Keywords

In the previous chapters, the politicisation of the EU at both the media and parliamentary levels was established in the six EU countries which form part of the study, for both debtor (Greece, Ireland, Portugal and Spain) and creditor countries (Belgium and Germany) during the Eurozone crisis. Having established the degree to which the EU has emerged as a growing issue both at the level of mainstream media and parliamentary debates across Europe since the onset of the crisis, in the remainder of the book we turn to the consequences which politicisation may have for voting in legislative elections.

In this chapter, we combine data from different sources. First, we test the relevance of EU issue voting in six EU countries using an online survey experiment. This design allows us to cope with possible endogeneity problems. Second, we investigate how the politicization of the EU affects electoral behaviour. To this aim, we integrate measurements of EU politicisation in the analyses of the experiment’s results. Differently from previous chapters, where the data was longitudinal and covered the whole period since the creation of EMU (2002–2017), we perform a synchronous analysis using data from 2019. Moreover, in this chapter, we consider the two components of politicisation (salience and tone) separately, since they may have different effects on EU issue voting.

The chapter is organized in the following way. We first present succinctly the extant literature on EU issue voting and its conditionality. Second, we justify the choice of the experimental design. Third, we present the data and the conjoint design. Fourth, we analyse the results and discuss the implications of the findings for the importance of EU issue voting in national elections.

EU Issue Voting in Legislative Elections

Citizens can use elections to select political representatives that share their political views. Through this selection mechanism, voters provide democratic mandates to politicians and parties for implementing specific policy platforms. Such a mandate is fundamental for democratic representation because it creates a connection between policy outputs and voters’ policy positions. In other words, when citizens choose representatives with whom they share policy preferences, the latter can legitimately pursue their policy agenda (De Vries & Hobolt, 2016). The extent to which this legitimacy mechanism works for EU preferences and EU policies seems to have changed over time.

The classical debate about the democratic deficit in the European Union has often stressed the lack of such a link between voters’ and parties’ EU preferences (Follesdal & Hix, 2006; Hix, 2008). According to this view, EU citizens do not have the possibility to select representatives that share their EU positions, given that no political contest incorporates EU issues. Indeed, the elections for the EU Parliament are considered “second-order” elections that are treated by parties as a test for their activity at the national level (Reif & Schmitt, 1980). Similarly, national elections fail to create a link between voters’ and parties EU preferences because EU issues are rarely discussed during electoral campaigns. Consequently, EU integration remains the “sleeping giant” of the national political debate (Van der Eijk & Franklin, 2004).

In more recent times, however, this perspective has been partially challenged by studies showing that citizens may consider their EU preferences when deciding their vote at the national level. At least under certain circumstances, there is evidence of what has been labelled “EU issue voting” (De Vries, 2007).Footnote 1 Using data from the European Elections Study in 2009, De Vries and Hobolt (2016) have shown that in the vast majority of EU member states citizens’ likelihood to vote for a specific party in national elections decreases when the distance between them and the party on the EU issue increases. Similarly, De Vries and Hobolt (2012) have found that voters’ likelihood to vote for a challenger party in the national elections is strongly and significantly affected by their attitudes towards European integration. Moreover, according to De Vries and Tillman (2011), EU attitudes have an effect on the national vote in both East-Central and Western Europe, even though EU issue voting seems to be larger in the former case.

However, the literature has also highlighted that the presence of EU issue voting cannot be taken for granted and seems to depend on contextual factors. In particular, EU issue voting seems to depend on the political competition over EU issues and on the availability of EU information. De Vries (2007), for example, has shown that voter-party proximity on the EU dimension affects electoral choices at the national level only when citizens perceive EU issues as important and political parties have different positions on them. Similarly, De Vries (2009) found that in the Netherlands the 2005 EU referendum increased both the salience of EU issues and the inter-party conflict over the EU, resulting in a more likely EU issue voting in the subsequent national elections. Focusing on European elections, De Vries, Van der Brug et al. (2011) showed that EU issue voting is stronger in those contexts where EU-related information is more available, while Hobolt and Spoon (2012) highlighted that citizens are more likely to base their vote on EU considerations when the degree of pollicization of the EU in the national political arena is higher. Similarly, De Vries and Hobolt (2016) found that EU issue voting in European elections is stronger when EU issues receive more attention in the media. Finally, Hobolt (2005) and Beach et al. (2018) showed that when the campaign is more intense voters are more likely to rely on EU attitudes to vote in EU referendums or European elections.

