Careless whisper: Political elite discourses activate national identities for far-right voting preferences

While exclusionary national identities are widespread among Europeans, relatively few people vote for the far right in most countries. Thus, an exclusionary identity in many cases does not lead to voting for the most nativist types of parties. We explain this empirical puzzle by showing that these identities need to be activated to become behaviourally relevant. To this end, we analyse longitudinal comparative data of over 135,000 individuals across more than 26 years and 26 countries combining different survey programmes and manifesto data. We use latent class analysis to show that over half of respondents hold exclusionary conceptions of nationhood. Moreover, this type of national identity predicts voting far right. Using multi-level modelling and within-country estimators, we further demonstrate that this relationship is significantly stronger when a country's political elites across all parties become more exclusionary. Taking the activation hypothesis to the test in a European context, we conclude that the effect of national identity is conditional on its prior activation.


| INTRODUCTION
In the 2016 US election, Donald Trump put American nationalism and the threat of immigration at the top of his campaign agenda and succeeded.Specifically, the prominence of anti-immigration rhetoric supposedly activated a specific American national identity as an important driver of vote choice in this election (Bonikowski et al., 2021;Garand et al., 2020;Sides et al., 2019;Thompson, 2021).In the European context, similar arguments may be made regarding right-wing parties such as Rassemblement National (FR), Legal Nord (IT) or Alternative für Deutschland (DE), which argue for the (re-)nationalisation of politics in the face of increasing globalisation and migration flows.Picking up on this reasoning of latent national identities becoming behaviourally relevant through the discourses of the political elite, we test the activation hypothesis in the European context in a longitudinal comparative analysis.
Previous research has repeatedly found ethnically charged, exclusionary conceptions of nationhood to be associated with anti-immigrant sentiments (e.g., Helbling et al., 2016;Hjerm, 1998a;Kunovich, 2009;Trittler, 2017;Wright, 2011;Wright et al., 2012), which, in turn, are related to far-right voting (Arzheimer, 2008(Arzheimer, , 2009;;Arzheimer & Carter, 2006;e.g., Abou-Chadi et al., 2022).Paradoxically, though exclusionary national identities are consistently prevalent, right-wing voting is less frequent (see below) than one would assume given that right-wing parties are usually the most nativist and anti-immigrant.The key question is thus when national identity is especially salient.While few works concern the role of national identity in voting at all (Bonikowski et al., 2021;Garand et al., 2020;Lubbers & Coenders, 2017;Mader et al., 2021;Pesthy et al., 2021;Thompson, 2021), evidence for the mechanism is to our knowledge scarce.So far, analyses have been mainly limited to the American context that relied on Trump's ethno-nativist rhetoric as an activating factor, without explicitly testing for this relationship (Bonikowski et al., 2021;Garand et al., 2020;Thompson, 2021), to analyses of municipal-level data suggesting latent nativist potentials in Germany (Schulte-Cloos, 2022) or to aggregated survey data measuring public opinion, which increased the relevance of national identity in voting decisions (Lubbers & Coenders, 2017).Drawing on social-psychological reasoning and works on political messaging, we use contextual data and analyse the role of exclusionary political elite discourses in activating national identities in voting decisions.We argue that both national identity and exclusionary political elite discourses have the potential to influence far-right voting preferences but that the combination of the two is particularly potent.
Public opinion on immigration correlates with the way political elites frame and discuss immigration-related issues (Bohman, 2011; but see Bohman & Hjerm, 2016;Heizmann & Huth, 2021;Mitchell, 2021;Schmidt-Catran & Czymara, 2023).At the same time, mainstream parties can have a catalysing effect on accommodating radical rightwing issues and, in consequence, increase support for such parties (Krause et al., 2023;Lochocki, 2018).We combine both arguments by showing that the political climate in a country in general affects the likelihood of voting preferences for a far-right party specifically.Advancing works in the American context, we analyse the role of national political party elites that embrace exclusionary issues in activating national identities.Building on recent developments in national identity research (Bonikowski & DiMaggio, 2016, 2022;Eger & Hjerm, 2022a, 2022b;May, 2023), we analyse how specific conceptions of nationhood drive this relationship.
We test whether exclusionary political elites increase the link between individual exclusionary national identity conceptions and far-right voting preferences based on comparative longitudinal data over more than 25 years.To this end, we analyse a rich dataset that harmonises various multi-national survey programmes compiled by the ONBound project (Bechert et al., 2020) and manifesto data from the Manifesto Project Dataset (Volkens et al., 2021b) from 1995 to 2020.Our analysis combines latent class analysis (LCA) on the individual level and within-country estimators on the national level.We show that, first, exclusionary national identity is widely prevalent in many European countries and correlates with far-right voting.Second, voters are more likely to choose far-right parties when political elites in general become more exclusionary.Third, we find that the effect of national identity is conditional on its prior activation through political elites: The within-country effect of anti-immigrant and pronational discourses on far-right voting preferences is particularly strong among those with an exclusionary national identity.Put differently, those with a narrow image of their national in-group are significantly more likely to vote for a far-right party if national political elites are more exclusionary.We conclude that otherwise latent national identities are activated through political elites across parties, which thereby leads to perhaps unintended negative consequences for social cohesion, political systems and society at large.
