Mainstreaming the populist radical right? Online news exposure and voting behavior in the 2019 European Parliament election

Abstract

How does media exposure relate to support for radical right populist parties (RRPPs)? We contribute to this classic debate by analyzing the web browsing histories and survey responses of six thousand study participants in France, Germany, Italy, Spain, and the UK during the 2019 European Parliament election. Linking direct measures of online news exposure to voting behavior allows us to assess the effects of the salience of issues politicized by RRPPs on their electoral support. The likelihood to vote for RRPPs was higher when the EU was more salient in individual media diets, while exposure to the less salient issue of immigration did not increase the propensity to vote for RRPPs. Alongside consistent results for other party families and interactions with pre-existing voting intentions, the findings indicate that the electoral effects of online media are contingent on the overall salience of a specific issue and voters’ predispositions.

Introduction

Radical right populist parties (RRPPs) have consolidated their presence in liberal democracies. For more than forty years, scholars have tried to understand the societal and political drivers of their success but one piece of the puzzle remains disputed: the role of the media. This is even more relevant considering that in contemporary ‘audience democracies’ most citizens learn about politics from the media (Kriesi 2004: 184), and this also applies to information about immigration and EU integration that are at the core of RRPPs’ campaigns. Still, there is no comprehensive account about the type of media content that may change support for RRPPs. To be sure, many studies have considered the role of news organizations in RRPPs’ success (among others, Walgrave and De Swert, 2004; Vliegenthart et al., 2012; Sheets et al., 2016; Berning et al., 2019; De Jonge, 2019; Van Spanje and Azrout, 2019), demonstrating that the amount and type of coverage in news stories are associated with growing support for RRPPs. Nonetheless, there is little empirical research directly linking media content to individual-level behavior. In other words, ‘we do not know how many and what kind of people are exposed to which messages’ (Prior, 2013: 102). As a result, the connection between media exposure at the individual level and voting for the radical right remains a blind spot in an otherwise prolific field.

This paper tries to fill this gap by investigating the role of individuals’ media exposure in voting for RRPPs in Western Europe. This is a crucial research endeavor, as the relationship between news exposure and support for RRPPs raises important questions about the influence of news organizations in setting the conditions for promoting political party agendas, shaping people’s understanding of a party and its goals, and ultimately amplifying the attractiveness of the radical right (Brown and Mondon, 2021).

Our theory combines insights from the literature in political science and political communication to examine hypotheses about the influence of news content that individuals may use to acquire information. We argue that individuals’ exposure to issues that are at the core of RRPPs’ campaigns in the news plays a pivotal role in broader processes of mainstreaming (and legitimation) of radical right populist politics. Still, media effects can be moderated by individuals’ pre-existing voting intentions. Specifically, we contend that exposure to media content featuring stories related to immigration and EU integration can fuel discontent, ultimately rewarding the radical right electorally, especially among undecided voters. Empirically, the analysis covers a period of three months in five European countries (France, Germany, Italy, Spain, and the UK) where media landscapes and the institutional position of the radical right in the political system vary greatly. We analyze the web browsing histories (‘web tracking data’) and panel surveys of more than 6,000 study participants during the 2019 European Parliament (EP) election. Crawling the content of each news article visited by participants and training a deep learning model based on data annotated by the Comparative Agendas Project (CAP) allows us to classify the political issue for each news visit. By that, we construct more reliable measures than the commonly used self-reports of media exposure that have well-known flaws (Prior, 2009; Scharkow, 2016). This research design allows us to precisely measure which digital contents were seen by which individuals and how media exposure relates to individual-level characteristics and voting behavior (Stier et al., 2020).

The results show that a higher salience of the EU in individual news diets was consistently associated with voting for RRPPs, even after controlling for individual-level covariates and voting intention before media exposure. This relation could be observed in particular for voters who were undecided before media exposure during the campaign. Exposure to immigration was less relevant, amid a generally low salience of the issue. These findings shed light on the role of high-choice digital media environments in contemporary democracies and the ‘mainstreaming’ of radical right populist politics.