In sum, previous literature has highlighted that, at least under certain circumstances, party positions on EU issues are congruent with the positions of their voters. This congruence is good news for the legitimacy of European Union activity, to the extent, it is due to EU issue voting. If the proximity between voters’ and parties’ positions depends on citizens voting for parties that share their view on EU issues, EU policies are more likely to match voters’ preferences. However, as we explain in the next section, party-voter congruence on EU issues might also be due to alternative mechanisms.

Experimental Evidence of EU Issue Voting

The present chapter contributes to the literature that analyses EU issue voting. Differently from the most part of the previous works, our study uses experimental data to assess the effect of EU attitudes on the vote. The need for experimental studies derives from the fact that the literature on party cues has established that voters’ positions on EU issues are partially shaped by the positions of their preferred party (Steenbergen et al., 2007; Stoeckel & Kuhn, 2018; Torcal et al., 2018). Given the complexity of the EU political system and their general lack of information about it, voters rely on cues from more informed political actors to develop opinions on EU issues. In other words, voters tend to align their positions on EU issues with those of their party (Pannico, 2017, 2020). This means that the congruence between voters’ and parties’ positions on the EU can be both the cause and the consequence of citizens’ vote preferences. It can be due to voters selecting parties based on their EU positions or to citizens aligning to party stances to cope with the complexity of EU politics. Only the former represents a selection mechanism that provides a democratic mandate to politicians for their EU activity. However, observational data are not suited to cope with these endogeneity problems. The experimental manipulation of parties’ positions, on the contrary, allows us to analyse EU issue voting net of the effect of party cues.

To the best of our knowledge, there are only two studies that use an experimental design to analyse EU issue voting. Hobolt and Rodon (2020) run a conjoint experiment in the UK in 2017 to test the relative impact of the EU dimension and the left–right dimension on vote choice. They present the respondents with pairs of candidate profiles formed by eight attributes and ask them to vote for one of the candidates. Their results show that participants consider the proximity between their own positions on the EU and the candidates’ ones when making their choice. Moreover, in this experimental context, the impact of the EU dimension on vote appears to be greater than the impact of the left–right dimension. Using a similar design, Schneider (2019) runs a conjoint experiment in Germany in 2016 to look at the effect of attitudes towards the EU and attitudes towards specific EU policies (i.e. EU refugee policy and EU financial aid to Greece) on the vote. She presents respondents with pairs of hypothetical politicians’ profiles composed of six attributes. The results of this experiment show that the congruence between candidates’ and voters’ positions on the EU integration matters only for the vote of Eurosceptic respondents, while party-voter congruence on EU specific policies matters for both Eurosceptic and pro-EU participants.

The experiment we present in this chapter follows the same approach as previous experimental studies on EU issue voting, but at the same time aims to overcome some of their limitations. First of all, it goes beyond single-country cases, given it was run simultaneously in six EU countries: Belgium, Germany, Greece, Ireland, Portugal and Spain. This multi-country design allows us to validate our results beyond context (Bruter, 2009), increasing the external validity of the experiment. In this regard, it is particularly relevant the fact that our case selection includes countries with different economic paths within the Eurozone.

Secondly, different from previous experimental works, our design exposes respondents to a clear and comprehensive range of party positions on the EU. The profiles used by Hobolt and Rodon (2020) in their experiment in the UK include the candidates’ position on the 2016 Brexit referendum (Leaving the EU/Remaining in the EU). This operationalization of the candidate position on the EU is strongly related to the British context and is hardly replicable in countries where a referendum on the EU membership has never been held. Moreover, being framed as a dichotomy between the status quo and Brexit, the operationalization disregards the cases in which the candidate and/or the respondent would prefer a deeper integration of the country in the EU. At the same time, in the experiment run by Schneider (2019) candidates’ positions on EU integration are not part of the profiles (i.e. it is not indicated if they are pro or against the EU). The author assumes that their position on EU-specific policies also indicate to the respondents the candidates’ position on the EU integration (i.e. the author assumes that candidates that favour more immigration or an additional bailout of Greece are seen as pro-EU candidates by respondents). On the contrary, our experimental design incorporates all possible EU party positions and clearly assigns them to the parties presented to participants.

Data and Experimental Design

The conjoint experiment was embedded in an online survey run in February–March 2019 in six European Union member states: Belgium (n = 3076), Germany (n = 2568), Greece (n = 1507), Ireland (n = 1514), Portugal (n = 2049) and Spain (n = 2026). For each of the countries, the sample was recruited through a crossed-quota design based on the 2011 census data.Footnote 2 The survey included different experimental manipulations. For this chapter, we only rely on respondents that participated in the conjoint experiment without being exposed to any other previous manipulation. Their number corresponds to approximately one-sixth of the original national samples.Footnote 3 Given that the respondents’ membership in these sub-groups has been randomly assigned, the final sub-samples are in principle as representative as the original national samples.