We follow Golder (2016) and use the term 'far right' as umbrella term, including 'radical' and 'extreme' right-wing parties.Individual-level predictors of support for such parties include lower education, economic vulnerability and anti-immigrant views (cf.Arzheimer, 2009;cf. Stockemer et al., 2018).Especially anti-immigrant views are closely related to specific conceptions of nationhood.Yet, this important aspect of political identity has only been sparsely researched as one of the drivers of the success of far-right parties (Lubbers & Coenders, 2017).
Classical national identity research links conceptions of nationhood to issue positions that influence voting.Most of this research builds on the distinction of ethnic versus civic nationalism that has been derived from state-level resources, for example, legal documents, public education and ideological histories of nation-states (Brubaker, 1994;cf. Helbling et al., 2016;Kohn, 2005;Meinecke, 1970;cf. Reeskens & Hooghe, 2010;Smith, 1987Smith, , 1991)).Transferring the distinction to the individual level, individual-level civic conceptions are defined by placing importance on voluntaristic criteria, such as law-abidingness.In contrast, individuals holding ethnic conceptions are expected to favour objected, ascribed and thus exclusionary criteria, such as descent, blood relations and nativity (cf.Helbling et al., 2016;cf. Jones & Smith, 2001;cf. Wright et al., 2012).Especially the latter has been found to be related to hostile policy positions towards immigrants (Hjerm, 1998a;Jones & Smith, 2001;Kunovich, 2009;Wright, 2011;e.g., Bonikowski & DiMaggio, 2016;Filsinger et al., 2021;Hochmann et al., 2016;Mader et al., 2021), which are connected to far-right voting (Arzheimer, 2008(Arzheimer, , 2009;;Arzheimer & Carter, 2006;Stockemer et al., 2018;e.g., Abou-Chadi et al., 2022;Mason et al., 2021).Specific forms of nationalism affect anti-migrant attitudes that, in turn, translate into right-wing voting.Ethnic national identity has been found to be positively related to both the Trump vote in the 2016 US election (Thompson, 2021) and votes for Alternative für Deutschland (AfD) in Germany between 2015 and 2017 (Mader et al., 2021;Pesthy et al., 2021).In addition to studies on conceptions of nationhood (that are, e.g., based on descent, birthplace, language proficiency etc.), others find that the strength of identity (Garand et al., 2020), or both general national pride and ethnically charged conceptions (Lubbers & Coenders, 2017), play a crucial role for right-wing votes.Bonikowski et al. (2021, p. 530) highlight the specific combinations of domain-specific pride, national chauvinism, strength of identification and the classically used conceptions that 'provided a crucial base of support for Donald Trump […]'.
Building on these findings, we theorise that national identity builds a basis for far-right voting.Because of the questionable validity of the ethnic-civic distinction in national identity, however, we move beyond this distinction.This is because, first, it is conceptually problematic to infer an individual's national identity from policies (cf. Reijerse et al., 2013).Second, ethnic and civic identities are empirically highly correlated and thus hardly differ in their exploratory power concerning policy positions (Janmaat, 2006;Wright et al., 2012;cf. Hochmann et al., 2016).Third, placing measured criteria in either category is often theoretically and empirically ambiguous (comparing results of Ariely, 2020;Berg & Hjerm, 2010;Hadler & Flesken, 2018;Hochmann et al., 2016;Jones & Smith, 2001;Kunovich, 2009;Reeskens & Hooghe, 2010;Wright et al., 2012).In line with a new person-oriented directive in national identity research (Bonikowski & DiMaggio, 2016;Ditlmann & Kopf-Beck, 2019;Hjerm, 1998b;Trittler, 2017), we argue that conceptions of nationhood are more complex, because individuals employ and combine criteria based on underlying images beyond this distinction.In fact, Helbling et al. (2016, p. 752) and Hjerm (1998a, p. 341) find that either only a few respondents or no respondents favour ethnic criteria without also embracing civic criteria, suggesting a continuum from inclusive to exclusive conceptions, while Trittler (2017), May (2023) and Bonikowski and DiMaggio (2016) demonstrate differently understood criteria based on these underlying images.
Additionally, we argue to retain the theoretically distinct dimensions (cf.Eger & Hjerm, 2022a, pp. 3ff.) of national identity that have been merged in more recent works (Alemán & Woods, 2018;Bonikowski et al., 2021;Bonikowski & DiMaggio, 2016;Soehl & Karim, 2021).National identity includes (i) the individual's awareness of being part of a nation, (ii) an affective dimension (e.g., pride and national chauvinism) resulting from this identification and (iii) a content dimension (Citrin et al., 2001) that encompasses national boundary-making and thus the conceptions of nationhood.This latter dimension describes the criteria used to distinguish between compatriots and others and, thus, determines who is recognised as part of the in-group (Bail, 2008;Helbling et al., 2016, p. 746;Lamont & Molnár, 2002, p. 168).We expect the content dimension to drive the relationship.Who is considered an 'outsider' depends on the criteria one applies (Hjerm, 1998a, p. 337).Distinction, in turn, is a key precondition for the devaluation of others (see Tajfel & Turner, 1979, on Social Identity Theory [SIT]).Voters who perceive immigrants as a threat to the country's culture, shared values or the welfare state will be more likely to vote for the far right.Crucially, the degree to which immigrants are recognised as outsiders and as threatening should depend on the narrowness of the idea of the national in-group.Supporting our decision, previous works find that exclusionary conceptions are strongly related to national attachment, pride and chauvinism (May, 2023) and that conceptions of nationhood outperform the other dimensions in strength when analysing the effect of different dimensions on right-wing voting (Lubbers & Coenders, 2017) and attitudes connected to right-wing voting (Hjerm, 1998a).
Taken together, we expect that voters with a narrower and thus more exclusionary definition of national identity are more likely to hold attitudes that align with far-right parties.Just as right-wing parties derive their positions from nationalist and anti-migration stances (Golder, 2016, p. 480), voters are also expected to derive their policy positions from their national identity and associated negative attitudes towards foreigners (Bonikowski, 2017, p. 188), making far-right voting more likely for those individuals.We hypothesise that: Hypothesis 1.Those with an exclusionary national identity are more likely to vote for the far right.