Media content and party support

The idea that developments in the media landscape are related to the rise of RRPs is not new in political science and political communication (Norris, 2000; Aalberg et al., 2016; Ellinas, 2018), but only a few empirical investigations focus on individual-level exposure to media content. In our view, such a perspective is needed in order to arrive at a comprehensive understanding of the role of media in the success of the radical right. By media content we refer to the salience of issues in news coverage. We consider salience to be a necessary precondition to achieve other important political outcomes such as priming and framing (Iyengar and Kinder, 2010). ¹ Media content streams through different channels, including television, the press, radio, and sources generated by actors themselves. More recently, digital media have become of utmost importance in political communication processes and a core channel for RRPPs to reach voters who distrust the legacy media (Schroeder, 2018).

A large body of research suggests that media content, as the primary mechanism through which the majority of the electorate receives information about political parties and their campaigns (Kriesi, 2004), affects different aspects of voting behavior (Beck et al., 2002). The literature on RRPPs’ voters and campaigns contends that issues associated with the cultural cleavage of globalization (Kriesi et al., 2006) – notably EU integration and immigration – are more strongly associated with support for RRPPs even if these issues are not anymore the monopoly of radical right parties (Hutter and Kriesi, 2019; Castelli Gattinara and Froio, 2022).

Research on the aggregate level of news media systems highlights that the quantity of coverage given to specific issues can drive public support for political parties. In a case study of Belgium, Walgrave and De Swert (2004) found that coverage of immigration in newspapers and television stations helped increase the electoral results of the Vlaams Blok. In another study, Koopmans and Muis (2009) illustrated that media coverage of Fortuyn appears to have enhanced the party’s polling performance before the 2002 election. Similarly, Vliegenthart et al. (2012) demonstrated that media coverage of issues related to immigration and integration is associated with subsequent changes in public attitudes toward immigration. In a study in the Netherlands, Boomgaarden and Vliegenthart (2007) showed that the quantity of media coverage of immigration-related issues is associated with a subsequent increase in the vote share of RRPPs. These findings have been confirmed in a comparative study of Belgium, the Netherlands, and Germany (Vliegenthart et al., 2012) and more recent findings on the Dutch Freedom Party (Damstra et al., 2021).

Research on the individual level of citizens has expanded these classic theories examining media effects on party support. Bos et al. (2011) suggest that media coverage drives perceptions of right-wing populist and mainstream politicians in the Netherlands. Media coverage has also been found to help explain individual-level party preferences in Germany (Semetko and Schoenbach, 2003). Scholars recognize that by contributing to political knowledge and blame attributions, media coverage influences parties’ support even if these effects are mediated by medium characteristics and individuals’ real-life events (Hobolt and Tilley, 2014). Individual-level studies also demonstrate that the nature of the effects of media coverage on parties’ support depends on the content of news coverage – notably if news stories are about immigration and integration issues – as well as on individuals’ pre-existing attitudes towards immigration and EU integration (Brosius et al., 2019).

In our understanding, there are two key limitations in previous research. The first is the imprecise linkage between media coverage, individual exposure, and individual behavior. As pinpointed by Ellinas (2018) and formalized by Arzheimer (2018: 57), ‘matching media with micro-level data is next to impossible, because mass opinion surveys do not normally collect detailed (i.e., per item) information on media consumption’. As reviewed above, two types of research designs are therefore predominant in the literature: either macro-level correlations between aggregated media coverage and voting behavior that do not allow for individual-level explanations, or survey-based self-reports of media exposure that are not reliable in the case of offline media consumption and perform even worse in contemporary (digital) high-choice media environments (Prior, 2009; Scharkow, 2016).

As a reaction to these measurement challenges, the present research design allows us to precisely measure which digital news contents were seen by which individuals and how exposure relates to individual-level characteristics and behavior. In related work, Guess et al. (2020) used the web browsing histories of a large sample of USA citizens to explain voting behavior in the 2016 USA presidential election. They find that exposure to untrustworthy conservative online content was not significantly related to voting for Donald Trump. Both Wojcieszak et al. (2021) and Guess et al. (2021) used web tracking data to test the effects of partisan news exposure on polarization, again with mostly null effects. Richter and Stier (2022) found that exposure to online news articles featuring European transnational candidates can increase knowledge about these leading politicians, indicating that detailed content measures going beyond general news exposure allow for detecting more precise theoretical and empirical linkages between media exposure, attitudes, and behaviors.