Participants were presented with the profiles of two hypothetical parties (“Party A” and “Party B”) that were defined by a set of six party attributes with independently randomly assigned categories. Respondents were asked to indicate for which of the two parties they would vote in a hypothetical general election. The experiment had two rounds, meaning that after making their first vote choice, respondents were presented with other two randomly defined party profiles and again asked to vote for one of them.

The party profiles were formed by the attributes shown in Table 5.1. The table also provides, for each attribute, the different categories it could assume within the party profiles, and the corresponding text (in English).Footnote 4 Given the aim of this chapter, the most relevant attribute is the one that indicates the party’s “EU position”. It was randomized over three categories: pro-integration, neutral, and against integration. The table shows that these categories refer to the fact that the party supports, respectively, a deepening of the EU integration, the status quo, or a reversal of the EU integration. As shown in the next section, these party positions on the EU will be matched with the positions of the respondents to assess the effect of the EU issue on the vote.

Table 5.1 Attributes, categories and text for the conjoint experiment

The profiles were composed of five additional party attributes. The party leader attribute was randomized over three categories that indicate different levels of leader’s experience in public governance. The ideology attribute refers to the party position on the left–right axis, while the “Economic performance” attribute informs the respondent about the trend of the economy during the party’s last term in office. Finally, the immigration and corruption position attributes refer to the party’s stance on these issues and were included in the party profiles because they were relevant political issues in several EU countries at the time of the fieldwork.

Within each party profile, the categories of the attributes have been completely randomized. This means that any combination was possible. Respondents may have been exposed to party profiles with unusual combinations of characteristics (e.g. a party from the right that supports a more open immigration policy). However, we opted for full randomization to break the correlation that exists in the real world between party characteristics and isolate the effect of each of them on the vote (Hainmueller et al., 2014). Finally, the order of the attributes in the party profiles have been randomized between respondents, but kept constant between rounds for the same respondent.

The use of this conjoint design was motivated by the need to maximize both the internal and the external validity of our study. The random assignment of party EU positions allows us to estimate the effect of respondents’ EU opinions on the vote accounting for potential sources of endogeneity (i.e. the effect of party cues on voters’ opinions). At the same time, the complexity of the party profiles and the manipulation of several attributes provide respondents with different elements on which to base their vote, approximating the task voters face in the real world (Sigelman et al., 1991).Footnote 5 Moreover, the experiment presents participants with two party profiles and asks them to vote for one of them. This design simulates the choice respondents face on the ballot paper, where they have to choose one of the candidates or parties that compete in the election.

Results

To assess the effect of the different attributes on respondents’ vote, we created a stacked dataset consisting of four observations for each respondent, one for each party profile s/he saw in the two rounds of the conjoint. Using this dataset, we estimate the following model:

$$\begin{aligned}{V}_{ijk} & = \beta 1*{Leader}_{ijk}+\beta 2*{Ideology}_{ijk}\\ & \quad + \beta 3*{Econ. Performance}_{ijk} +\beta 4* {EU Distance}_{ijk} \\ & \quad + \beta 5*{Imm. Position}_{ijk} + \beta 6*{Corr. Position}_{ijk}+{e}_{ijk} \end{aligned}$$
(5.1)

In other words, we use the party’s attributes to predict the vote of participant i for party k in round j. Given that the categories of the party’s attributes have been randomly assigned, the average marginal component effects (AMCEs) can be estimated using linear regression with standard errors clustered by the respondent (Hainmueller et al., 2014). The pooled model also includes country dummies.

As shown in Eq. (5.1), the model does not include the “EU position” attribute in its natural form. We are not interested in whether voters are more likely to vote for Eurosceptic, pro-integration or neutral parties. To provide evidence of EU issue voting, the model should show that the congruence between parties’ and voters’ positions on the EU has an effect on citizens’ electoral choices. In other words, respondents should be more likely to vote for a party that has an EU position similar to theirs than for a party that does not. To test this possibility, we recoded the “EU position” attribute so that it would indicate the proximity between the respondent’s and the party’s position on the EU. To collect respondents’ positions on the integration process, we used the following question included in the survey before the experiment: “Some people believe that the process of European integration should move forward to the creation of the United States of Europe. Others believe that the European Union should be dissolved in order to return to a situation in which states are fully sovereign. In which point of the following scale would you place yourself?”. To provide an answer, participants used a scale ranging from 0 “The EU should be dissolved” to 10 “The EU should move towards the United States of Europe”. We coded as “Pro-integration” respondents the participants that placed themselves in the 7–10 range, while participants in the 4–6 and 0–3 ranges were coded as “Neutral” and “Against integration”, respectively. Finally, matching the party position with the respondent one, we created the “EU distance” party attribute, consisting of the following three categories: “More pro-EU than the respondent”, “Same position”, “More Eurosceptic than the respondent”. Capturing party-voter congruence, this party attribute is more suitable for the estimation of EU issue voting than the “EU position” attribute. For this reason, the former rather than the latter is included in the model.