| Political elite discourses and far-right voting
From a cross-national perspective, issues of national identity and immigration are often prominently discussed in European societies.Political elites are one of the key players in these national discourses around immigration-related issues.By providing information about ethnic minorities, political elites play a key role in shaping discourses (Flores, 2018).They inform the public about social and political circumstances, interpret and contextualise arguments, and can thus affect people's understanding of reality (Careja, 2015;Chong & Druckman, 2007).Following Czymara (2020), we define political elite discourses as 'the sum of political elites' frames, arguments, and narratives' and as a 'contextual character, in the sense that they are a feature of a country at a certain time point' (p.1215).
Voters are directly or indirectly exposed to political information, for example, through mass media consumption (Meltzer et al., 2020), social media (Bail et al., 2018), interpersonal communication (Schmitt-Beck, 2003) or other sources.In other words, information and arguments put forward in elite discourses constitute an information environment to which individuals in a given context are exposed.Thus, such an information environment can influence individual perceptions, attitudes and behaviour beyond direct consumption.This does not imply, however, that everyone is similarly receptive of such information.In fact, our core argument is that these discourses resonate more with those holding an exclusionary identity.
Interestingly, Bohman and Hjerm (2016) show that the rhetoric of the radical right itself has little impact on public opinion.Instead, theories that see preferences on issues fixed among the electorate explain increase and decrease in vote shares for far-right parties based on whether other parties take up their electorates' demands for stricter immigration policies (Arzheimer, 2009).In contrast to the hypothesis that far-right parties are winning votes when no other party satisfies these demands, some scholars argue that the catering of these issues by other parties causes the far right to gain votes (Lochocki, 2018;e.g., Krause et al., 2023).By taking radical right-wing positions, established parties may remove taboos, which eases far-right vote choices.Furthermore, by putting issues on the political agenda by taking a position, voters may evaluate their own positions and include the issue in the decision-making process (Arzheimer, 2009).Empirically, it has been shown that political messaging of established parties on immigration issues are a sufficient condition for far-right vote gains (Lochocki, 2018, p. 60).Moreover, far-right parties are found to benefit more overall from positional shifts towards far-right issues by mainstream parties (Krause et al., 2023).
Our contextual argument concerns a country's discursive climate overall.While some scholars argue that public issue attention affects parties' issue priorities in their campaign programmes (Klüver & Spoon, 2016), only relatively popular parties seem to engage in 'riding the wave' of virulent issues (Wagner & Meyer, 2014).Meanwhile, party positions appear to be unresponsive to voters' policy shifts (O'Grady & Abou-Chadi, 2019) and, overall, have been found to lead public opinion (Hellström, 2008), for example, through the media picking up on highlighted issues of electoral programmes (Merz, 2017).Taken together, increased salience of immigration-related topics in the electorate seems to motivate parties to take a stand on these issues, which in turn increases exclusionary and inclusionary positions (dependent on party positions) offered to the public overall.In connection with those on the far right benefitting from positional shifts towards their stances, we hence argue that the combined discourses of all political parties might lead to perhaps unintended consequences for far-right voting preferences.This is our second hypothesis: Hypothesis 2. Increasing exclusionary political elite discourses in a country are associated with a higher probability to vote for the far right.(Elite Discourse Hypothesis)