The second limitation in available knowledge is a predominant focus on the ‘usual suspects’, i.e., traditional media outlets that are often studied in isolation from one another, most notably quality newspapers or the tabloid press. Instead, we account for individuals’ exposure across different news outlets including public broadcasting and the recently emerging hyperpartisan news websites that may contribute to RRPPs’ success. This allows us to consider the (in)advertent role of media exposure to explain RRPPs’ support, a step that is needed to account for the growing diversification of contemporary media environments (Chadwick, 2013; Stier et al., 2022). In the next section, we formulate hypotheses about the relationship between media exposure and support for RRPPs.

How media exposure relates to voting for the radical right

The content of public debates can impact support for RRPPs because these may resonate with issues that are at the core of the programs of these parties, notably issues related to immigration and EU integration (Mudde, 2016; De Jonge, 2019). We argue that individuals’ exposure to such issues in the news plays a pivotal role in broader processes of mainstreaming (and legitimation) of radical right populist politics because news content can influence voters not only through the slant of a particular story but also merely by choosing which stories to cover. Still, we contend that media effects can be moderated by the intensity of individuals’ pre-existing voting preferences. In other words, media content can shape vote choices by setting a favorable agenda for RRPPs’ politics and the diffusion of their claims, especially for undecided voters. As such, we ask what characteristics of news content are associated with higher levels of support for RRPPs?

Two main mechanisms can be identified related to (1) the issues covered and (2) pre-existing voting intentions. To begin with, the issues featured in the news. Even if RRPPs are not single-issue parties (Mudde, 1999), it has been demonstrated that these parties have triggered, and then took advantage of, the emergence of specific issues in West European Politics that have restructured political conflict, starting from the 1980s (Kriesi et al., 2012). This is particularly the case with regard to immigration and EU integration issues. Two main classic theories help in qualifying these mechanisms: agenda setting (McCombs and Shaw, 1972) and issue ownership (Petrocik, 1996). Combining these two, we expect that media exposure to issues that feature prominently in RRPPs’ platforms is associated with support for these parties.

Agenda-setting theory posits that audiences perceive issues emphasized by the news media as important. This implies that the salience of an issue in the news can transfer from the media agenda to the public agenda, often referred to as ‘public agenda-setting effects’ (McCombs and Shaw, 1972). However, not all news media content is expected to matter for the support of RRPPs. The theory of issue ownership posits that in voters’ minds, some political parties are in general more associated with specific issues than others and that they are also considered as being more competent to deal with them (Petrocik et al., 2003; Walgrave et al., 2012). In this interpretation, voters do not perceive parties to be equally competent on all issue-related problems. Differently, as Petrocik (1996: 826) put it: ‘a history of attention, initiative and innovation toward these problems … leads voters to believe that one of the parties (and its candidates) is more sincere and committed to doing something about them’.

Today, negative campaigns targeting migrants are not necessarily a prerogative of RRPPs (Hutter and Kriesi, 2019; Castelli Gattinara and Froio, 2022), but RRPPs are primarily associated with immigration-related campaigns because these parties have been among the first to emphasize this issue in European party systems (Mudde, 1999, 2007; Rooduijn, 2015). Studies on media populism and the activation of populist frames have expanded upon the classic agenda-setting and issue ownership dynamics. Esser and Strömbäck (2014) revealed that media populism, through partisan bias and sensationalism, can amplify populist messages and contribute to the growth of RRPPs. Similarly, Müller et al. (2017) illustrated that populist messages in media coverage could activate populist attitudes among voters. Other studies found that media coverage of immigration and EU issues could shape public perceptions and contribute to the emergence of populist beliefs and attitudes (Rooduijn et al., 2014; Müller et al., 2017). We thus expect that exposure to news media content on immigration-related issues upon which RRPPs enjoy most ownership will more easily attract support for RRPPs. Differently, when media content addresses other topics, support for RRPPs is less likely as the public will not recognize their legitimacy and credibility to intervene on these debates.