Figure 5.1 plots the results from the model for the pooled sample.Footnote 6 The coefficients in the graph can be interpreted as the average change in the probability that the party profile will be chosen by the respondent when it includes the listed characteristic value instead of the baseline characteristic value. For example, the graph shows that, keeping all other attributes constant, advocating for a more open immigration policy decreases by approximately 7 percentage points the probability of the party profile being chosen by the respondent, compared to the case in which it advocates for a more restrictive policy.

Fig. 5.1
A graph plots the results of a vote comprising the leaders, ideologies, economies, E U distances, immigration, and corruption. The better-performing economy has the maximum number of votes and the worst-performing economy has the least votes.

The predictors of the vote in the pooled sample

The coefficients that refer to the “EU distance” attribute reveal two interesting characteristics of respondents’ votes. First of all, the figure shows that participants do take into account their attitudes towards the EU when making electoral choices. When the EU position of the party does not match the position of the respondent, either because it is more Eurosceptic or because it is more pro-EU, the participant is less likely to vote for the party, compared with the situation where the respondent and the party have the same EU position. In other words, the graph shows evidence of EU issue voting. Given that parties’ EU positions were randomly assigned, our experiment excludes by design potential sources of endogeneity. Differently from the case of observational results, we can confidently state that the proximity between parties’ and voters’ positions is the cause, not the consequence, of respondents’ electoral choice.

Secondly, the results also show that EU issue voting is somehow asymmetric. Parties that are more Eurosceptic than the respondent are punished more than parties that are more pro-EU. The decrease in the probability to be voted is 8 percent points in the former case and 5 percentage points in the latter one. A Wald test for the equality of the two coefficient produces a F statistic of 4.21 and a p value of 0.04, confirming that the two effects are actually different from each other. Therefore, not all types of incongruence between parties’ and voters’ positions are punished to the same extent.

The disaggregation of the results by country provides a finer-grained picture of EU issue voting in our sample. Figure 5.2 plots the “EU distance” coefficients for each country, while columns 2 to 7 of Table 5.2 in the Appendix provide the full models. In four of the countries, EU issue voting seems to follow the same general pattern we identified in the whole sample. In Germany, Greece, Ireland and Portugal, participants take into account EU positions when making their electoral choices. Moreover, the electoral punishment for parties that are more Eurosceptic than the respondent seems to be larger than the punishment for parties that are more pro-EU than the participant. This is clearer in Greece, Ireland and Portugal, where there is no significant punishment in the latter case. In Germany, the results also point towards the same asymmetry in EU issue voting. However, in this case, both coefficients are significant, and the Wald test does not allow us to conclude that they are different from each other.

Fig. 5.2
A graph plots the E U distance coefficients of Belgium, Germany, Greece, Ireland, Portugal, and Spain. The coefficient parameters are more pro-E U than the respondent, in the same position as the respondent, and more Eurosceptic than the respondent.

EU issue voting in six EU countries

Finally, Belgium and Spain are, to different degrees and for different reasons, particular cases. In Belgium, the results show both the existence of EU issue voting and its asymmetric nature. However, in this country, voters do not significantly distinguish between parties that match their EU preferences and parties that are more Eurosceptic than them. Their electoral punishment only targets parties with a more positive perspective on the EU than their own. Therefore, the asymmetry in Belgium is reversed compared with the other countries. In Spain, voters do not seem to take into account EU positions when choosing a party. Both coefficients are not statistically significant. For Spanish respondents, the congruence between their EU positions and the position of the party did not matter when making their vote choice. This is, therefore, the only country in which we found no evidence of EU issue voting.