| The activation of identities through elite discourses
So far, we have hypothesised that national identity and exclusionary elite discourses affect far-right voting independently.Yet, national identity is a rather stable feature of individuals (cf.Mader & Schoen, 2023) in modern democracies and is therefore likely to have a continuous impact on voting decisions of the far right.In contrast, far-right voting is less stable and has only recently increased.Findings from the US presidential election in 2016 specify the relationship further.Studies (Bonikowski et al., 2021;Garand et al., 2020;Mason et al., 2021;Thompson, 2021) suspect Donald Trump to have played a crucial role in activating a nationalist base through 'intense immigration rhetoric' (Garand et al., 2020, p. 15), meaning the congruence between specific long-held ideas in the population and his ideas, respectively (Bonikowski et al., 2021).We test this activation hypothesis beyond the specific Trump context.Social-psychological theories agree that, in order to become behaviourally relevant, identities require activation (Carter, 2013, p. 204;Stets & Burke, 2000, p. 231).Activation is theorised as a function of situational factors, for example, the fit of an identity's meanings with a situation and the accessibility of these meanings within the selfconcept, creating centrality or salience of an identity (Carter, 2013, p. 206;Stets & Burke, 2000, p. 230).Identity Theory (IT) emphasises the organisational structure of identities in a salience hierarchy in which lower identities are less likely to guide activity across situations (Stets & Burke, 2000, p. 12f citing Stryker, 1980, and Stryker & Serpe, 1994).
In contrast to person-based (e.g., moral identity) and role-based (e.g., mother and student) identities, group-based identities such as national identity are rather low in this hierarchy and, thus, require explicit activation through situational factors and social cues that align with individually held meanings attached to the nation.Elite discourses provide such social cues.Exclusionary political elite discourses have thereby been shown to foster the salience of national identity for individuals (Helbling et al., 2016).Crucially, persons with narrow images of the national in-group should be most affect by such discourses, since their individually held images and the meanings of the exclusionary elite rhetoric align easily.In turn, the discourses should increase the significance of national identity for these individuals specifically.
Emphasising the categorisation processes of the self and others as well as the inherent positive in-group evaluation of this process, SIT furthermore emphasises the possibility of re-categorisation and the consequences of activated identities.In a survey experiment, respondents who belonged to a smaller ethnic group and an overriding national group preferred the policy positions that would benefit the group that had been previously primed through the questionnaire's wording (Transue, 2007).Hence, identities not only develop political potential but also affect policy positions dependent on which identity is primed.In accordance with SIT, voters will generally select the option that most benefits the group to which they belong.Yet, which of these groups is used as a reference depends on the activation.
Anti-immigrant rhetoric among political elites contributes to the overall salience of national identity.We expect that the priming effect increases for individuals with exclusionary conceptions of nationhood because meanings concerning national boundaries align.Moreover, since these individuals are more likely to perceive immigrants as a threat and favour those policies benefiting 'their' national in-group, anti-immigrant rhetoric should increase the attraction of stricter migration policies and, thus, of far-right parties.Our final hypothesis regards this interplay between individual views on national identity, national political elite discourses and far-right voting: Hypothesis 3. The relationship between exclusionary national identity and far-right voting is conditional on its prior activation through highly prevalent exclusionary political elite discourses.(Activation Hypothesis)

| DATA AND METHODS
We use data from the ONBound project, which combines harmonised data from the International Social Survey Programme (ISSP, 1998(ISSP, , 2012(ISSP, , 2015) ) and the European Values Study (EVS, 2020a(EVS, , 2020b).Since we are interested in voting, we only include natives and individuals older than 18.After listwise deletion of missing data based on the variables in our model (see below), our analysis is based on 127,409 respondents observed in 26 countries and 26 years, with a total of 104 country-years.Table S1 shows the final samples sizes for each country in each year after data cleaning. 1

| Outcome
To classify parties as far-right, we draw upon the definition of the PopuList project that classifies parties as 'far-right' (Rooduijn et al., 2019).Since we harmonise data over various programmes, we created a binary variable with value 1 if a respondent indicates to vote in the next election for (ISSP 1995, ISSP 2003, EVS 2008), has voted in the last election for (ISSP 2013) or prefers (EVS 2017) a far-right party and 0 otherwise.As we combine these different concepts, we refer to our outcome as a far-right voting preference.