Hypothesis 1: More media exposure featuring immigration is associated with a higher likelihood to vote for a RRPP.

In addition, in European democracies, radical right parties prosper not only on campaigns on the issues they ‘own’ but also on a rhetoric centered around a deep mistrust in the process of EU integration. Eurosceptic discourse does not only benefit the radical right (Pirro and Taggart, 2018). Still, the radical right tends to be rewarded by voters specifically when mistrust of the Union’s supranational decision-making structure and its global outlook is related to a cultural understanding of Europe, associating EU integration with immigration-related issues (De Vries and Hobolt, 2020). In other words, the EU integration process is criticized for going against RRPPs’ mission to defend national identity, borders and ensure security (Vasilopoulou, 2011). Researchers suggest that the Great Recession has given RRPPs the opportunity to reshape their critique of EU integration accounting for economic arguments especially in countries that suffered severe financial distress and austerity measures (Pirro and Taggart, 2018). Still, the so-called refugee crisis and then the Brexit have brought back culturally inspired forms of Euroscepticism that are believed to favor RRPPs. Accordingly, we expect that:

Hypothesis 2: More media exposure featuring the EU is associated with a higher likelihood to vote for a RRPP.

Moreover, it can be expected that the role of media exposure differs depending on pre-existing political leanings. On the one hand, studies find small persuasive effects of media and candidate or party messages during election campaigns (Kalla and Broockman, 2018) and rather stable political attitudes within the general public over time, e.g., on immigration (Kustov et al., 2021). On the other, there is evidence that media exposure matters most for citizens with weaker pre-existing political preferences (De Benedictis-Kessner et al., 2019). When it comes to voting behavior, media exposure on immigration or the EU might therefore be most relevant for likely voters who have not yet committed to a party. Vice versa, likely voters who make up their mind early during a campaign should be less swayed by coverage of immigration and EU issues. An alternative stream of research would predict that due to motivated reasoning and confirmation biases, voters with pre-existing party preferences should be emboldened in their partisan views when exposed to issue-specific information (Müller et al., 2017; Peterson and Iyengar, 2021). Media exposure might therefore reinforce a party preference over the course of a campaign. In light of these contradictory expectations, we do not formulate a hypothesis but pose an open research question regarding the role of pre-existing predispositions.

RQ1 Does the association of media exposure and voting for a RRPP differ for undecided voters or voters who already preferred a RRPP at the beginning of the campaign?

Data and methods

This study combines surveys with web tracking data and automated content analysis. Below, we describe the process of data collection and the methods used in the analysis. Further information is available in the Online Appendix.

Web tracking

The web tracking data used in this paper were collected by Netquest, a market research company, in compliance with the EU General Data Protection Regulations (GDPR). ² Panelists were incentivized to install a tool that tracks their online activities on desktop computers. Participants were informed about the scope of the data collection and gave their explicit consent to participate in surveys and web tracking. One key advantage of web tracking data is that it provides direct measurements of exposure to political content, instead of survey-based self-reports used in most previous research. The web tracking dataset consists of more than 116 million website visits contributed by 6,002 study participants from 15 March to 16 June 2019. Due to the non-probability sampling of study participants, the sample diverges from the general population in several regards (see demographic statistics for the sample in Online Appendix Section S1). As a validation, Online Appendix Section S4 demonstrates that study participants are comparable to the general population in terms of their online news exposure, their exposure to offline news via newspapers, television, and radio as well as their privacy attitudes (among the subset of German participants).

To identify exposure to political issues on online news outlets, we took two steps. First, we coded the five thousand most visited domains per country and identified 627 news domains with 1,520,717 million news website visits by participants (Table 1; see Online Appendix Section S2 for a description of the coding). Second, the content of the websites was scraped using the R library rvest (Wickham, 2020) and parsed with the newspaper package in Python (Ou-Yang, 2013). The parsing extracted both the headline and main text of the news articles. ³

