavior (Tingsten 1937;Lazarsfeld et al. 1944;Berelson et al. 1954;Campbell et al. 1960) indeed provided a microlevel basis for earlier studies on cleavages and social conflict (Coser 1956;Dahl 1956;Lipset 1959;Rae and Taylor 1970). 4. See also Wagner (2021), Harteveld (2021aHarteveld ( , 2021b, and Hernandez et al. (2020) for recent scholarship on affective polarisation in a comparative perspective. However, most of these studies focus on affective polarisation over political parties, except for a few studies. 5. See also Bornschier et al. (2021) and Helbling and Jungkunz (2020) for studies on political cleavages with a similar emphasis on social identity. 6. See, for example, Iyengar et al. (2012) that contrast policy-based division with affect-based division. 7. See also recent studies on European identity published in special issues on Journal of Common Market Studies (e.g. Kuhn and Nicoli 2020) and European Union Politics (e.g. Clark and Rohrschneider 2021). 8. While existing literature in social psychology have largely concluded that in-group favouritism is more dominant, there is also evidence that partisan bias is prompted more by out-group hostility in the United States (Iyengar and Westwood 2015;Carlin and Love 2018;Westwood et al. 2018). 9. See also Hahm (2016) that highlights the role of institutional configurations that affect partisanship both at the mass and elite levels. 10. We conducted the survey on a sample collected by Dynata (formerly SSI).
This sample was selected to be broadly representative of the adult population with respect to age and gender in each of the 25 member states (see Tables A1 and A2 in the online appendix). 11. As described, we used the dictator games with hypothetical money and imaginary recipients. Recent studies in economics and psychology show some consistent evidence that subjects give similar amounts and discriminate between in-group and out-group to similar degrees in the hypothetical and incentivized dictator games (see Ben-Ner et al. 2008, 2009. Moreover, existing studies find that stakes tend to have little effect on average behavior and that games with larger stakes tend to generate data with less variance (Camerer and Hogarth 1999;Forsythe et al. 1994;List and Cherry 2008;Carpenter et al. 2005). Therefore, we expect that hypothetical dictator games would provide us with conservative estimates, which bolster our confidence in our findings. With respect to any potential ethical concerns, this study was reviewed and approved by the committee of the University of Mannheim and the data were collected in full compliance with the European General Data Protection Regulation. Before respondents started answering survey questions, they were informed that the survey is part of a scientific project through the informed consent document. The survey proceeded only after getting this consent. At the end of the surveys, we also debriefed them by stating that the content used in the experiments may not be based on facts. 12. We conduct several robustness checks based on different ways to categorize individuals with/without EU attachment. The results are consistent with the main analysis (see Figures A5 and A6 in the online appendix). 13. This finding compares with prior studies that show the variation in the salience of the conflict over European integration. For example, Hutter and Kriesi (2019) found a notable lack of politicization of European integration in Central and Eastern Europe. 14. We also replicate this study using trust games and find consistent results.
Results are available upon request. 15. There exists a moderately high correlation of 0.49 between the two variables, elite polarisation and the strength of Eurosceptic parties. Despite a conceptual distinction, we use these two variables separately in alternative models to empirically clarify the respective effect of each variable. 16. In online appendix, we run the single-level equivalent, with standard errors clustered by respondents, to check the robustness of this model (see Table A5). 17. The intercepts therefore can be interpreted as the number of tokens allocated to Player 2 if he or she is in the in-group on the question of European integration, under otherwise typical conditions. Depending on the model, we see that respondents would allocate roughly 3.77 tokens, on average, to their in-group (refer to Models 6 and 7). 18. Due to the available number of countries, we are careful in drawing definitive conclusions about the role of country-level factors. Thus, our findings emphasize that there is substantial variation, within and across countries, in the extent of political divisions based on EU identity. The level of bias varies meaningfully with variables of electoral context.