| Individual-level predictor: National identity
Based on recent advances (Bonikowski & DiMaggio, 2016) recommending person-centred methods to analyse individually held images of the national in-group, we follow May (2023) and employ LCA together with cluster analysis (CA) in a cross-country design.
ISSP and EVS measure conceptions of national identity by asking respondents to rate the importance of different criteria for being truly 'national' on a 4-point Likert scale.The survey programmes differ only in the surveyed criteria, which are as follows (Table 1).
Figure 1 plots the results for each survey.Since the variables have been dichotomised to represent either support or rejection of any criterion, values closer to 1 indicate very high conditioned probabilities and high within-class homogeneity.Thus, all members of this type agree on the importance of the criteria with a probability larger than .8. Accordingly, values closer to 0 indicate that the respective criterion is rated as at least 'not very important' but yields also high homogeneity within the classes.Consequentially, all members of a respective type agree on the rejection of this characteristic to describe the national in-group with a probability of less than .2.Values between 0.2 and 0.8 only indicate tendencies concerning the importance of items (0.2-0.4: weak disagreement; 0.6-0.8:weak agreement).Values close to 0.5 suggest strong internal heterogeneity concerning the criterion's importance.Note that lines connecting the points do not imply trends or causal relations.
Even though items vary across the survey waves, we identify comparable types across all five survey waves.Particularly, in each wave, we identify one type (indicated by red areas within the radial plots) that places high importance on each surveyed criterion, except for religion.We call this type the exclusionary identity, since respondents belonging to this type set high barriers to national membership.This type consistently appears with a very clear response pattern and thus displays a coherent national identity.Below, we test the relationship of this type to the far right and its interaction with political elite discourses, since we argue that those with narrow conceptions set high barriers to the national in-group are more likely to prefer far-right parties that offer harder stances on migration.

| Country-level predictor: Exclusionary political elite discourse
To capture an exclusionary political elite discourse, we use data from the Comparative Manifestos Project (CMP, version 2021a; Volkens et al., 2021b).The CMP offers quantitative content analyses of election manifestos for parties competing in democratic elections after World War II in OECD nations, including EU member states and several other countries (Volkens et al., 2021b, p. 2).The content analysis of party manifestos is done through manual coding performed by trained humans working with a fixed set of coding rules.Party manifestos content analyses are  particularly suited to capturing the political stances of parties because manifestos represent parties as a whole, cover long time periods, allow comparisons within and between parties and countries and are replicable (see Dancygier & Margalit, 2020, p. 743).In particular, coders of the CMP quantify how much of a manifesto is devoted to certain predefined topics (Klingemann et al., 2006), theoretically ranging from 0% (issue not mentioned in a manifesto) to 100% (no other issues mentioned in a manifesto).These measures are based on the same coding instructions for all countries and are thus particularly well suited for cross-national comparative research (Bohman, 2011;Careja, 2015;Schmidt & Spies, 2014).We use the statements of party manifestos to proxy for the positions of political elites belonging to each party, whose arguments often have high visibility (Helbling et al., 2016, p. 752).Our measure reflects the prominence of exclusionary issues on the agenda (Arzheimer, 2009, p. 265).Research on the crossvalidation of the CMP data and expert surveys conclude that both approaches measure party positions similarly (Marks et al., 2007;Netjes & Binnema, 2007).Moreover, using CMP has the advantage of bypassing endogeneity bias due to party signals reacting to other parties' positions (Arzheimer, 2009, p. 265).
We combine two items from the CMP data: first, positive statements about a national way of life (per601), which cover 'Favourable mentions of the manifesto country's nation, history, and general appeals' and include appeals to patriotism, national pride or nationalism, as well as 'Statement [sic] advocating the restriction of the process of immigration, i.e. accepting new immigrants', and second, negative statements about multiculturalism (per608), which include 'The enforcement or encouragement of cultural integration' and 'Calls for immigrants that are in the country to adopt [its] culture and fully assimilate' (Volkens et al., 2021a).For our final measure of an exclusionary political elite discourse, we take the values of these two items for each party in each election, weighted by with the respective party's vote share to account for the fact that an argument is likely to be more visible in the national discourse if it comes from a more popular party.Similar to Helbling et al. (2016) and Czymara (2020), we generate the countryspecific mean for the two weighted items and sum them up.Finally, we standardise the variable with a mean of zero

| Controls
It might be that political elites primarily address actual immigration developments and that people turn to far-right parties during times of high immigration (cf.Arzheimer, 2009).Because we are interested in the effect of discourses beyond the influence of actual immigration, we control for national immigration rates as measured by the United Nations High Commissioner for Refugees (UNHCR, 2021), which includes data on refugees, asylum-seekers and stateless persons.To the best of our knowledge, other datasets on migration unfortunately do not include data prior to 1998 for all countries included in our analysis or only provide data for every 5 years.Thus, we decided to use harmonic annual data covering the whole time period for all countries.We standardise these numbers by total population size.Second, it is possible that political elites turn to immigration issues in times of economic hardship as a scapegoating strategy.Following realistic conflict theory, some people are also more likely to vote far right during such times (cf.Arzheimer, 2009;cf. Golder, 2016).To account for the national state of the economy, we control for unemployment rates (as per cent of total labour force) and gross domestic product per capita (based on purchasing power parity and the 2017 international US dollar).We take these data from the World Development Indicator database (World Bank Data, 2021).
On the individual level, we account for demographic and socio-economic characteristics that can influence national identity and voting.Far-right vote choices are more likely for less educated, younger, male and economically vulnerable individuals due to a higher likelihood to compete for resources and feeling threatened by immigrants (cf.Arzheimer, 2009;cf. Stockemer et al., 2021).In contrast, religiosity (Siegers & Jedinger, 2021) and union affiliation (Mosimann et al., 2019) can have an 'immunising' effect on far-right voting decisions.Thus, we control for education, work status, a non-linear age effect, gender, church attendance and union membership.Table 2 gives an overview of the descriptive statistics for the variables in our model.