Table 1. Website visit statistics

Panel surveys

Participants in the web tracking panels were invited to participate in panel surveys on media and politics. The sampling of panelists was determined by national census statistics when possible. In some countries (France, Spain) where the web tracking panels were big enough, approximately 1,500 participants were invited according to population margins, but some quota cells still remained empty. In countries where the number of web tracking panelists was generally lower, all potential study participants were invited (Germany, Italy, UK). Because we observed pronounced deviations from the national population margins for some demographic groups, we post-stratified our samples in accordance with population weights and used these weights in our analyses. The surveys were in the field in late April and early May 2019 (wave 1, in the following ‘W1’), approximately four weeks before the EP election and in the two weeks immediately after the election on 27 May (wave 2, in the following ‘W2’). 6,644 participants participated in W1 and the web tracking. 6,002 participants also participated in the post-election survey W2. The 5,627 participants who did not definitely rule out in W1 to vote in the forthcoming EP elections constitute the final sample for our study (see Online Appendix Section S1 for more information on the sample).

The analyses include an extensive set of control variables that can confound the relationship between media exposure and voting for RRPPs (Rooduijn, 2015; Arzheimer, 2018). These include the demographic controls age, gender, education (standardized across countries based on the International Standard Classification of Education, ISCED), being unemployed and the political confounders political ideology (left/right), satisfaction with democracy and political interest. In addition, we control for respondents’ pre-existing attitudes toward the cultural benefits of immigration and their attitudes towards further EU integration (with higher values indicating more support for immigration and EU integration). We also asked respondents about their offline news use by implementing lists of newspapers, television, and radio programs from the Reuters Digital News Report (Newman et al., 2019) in our surveys. As the measure of offline news exposure, we took the maximum number of days a respondent tuned into her/his most frequented news source in the previous week. Finally, to classify respondents’ voting intention in W1 and reported voting behavior in W2, we use the party classifications from the PopuList (Rooduijn et al., 2020). Accordingly, we defined as RRPP parties the Italian League (former Northern League), Fratelli d’Italia, the UK Independence Party (UKIP), the Brexit Party, the French National Rally (NR, former Front National), Debout la France, the Alternative for Germany (AFD), and the Spanish Voice (VOX). Descriptive statistics for all used variables can be found in Online Appendix Section S3.

Language model

To capture the topic of the news content, we rely on a deep learning approach (see Online Appendix Section S5 for a detailed description and validation). As training data for the automated text analysis, we used annotated party manifesto and media (when available) texts from CAP for the six countries under study. A total of 295,353 texts were used as the training data. Using CAP-annotated texts as training data allows us to classify the news articles into the 20 topics as described in the CAP codebook (Froio et al., 2017). ⁴ Concretely, we employ a state-of-the-art language representation model, Bidirectional Encoder Representations from Transformers (BERT). Due to the multilingual nature of our data, we use multilingual BERT (mBERT), a version of BERT trained simultaneously with a Wikipedia dump of the top 104 languages (Azunre, 2021). The training data were split into a training set and a validation set with stratified sampling based on the topic category. Training was continued until no improvement in validation accuracy was achieved.

The model was then used to predict the topic of the visited news URLs in the web tracking data. To validate the results, a codebook was developed and a sample of 250 articles in the theoretically relevant CAP topics Immigration and International Affairs was hand-coded. The validation shows that our approach is able to correctly capture the topics of news articles with 76.8% accuracy (Table 2). As there is no EU-specific topic in CAP but only the more general topic ‘International Affairs’, we use country-specific EU dictionaries that contain terms related to EU institutions and governance in English, French, German, Italian, and Spanish that were developed by country experts (Maier et al., 2021). We also added the term ‘Brexit’ that was a major EU-related theme in news coverage. ⁵ Filtering articles by topic before applying the EU dictionary maximizes precision. In turn, due to its limited recall, the approach will miss out on EU-related mentions in other topic categories. Nonetheless, as the rate of false positives identified by the dictionary can be really high depending on the topic category, we chose this conservative two-step measurement approach. ⁶

Table 2. Accuracy of topic classification

Before applying the topic classifier, we merged subsequent visits of the same URL to account for automatically refreshing browser tabs and restricted the media exposure measures to the one month between the first survey on 23 April 2019 and the EP election date on May 26. The aggregated exposure measures derived from the web tracking were matched with the survey responses via a pseudonymized unique participant ID.