These results show that voters partially base their electoral choices on EU attitudes. They are more likely to vote for a party if they share with it a common vision of EU integration. This is true in almost all the countries analysed and when accounting for possible endogeneity problems, as in the case of our design. Therefore, the indirect path of representation in the EU looks effective. In national elections voters tend to select politicians who represent their EU positions within the intergovernmental institutions of the EU. This selection mechanism makes it more likely the implementation of EU policies that match citizens’ preferences.Footnote 7

The second step of our analysis is aimed to investigate the effect of contextual factors on EU issue voting. In particular, following previous literature, we are interested in the availability of EU information and in the level of political competition over EU issues. In order to capture part of this contextual variation, we use two measures of EU politicization in the national political debate: the salience and the tone of EU news. In each country, for the 30 days before the fielding of the experiment, the content of all the online articles of two mainstream newspapers has been recorded (see Table 5.3 in the Appendix). For the salience measure, we calculated for each newspaper the percentage of articles “about the EU” on the total published during the 30-day period. An article was considered “about the EU” if its title contains at least one of the keywords listed in Table 5.4, or if its body contains at least two of them (or only one, but repeated at least twice). We then averaged the salience value of the two mainstream newspapers at the country level to proxy the EU salience in the general national media environment.

The tone measure refers to the direction of EU discourse. For each article, all the sentences mentioning the words listed in Table 5.4 have been translated into English. For each sentence, a sentiment score has been calculated using the R package “sentimentR”. The sentiment score has a negative sign if the sentence has an overall negative tone and a positive sign if the tone is positive. For each article, an average sentiment score has been calculated using all the sentences “about the EU”. For each newspaper, an overall tone for the 30 days has been calculated as the percentage of articles with an average negative tone on the total of articles “about the EU”. The tone at the national level has been calculated by averaging the values of the two newspapers. Table 5.5 in the Appendix provides the average country values of both EU salience and EU tone.Footnote 8

To explore the relationship between EU politicization and EU issue voting, we plot the national coefficients from Fig. 5.2 against the national values of EU salience and EU tone. The left-hand panel of Fig. 5.3 suggests that the availability of EU information in the national media context is relevant for the role that EU attitudes play in citizens’ voting. The figure shows an overall negative relationship, meaning that the higher the salience of the EU, the greater the electoral punishment suffered by the party for having a different position in the EU than the respondent. As might be expected, the relationship is stronger in the case where the party is more Eurosceptic than the respondent than in the opposite situation, but in both cases, the correlation coefficients are significant (–0.61 vs. –0.28). The results are less clear in the case of EU tone (right-hand panel). The coefficients are more dispersed in the figure and it is more difficult to identify a clear relationship. Indeed, the correlation between the EU tone and EU issue voting is strong in the case where the party is more pro-EU than the respondent (–0.79), but very weak in the opposite case (–0.03).

Fig. 5.3
2 graphs estimate the effect of E U distance on the vote versus E U salience and E U tone, respectively for Spain, Portugal, Belgium, Greece, Ireland, and Germany. Spain exhibits the highest values in both graphs and Germany exhibits the lowest value.

Relation between EU politicization and EU issue voting

In sum, the results of our cross-country experiment support previous observational findings. On the one hand, with the notable exception of Spain, we found that in all countries considered voters take their EU positions into account when deciding which party to vote for. On the other hand, our analysis suggests that EU politicization matters for EU issue voting. This is the case first of all when we consider the salience of EU information, while the relevance of EU tone is more doubtful.

Conclusions

In this chapter, we considered the importance of EU issue voting. Using a conjoint experiment design, we investigated whether citizens use national elections to select representatives who share their views on the EU. Such a selection mechanism would be important for the democratic quality of the EU, because it would create a link between policy output and voters’ policy positions.

Our study provided two main findings. First, EU citizens do take into account their attitudes towards the EU when making electoral choices. The fact that these findings were produced using experimental data shows that party-voter congruence on EU issues can be a cause, and not only a consequence of voters’ preference for a specific party. We also found evidence of asymmetry in the EU issue voting. Parties that are more Eurosceptic than the respondent are punished more than parties that are more pro-EU. This may suggest that not all types of incongruence between parties’ and voters’ EU positions are punished to the same extent. These patterns are present in different countries, very heterogenous in terms of trajectories within the EU and Eurozone. With the relevant exception of Spain, we found evidence of EU issue voting in all the countries analysed, being them creditor countries like Germany and Belgium, or bailout ones, like Portugal or Greece. Taken together, these results present a decisive confirmation of EU attitudes as an exogenous factor of voting behaviour.

Following previous literature, we also linked the media context to EU issue voting. The EU salience and EU tone of two mainstream newspapers were used as proxies for the information environment of the different countries. The highlighted relationships suggest a connection between the availability of EU information and the strength of EU issues voting, while the tone of EU news appears less relevant. However, the low number of countries considered does not allow us to draw statistically valid conclusions from this part of the analysis. Further studies considering a larger number of news contexts are needed.