| Statistical model
We employ country-fixed effects modelling with cluster-robust standard errors to estimate within-country effects of political elite discourses.These estimates are not plagued by omitted variable bias due to time-constant aspects on the country level (Bell et al., 2019;Fairbrother, 2014;Schmidt-Catran et al., 2019), which include stable differences in political or legal factors, such as the citizenship policies or historic path dependencies mentioned above.We employ fixed effects linear probability models.Results are similar for non-linear (logistic) probability models.
We are particularly interested in the question of whether an exclusionary political elite discourse has a stronger effect on far-right voting preferences for individuals who hold a nationalistic, exclusionary identity.To test this hypothesis, we use the cross-level interaction estimator developed by Giesselmann and Schmidt-Catran (2019).That is, we do not only interact the variables political elite discourse and identity but also country dummies with both the variables political elite discourse and the identity variables.These additional interaction terms absorb all potential between-country variances.Thus, this estimator provides a genuine within-estimator for the cross-level interaction term as well.This means that the interaction effect uses only variation within countries over time that, again, is not plagued by unobserved heterogeneity on the country level.

| Descriptive overview
A striking 53% of respondents hold an exclusionary national identity (see Table 2).Seemingly in contrast, merely 7% in our data prefer a far-right party.This implies that, although highly prevalent in Europe, an exclusionary identity correlates with far-right voting far less than one would usually expect.In the models below, we test our hypothesis that these identities need to be activated.
Figure 2 shows trends in exclusionary political elite discourses in European countries over time.First, countries differ significantly in their average level of exclusionary discourses among political elites.While Scandinavian countries tend to exhibit little exclusionary discourse at all time points, such discourses tend to be more prevalent in Eastern Europe.Second, there are differences in trends.While lines are almost flat for Spain and Italy, there is a clear upward trend for Latvia and Denmark.The highest value of all countries, however, is Hungary in 2018, after an inflow of refugees to Europe, a development that lad to particularly negative attitudes among Hungarians (Czymara, 2021).
Figure 3 plots the average probability of far-right voting preferences on the aggregate level against a country's level of exclusionary political elite discourses for each time point.We see a clear positive correlation: The average probability is larger in a country during times when political elites are more exclusionary.An extreme case is, again, Hungary in 2018, where the average probability reached 47.2%.One explanation for this is that Hungary is one of the few cases where the largest (and governing) party, Fidesz, is classified as far right.The bivariate correlation between average far-right voting probability and exclusionary elite discourses amounts to .46.This correlation is to a large degree driven by Hungary.Removing Hungary also leads to a positive, but lower, correlation of .24.
To test these relationships more thoroughly and to examine whether the impact of an exclusionary political elite is stronger for natives with an exclusionary national identity, we now turn to the longitudinal fixed effects models.

| Longitudinal fixed effects models
The results of the fixed effects models are shown in Table 3. Model M1 includes the two main predictors: national identity on the individual level and political elite discourse on the country level.We see that both have a clear and statistically significant positive effect on the probability to prefer voting for a far-right party.Holding exclusionary conceptions of the national in-group is associated with a 4-percentage point increase of the probability of voting preference for a far-right party (p < .001).Similarly, an increase of one standard deviation of the exclusionary political discourse variable is associated with a 5% increase of far-right voting probability ( p < .001).Adding country-level controls reduces the within effect of political elite discourse to 4 points, while the difference between those with an exclusionary national identity and others remains similar, as model M2 shows.In model M3, adding control variables on the individual level does not change these coefficients at all.Table S2 includes the coefficients of all variables.
T A B L E 3 Fixed effects models.F I G U R E 3 Far-right voting preferences and political elite discourses in Europe.
12 MAY and CZYMARA To test whether the effect of exclusionary political elites differs for exclusionary national identity, we estimated the respective cross-level interaction.Figure 4 plots the results of the interaction by showing the association between political elite discourses and the predicted probability of far-right voting preferences depending on national identity.Even after netting out unobserved effect heterogeneity on the country level, there is a strong and statistically significant interaction effect of 2.2 percentage points ( p < .001).There are practically no differences in the probability of far-right voting preferences between exclusionary identities and non-exclusionary identities in country-years with little exclusionary political elites.The predicted differences amount to merely 1.4 points ( p = .019)when elite discourses are at their minimum value (À1).This difference strongly increases when political elites are very exclusionary (6): For the maximum value of political elite discourse, the difference predicted by the model is 15.9 percentage points and clearly significant ( p < .001,although the confidence intervals are widened due to a smaller number of observations for these levels, as shown in Figure 4).
In sum, both exclusionary national identity conceptions and exclusionary political elite discourses are positively related to far-right voting preferences.Both aspects also interact, with the exclusionary rhetoric of political elites fostering the relationship between exclusionary national identity and far-right voting.