Results

Overall, 38% of the sample got exposed to news coverage either about immigration or the EU. Figure 1 shows that exposure is distributed unequally across the two topics. 33% of study participants saw at least one news article about the EU, yet only 18% read online news about immigration.

Figure 1. Exposure to news coverage on the EU and immigration.

The only moderate salience of these issues in citizens’ online media diets is not entirely surprising, as EP campaigns are known to be of secondary importance to voters. The amount of exposure also has a characteristically skewed distribution, i.e., a few people were exposed to a substantial amount of news coverage on these issues, albeit a generally limited amount of exposure or no exposure at all. While these numbers are a more accurate reflection of citizens’ political exposure to political affairs coverage than survey self-reports, it also has to be kept in mind that these measures do not capture exposure to politics via traditional media channels.

Even under conditions when citizens are rarely exposed to immigration and EU coverage, few points of contact might already have relevant political effects (De Benedictis-Kessner et al., 2019). The unique advantage of web tracking data is their high granularity, allowing us to subset the time period to only the one-month period between our pre-election survey and the actual election date. Equipped with such ecologically valid measures, we start with a descriptive look at the relation between vote choice in the 2019 EP election (reported in W2) and exposure to news coverage on immigration and the EU as well two additional issues that were salient among study participants (see Online Appendix Section S5), namely civil rights and the environment (including climate change). Voters of RRPPs were exposed most frequently to news articles about the EU, with significant differences compared to several other party families (Fig. 2). In contrast, differences between party families were less clear-cut for immigration exposure. More generally, it is noteworthy that the EU was more prominent in participants’ media diets than immigration. This discrepancy reflects the dynamics during the 2019 EP campaign when immigration was generally less relevant compared to issues such as climate change or the ongoing Brexit negotiations between the European commission and the UK.

Figure 2. Voting in the 2019 EP election and exposure to EU and immigration news.

We next move to a more comprehensive modeling strategy that exploits the advantages of our pre- and post-election panel surveys. This allows to overcome some of the confounding that traditionally haunts explanations of voting for RRPPs (Arzheimer, 2018). We use logistic regression models with voting for RRPPs in W2 as dependent variable, a set of relevant confounders measured in W1 and media exposure in between the two survey waves as the main explanatory variables. Due to a lagged dependent variable controlling for RRPP voting intention in W1, this is a conservative test of the relative role of media exposure on voting for RRPPs. A similar modeling strategy was applied in previous studies using a two-wave panel design in combination with web tracking (Guess et al., 2020; Richter and Stier, 2022). RQ1 on heterogeneous effects is tested by interacting the media exposure measures with a RRPP voting intention or being undecided in the W1 pre-election survey. We included the logged number of visited news articles per participant to separate specific issue exposure from general news exposure and also control for the different contexts by including country dummies. Since the topic exposure variables have skewed distributions (see Fig. 1) and to account for excess zeros, we use the natural logarithm of content exposure, after adding 1 to each content measure.

Table 3 presents the results of four models that include media exposure measures for immigration and the EU as well as the respective immigration/EU issue attitude battery as an additional control variable. Model 1 shows no statistically significant relationship between exposure to immigration-related news and voting for RRPPs. Meanwhile, EU-related news is a statistically significant predictor of voting for the radical right in Model 2. The respective effects become stronger – and immigration exposure also statistically significant – in Model 3 when both exposure measures as well as the respective issue-specific attitude variables are included. To gain a better understanding of the magnitude of these effects, Fig. 3 visualizes the predicted probabilities from Model 3, as well as the histogram of the distribution of the sample along the media exposure measures (similar to Fig. 1). At the higher end of the exposure scales, the confidence intervals become considerably wider due to a limited number of data points. The negative binomial regressions take into account the skewed distribution of the exposure scale and therefore the confidence intervals increase. As only 13 persons had a media exposure score higher than 4 on the logged x axis for EU exposure (the equivalent of 54 EU-related news website visits), only the range until a logged value of 4 will be shown and interpreted in the prediction plots. Moving from no EU exposure to 54 news articles on the EU during the one-month research period would increase the likelihood of voting for a RRPP from 4% (95 percent CI: [0.03, 0.05]) to 20% (CI: [0.11, 0.34]). In contrast, for immigration exposure, the confidence intervals of the predicted effects overlap along the range of the immigration exposure scale. Taken together, these results show that higher online media exposure featuring the EU was related to a substantive increase in the likelihood to vote for a RRPP (supporting H2) while we cannot observe the equivalent relationship for immigration, speaking against H1.