| Robustness checks 2
Figures 2 and 3 show that Hungary has very high values regarding both our dependent and independent variables and could thus be viewed as an outlier.Although we believe that such strong cases also provide important information on the association of interest, we re-estimated the interaction model without Hungary to test the robustness of our findings.Interestingly, the interaction difference becomes slightly more pronounced when Hungary is excluded, amounting to 20 points ( p < .001).Thus, we conclude that Hungary is indeed an influential case in our data but that it is not driving our results.
A major objection against the theoretical model we proposed to explain the correlation between elite discourses and voting is reverse causality.It is also plausible that parties change their rhetoric in response to prior shifts in public opinion (although our discourse variable is constructed such that it always precedes our voting measure).If this is true, our choice to weight discourses by party success might be particularly problematic, because electoral success is the result of public opinion.To counter this objection, we ran the models with an unweighted discourse measure in which each party's stance contributes equally to the national context variable.While we are convinced that the weighted version of the variable is better suited to capturing the actual political climate in a given context, we obtain similar results if we use the unweighted measure.Similar to our findings above, the interaction remains significant, with an interaction difference of 20 points (p < .001)and an overall interaction effect of 2.1 percentage points.
Third, one can argue that far-right voting preferences are mainly possible in contexts in which a far-right party exists in the first place.Accounting for this in our dataset is complex because survey waves do not always overlap with elections and because our outcome includes questions on prospective voting and questions on general party preferences.To proxy the existence of a far-right party for a country at a given point in time, we re-estimated the models including only country-waves where at least one respondent chose a far-right party, assuming that if there was a far-right party, at least one surveyed person would choose it.Two-thirds of our respondents belong to a country-wave where at least one respondent exhibited a far-right voting preference.However, this strategy is likely overly conservative because the number of far-right voters might be too low in several countries to be included in the surveyed population by chance.Far-right voting is not very common overall and is additionally subject to social desirability bias.In any case, the results reported above are stable to this sample decision: Both the individual-level national identity effect and the macro-level exclusionary discourse within effect remain similar in size and statistically significant at p < .001.The same is true for the interaction, which remains at roughly 15 points ( p < .000).As an even more conservative test, we experimented with models that drop contexts in which fewer than 100 respondents voted far right.This dramatically reduced our number of cases, as it drops 80 country-waves.Even in this extreme F I G U R E 4 Effect of exclusionary political elite discourse conditional on national identity.
case, however, our results remain similar: The interaction term ranges from 0.021 to 0.024 and is always significant at least p < .01.
Fourth, one might argue that activation results from general issue salience, not exclusionary rhetoric only.When using mere issue salience as our predictor (defined as the sum of weighted exclusionary and inclusionary discourses), the estimate is positive and significant.Even though the effects of exclusionary and inclusionary discourses go in opposite directions, an increase in issue salience may activate reflections about the national in-group with different effects for types of national identity.However, the salience measure is composed of weighted inclusionary and exclusionary positions, with inclusionary positions taking up much less space within the party manifestos.The salience effect is thus largely driven by exclusionary stances.When exclusionary and inclusionary discourses are modelled as separate variables in the same model, the estimate of exclusionary discourses is positive and significant ( p < .001),while the estimate for inclusionary discourse is negative but not statistically significant ( p = .451).Hence, it is mainly exclusionary political elite discourses that play a role, not inclusionary ones.
Fifth, one might question the CMP as a proxy for exclusionary discourses.Dancygier and Margalit (2020) provide an alternative operationalisation that explicitly codes party manifesto positions on immigration issues and thus offers measures that are closer to exclusionary discourses.Re-running our analysis with these data yields similar conclusions to our main analysis: Again, exclusionary party positions correlate with far-right preferences, and again, this is particularly strong for respondents with an exclusionary identity (interaction = 0.32, p = .001).Despite the great benefit of being directly connected to immigration, we ultimately decided against this dataset after careful consideration, because its coverage of parties and elections is much more fragmentary than the CMP data.Concretely, using the data of Dancygier and Margalit (2020) would mean that, after listwise deletion, we would be left with less than half of countries (11 vs. 26), a third of country-years (36 vs. 102) and merely 12% of parties (190 vs. 1603).Yet, the fact that our main argument replicates with this data source that has been coded completely independently from the CMP increases our confidence in our findings.
Sixth, we estimated linear probability models for our binary outcome.Re-estimating models with logistic regression leads to the same conclusions: With an odds ratio of 1.07, the interaction is substantial and statistically significant ( p = .006).Thus, the link function does not affect our conclusion.
Finally, the classification of far-right parties is not always perfectly clear.As an alternative measure, we use the ParlGov database (Döring & Manow, 2021) classification scheme for our outcome (however, there is a 96.3% overlap between both classifications in our data).ParlGov categorises parties as 'right-wing' based on their position in the classical two-dimensional economic (state/market) versus cultural (liberty/authority) left-right scheme.The coefficients are smaller with this outcome (national identity effect = 2.2 points and exclusionary discourse effect = 1 point), but both remain positive and statistically significant at p < .001.The interaction term also remains statistically significant at p = .004.

| DISCUSSION AND CONCLUSION
Our results show that exclusionary identities are prevalent in Europe, whereas far-right voting is less frequent.It thus appears that the connection between such identities and far-right voting is far weaker than one might assume based on their ideological overlap.We can however explain this empirical pattern: Identities need to be activated to become relevant for voting. 3This happens through national political elites who employ exclusionary rhetoric.Based on the most extensive data available, we show that both national identity and political elite discourses influence farright voting preferences across countries and time in Europe.Within-country estimates demonstrate that the otherwise loose relationship of exclusionary identity and far-right preferences is significantly stronger in times of exclusive elite discourses across parties.Conforming to Krause et al. (2022) and Lochocki (2018), this implies that far-right parties benefit from an overall exclusionary political climate, in which national boundaries are on the political agenda.
Consequently, it appears that parties that are not far right are likely to hurt themselves when they use such rhetoric.
Our results add to the puzzle of who is particularly moved by a more exclusionary climate and show that individuals with narrow conceptions of nationality, who account for more than half of the respondents, are more likely to be affected and driven towards right-wing parties, translating to enormous voter potential for the far right.Political elites who use exclusionary rhetoric might thus cause unintended negative social and political consequences.Additionally, we showed that narrow conceptions of national identity are more likely to align with far-right positions and thus are prone to being activated through exclusionary discourses.
While we offer important insights on the mobilising potential of elite rhetoric on voting, our study has limitations.First, we argued that political elite discourses primarily trigger existing national identity conceptions into relevance, not that they trigger them into being.However, it could be that exclusionary political elite discourses lead voters to (re-)formulate their images of the national in-group (cf.Helbling et al., 2016).Adding to this point, individuals can also draw their conceptions of the national in-group from experiences with state institutions, such as citizenship policies (Almond & Verba, 1963).In times of exclusionary debates, those policies might be more prominently discussed.Future research may test the impact of top-down processes of policy frames such as Germany's policy reform in 2000.Second, the presence and success of right-wing parties overall increases the salience of far-right issues and thus may motivate other parties to deal with the issue.Third, because we harmonise data across different survey programmes, our outcome combines voting intentions, past voting and party preferences.Although we assume that these aspects are highly correlated, combining them into a single measure is not ideal from a methodological point of view.Fourth, while we observe the same set of countries over time, we are limited to pooled crosssections on the respondent level.Individual-level panel data would be ideal for testing our argument that national identity is activated by political elites.Unfortunately, we are not aware of any panel data that include the relevant items to analyse this and certainly not across countries.Finally, future research could analyse the relationship between other types of national identity, other kinds of political elite rhetoric and other voting behaviour.
Despite these limitations, we highlight the role of both national identity and political elite discourses and their interplay for far-right voting preferences in Europe.Exclusionary national identities are widespread in Europe and offer huge electoral potential for far-right parties.But exclusionary national identities become especially relevant in times of high prevalence of exclusionary discourses by the overall discursive climate in a country.Thus, when it comes to taking more exclusionary positions in the hope of gaining votes, it is likely that the far right benefits most.
T A B L E 1 Item availability across study waves.Some people say that the following things are important for being truly[NATIONALITY].Others say they are not important.How important do you think each of the following is …

F
I G U R E 1 Representations of national identity by survey wave.[Colour figure can be viewed at wileyonlinelibrary.com] and a standard deviation of one, so that the coefficient represents the change in the outcome associated with the movement of one standard deviation of the predictor.Since individual-level variables and the CMP are not always measured within the same year, we matched respondents reporting retrospective voting decisions with CMP values of the respective previous election.Missing values for years without national elections were interpolated using values of the previous elections.
Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/nana.12985 by Tel Aviv University, Wiley Online Library on [08/09/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/nana.12985 by Tel Aviv University, Wiley Online Library on [08/09/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Trends in exclusionary political elite discourses.Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/nana.12985 by Tel Aviv University, Wiley Online Library on [08/09/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 14698129, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/nana.12985 by Tel Aviv University, Wiley Online Library on [08/09/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/nana.12985 by Tel Aviv University, Wiley Online Library on [08/09/2023].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License