Table 3. Media exposure and voting for radical right populist parties

Note: Results from logit regression models. Low education is the reference category for education. France is the reference category for country dummies. Survey weights are included. ***P < 0.001; **P < 0.01; *P < 0.05.

Figure 3. Predicted probabilities of exposure to political issues on voting for radical right populist parties. Predictions from Model 3 in Table 3.

To investigate a potential subgroup heterogeneity (RQ1), we report the predicted probabilities in Fig. 4. A homogeneous picture emerges in the case of EU exposure. Among people with a non-RRPP party preference in W1 (top right panel), moving from no EU exposure to reading 54 articles on the EU would increase the probability of voting for a RRPP from 2% (CI: [0.02, 0.03]) to 12% (CI: [0.06, 0.22]). Along the same range, the RRPP voting probability of people with a consistent RRPP preference would increase from 35% (CI: [0.27, 0.44]) to 87% (CI: [0.61, 0.97]). Voters who were undecided in W1 (bottom right panel) also stood out from voters with an existing party preference: their probability to vote for a RRPP increased from 6% (CI: [0.04, 0.09]) to 62% (CI: [0.29, 0.87]), though with widening confidence intervals after a logged value of 1.5, which is the equivalent of exposure to four news articles on the EU. For immigration-related news exposure, we zoom in on a lower range of the scale, as uncertainty is higher due to less exposure to this issue than to the EU. The interaction with a pre-existing preference for a RRPP stood out (Fig. 4, top left panel). Moving from 0 to 2 on the logged immigration exposure scale (the equivalent of exposure to 6 immigration-related news articles) lowered the likelihood to actually vote for a RRPP from 41% (CI: [0.33, 0.50]) to 16% (CI: [0.07, 0.33]). Overall, amid a high amount of predictive uncertainty, the subgroup analyses provide inconsistent evidence depending on the issue and W1 voting preferences.

Figure 4. Predicted probabilities of exposure to political issues on voting for radical right populist parties. Interaction of exposure with a RRPP voting intention or being undecided in W1. Predictions from Model 4 in Table 3.

The surprising negative relation between a RRPP voting intention in W1 and immigration-related exposure could also be an artifact of the text classification. The model did not only learn the policy dimensions of immigration but also other genres such as cuisine and traveling where foreign influences are prominently featured. ⁷ To tailor the measure more to the electoral context at hand, we re-constructed the variable including only those immigration-related articles that also feature at least one term from the EU dictionary. The results in Table A8 and Figure A13 in the Online Appendix display the theoretically expected effects: immigration news with a cross-relation to the EU was positively related to voting for a RRPP party for those people who already indicated to do so in W1, even though not at a statistically significant level due to the meager exposure to EU-related immigration coverage. As further robustness tests, we reran the models including voting for the other major party families as the dependent variable (Online Appendix Section S6). Among the main effects, neither EU exposure nor immigration exposure were related to voting for radical-left populist parties, Conservative/Christian democratic parties, green parties, nor social democratic parties at a statistically significant level. EU exposure was positively related to voting for liberal parties (including pro-EU parties like La République En Marche!, +Europa/Radicali or the Liberal Democrats) at a statistically significant level. Additionally, two interactions stood out: participants with a green party voting intention in W1 switched to other parties with increasing immigration exposure. In contrast, more EU exposure increased the likelihood that these people actually vote for a green party. The findings for both party families, the Europhile liberals and greens, are in line with issue ownership theory.

Conclusion

How does media coverage relate to support for the radical right? The answer to this question depends on the extent to which the salience of public issues sets a favorable agenda for RRPPs and the diffusion of their claims. In contemporary ‘audience democracies’ media redistribute power between political actors, by granting or denying attention to their causes. If media coverage increasingly turns toward issues at the core of RRPPs’ campaigns, then the mainstreaming of these parties in contemporary democracies is at stake. Given the importance of these questions, the relationship between media coverage and voting for RRPPs has attracted much attention, with the key question being whether news coverage deliberately or inadvertently increases support for the radical right. The paper provided insights into this question by studying how exposure to online news coverage on immigration and the EU relates to voting for RRPPs. An exploratory subgroup analysis investigated the role of a pre-existing preference for a RRPP or being undecided before getting exposed to news coverage during the election campaign. Going beyond previous studies, we used digital behavioral data and panel surveys to capture individuals’ actual content exposure during the campaign, as well as their demographic attributes and voting preferences at different times of the 2019 EP electoral contest.

Against expectations, we found that the salience of immigration in individual news diets was not a mobilizing factor for voters to choose a RRPP. Among the group of voters who preferred a RRPP before media exposure, the likelihood to vote a RRPP even decreased. However, two caveats are the small sample size underlying the interaction models and the imprecise immigration measure, as results turned insignificant when narrowing down immigration exposure to EU-related content. Nonetheless, as the findings all point toward a low weight of immigration exposure in the minds of voters, they pour some cold water on the hypothesis of the ‘mainstreaming’ of radical right politics through the coverage of immigration. Coupled with an overall low exposure of immigration news, the non-results raise some questions for research relying on aggregate-level time series correlations of immigration issue salience and vote shares (e.g., Boomgaarden and Vliegenthart, 2007) that can mask considerable heterogeneity at the individual level. At the same time, the overall persuasive potential during the 2019 EP campaign was only modest, amid a limited salience of the immigration issue. Additionally, while our measures are more granular than aggregate-level data or self-reports of media exposure, they do not capture content exposure to immigration and EU news in traditional media channels such as television.

A higher salience of news articles on the EU in individual news diets was positively related to voting for RRPPs, even after controlling for individual-level covariates, voting intention at the start of the campaign and potential interactions with pre-existing party preferences before exposure. These results confirm that media coverage of issues mobilized by ‘entrepreneurs’ (de Vries and Hobolt, 2020) such as the radical right can contribute to their support. In terms of the role of the media in EU politics, the findings are in line with other research showing that news coverage can mobilize voters (Schuck et al., 2016), affect learning about EU actors (Richter and Stier, 2022), and trust in the EU (Brosius et al., 2019).

It should again be acknowledged, however, that EP elections are considered ‘second-order’ by most voters (Hix and Marsh, 2011). Consequently, the identified mechanisms should also be tested in ‘first-order’ and more controversially fought national elections. Additionally, our measures of news coverage are only informative of issue salience and cannot differentiate the tone of seen news articles. Future research at the content level is needed to uncover the effects of media framing on vote choices. Additionally, not even our ecologically valid measures, panel surveys, and an extensive set of individual-level controls allow for identifying causal effects of media exposure on voting. A related issue was the low prevalence of news exposure on the EU and immigration. On the one hand, this finding is in line with other studies using digital behavioral data that more accurately reflect citizens’ meager exposure to politics than inflated survey self-reports (Wojcieszak et al., 2021; Stier et al., 2022). On the other, a low prevalence of relevant phenomena is a limiting factor when using measures derived from online data as explanatory factors, at least outside of ‘first-order election’ campaigns. One promising path forward are combinations of ecologically valid experimental encouragement designs (e.g., with a news website or activist group as treatment) with web tracking data to observe compliance and behavioral effects (Guess et al., 2021).

Notwithstanding these limitations, our results imply that media coverage of issues made salient by challenger parties can contribute to their support. Still, the (limited) salience of issues in the political information environment – here immigration – remains a strong intervening variable (see also De Jonge, 2019). These findings hold broader implications not only for scholars of radical right politics and populism but also for those interested in understanding how political communication processes affect the politicization of EU integration and political behavior in European democracies. Nonetheless, the investigation of RRPPs can only be regarded as the beginning of a more general research agenda. In a broader theoretical and empirical framework, future studies should investigate the interplay of individual-level exposure, issue salience, and issue ownership across party families. Given that citizens increasingly receive political information from a multitude of digital channels, digital behavioral data measured at the individual level can play a major role in this research agenda.