The Impact of the Explosion of EU News on Voter Choice in the 2014 EU Elections

The European elections in 2014 were the first to be held after a long period in which EU-related news was prominent in the media. They were held after years of daily news about the euro crisis and after months of news about the popular uprising in the Ukraine against president Yanukovych, who had refused to sign the association agreement with the EU. This could have invited political parties to overcome the usual problem of low salience of EU issues by strongly profiling themselves on EU issues. Turnout at the 2014 EU elections, however, remained low, hinting that parties were unable to convert the attention for European issues into enthusiasm for their party at the European elections. This paper asks how vote choice was influenced by party campaigning on EU related issues. A news effects analysis based on a content analysis of Dutch newspapers and television, and on a panel survey among Dutch voters revealed that EU issues functioned as wedge issues: the more strongly parties were associated in the news with the euro crisis and the Ukraine, the less they succeeded in mobilizing voters.


Introduction
"Hier schlägt das Herz europäisch und dieser Herzschlag, der hier europäisch schlägt, wird auch bei uns in Europa, in Deutschland, in Berlin gehört. Uns ist das Schicksal der Ukraine nicht gleichgültig" [Here the heart beats European, and this heartbeat, which sounds European, is heard by us in Europe, in Germany, in Berlin. We are not indifferent to the fate of Ukraine]. German Minister of Foreign Affairs Guido Westerwelle in his speech to protesters at Euromaidan, December 4 th 2013 For the first time in the history of European elections since 1979, EU related news was prominent in news-paper and television news in the months preceding the European Elections of May 2014. Developments in the Ukraine became the foremost important topic in the news due to the popular uprising against president Yanukovych, who had cancelled the Ukraine's association treaty with the EU. News about EU support for the revolt followed, for example news about the visits to Euromaidan of among others the German foreign min- On the basis of the increase in EU related news one could have expected that political parties would have been able to convert the massive attention for European issues into enthusiasm for the EU issue positions of their party at the European elections: many studies showed or at least suggested that a poor EU visibility as indicated by a low amount of EU related news in previous EU election campaigns contributed to a low turnout in earlier EU elections (De Vreese, 2003;De Vreese, Banducci, Semetko, & Boomgaarden, 2006;Lefevere & Van Aelst, 2014;Schuck, Vliegenthart, & De Vreese, 2016;Schuck, Xezonakis, Elenbaas, Banducci, & De Vreese, 2011;Van Spanje & De Vreese, 2014;Wilke & Reinemann, 2005). The research question of this article is how vote choice was influenced by parties' profiling on EU related issues. In what direction and to what extent was the vote for a party affected by the news coverage of that party's stance on EU issues such as support for the EU debt nations to solve the euro crisis; or a treaty with the Ukraine, against the will of Russia?
This study adds to the recent literature which shows that issue voting matters in a European context (e.g. Hobolt & Spoon, 2012;Van de Wardt, De Vries, & Hobolt, 2014) and to the literature on effects of the visibility and the tone of EU related news (e.g. Azrout, Van Spanje, & De Vreese, 2012;Van Spanje & De Vreese, 2014). The unique contribution is to show that the amplification of parties' issue positions on EU related is-sues in the news media matters for electoral supportalbeit not in a straightforward "more is better" fashion.

The Ukraine and the Euro Crisis in the 2014 EU Election Campaign
The Netherlands is an interesting case to study the impact of the attention for European issues, because of strong variation over time. In the years before the Dutch 'no' to the European constitutional treaty in 2005 the EU was not an important or controversial issue in the news. During the euro crisis enthusiasm for the EU diminished further. The declining enthusiasm for EU politics can be seen from Figure 1, which shows the decrease in turnout from 58.1% in 1979 down to 36.8% in 2009. Figure 1 shows also the development of EUvisibility for Dutch citizens, as indicated by the amount of news coverage on the EU in De Telegraaf, which is the most popular newspaper in The Netherlands. 1 Although 1 De Telegraaf is the Dutch newspaper with the widest circulation, also among the lower educated segments of Dutch society. It's also the newspaper with the highest impact on politics, as measured by the number of Parliamentary questions based on news reports, and it is the only newspaper for which digital content is available from 1979 onwards is available. Attention for the EU in De Telegraaf was measured in each EU election year by the number of news articles about the EU and EU institutions in the five months preceding the EU elections, as operationalized by means of a search query (cf. Appendix) in the Amsterdam Content Analysis Toolkit AMCAT (Van Atteveldt, 2008, cf. https://amcat.vu.nl/navigator and https://github.com/amcat).

Figure 1.
EU visibility in the news preceding the EU elections (5 months) and turnout. EU visibility is defined as the amount of EU news items, as indicated by the number of news items on the EU in De Telegraaf, divided by 10 to arrive at a better scale for visualization. turnout is related to increases or decreases in EU related news, turnout increased only marginally from 36.8% in 2009 to 37.3% in 2014, in spite of the unprecedented amount of EU related news in 2014. Figure 1 shows a long-term negative relationship between the level of EU visibility and EU turnout (r=-0.35) in combination with a positive short-term relationship between their firstorder differences (r=+0.48). This suggests that voters who were not made enthusiastic by any of the political parties in a previous EU election, tend to stay less than enthusiastic, which is known as habitual (non-)voting in the research literature (Franklin & Hobolt, 2011). In 1999 both turnout and EU visibility were very low, but the news was nevertheless soaked with complaints about the "Brussels bureaucracy," without explaining what the Dutch parties, let alone the Dutch voters, could do about it. Figure 1 illustrates the central puzzle of this article: why did parties not mobilize their voters on the new EU issues in 2014, given the unprecedented amount of EU news in 2014? We focus on the vote for individual parties rather than on turnout, because this will show whether a specific party's profile with respect to the new EU issues can affect the mobilization of voters for that party.

Electoral Consequences of Party Emphasis on EU Issues
Party contestation at the national level is often more attractive for national political parties than competition on EU issues, even in EU election campaigns (Van der Eijk & Franklin, 2004). This is especially true for a multiparty system with coalition governments like the Netherlands. Research from Adam et al. (2014) on party press releases in the twelve weeks that preceded the 2014 European elections shows that only 7% of the Dutch press releases were devoted to international affairs, as compared to 16% of German press releases (Adam et al., 2014). EU issues may drive a wedge within the electorate of a party and within the governing coalition in a multiparty system. Competition on EU issues is primarily used by parties who have never been part of government coalitions ( Van de Wardt et al., 2014). In combination with the tendency of media to concentrate on negative news, this results in the paradox that more media attention for Europe may be detrimental for trust in Europe (Van Noije, 2010;Van Spanje & De Vreese, 2014;Vliegenthart, Schuck, Boomgaarden, & De Vreese, 2008).
The 2005 Dutch referendum on the EU constitution provides an example. Months before the vested political parties started their short pro-EU campaign, newspapers and television news programs came to report extensively about the expected French "no" because of the French fear for cheap labor from Eastern Europe. This huge increase in media attention for the EU long before the official campaign contributed to the shift from a pro-European stance towards the Dutch "no" to the EU constitutional treaty (Kleinnijenhuis, Takens, & Van Atteveldt, 2006). This shows the weak role of Dutch political parties in the news about European affairs. In the terminology of Koopmans and Erbe (2004) the low involvement of national actors in European affairs is described as weak vertical Europeanisation. Almost all news about the euro crisis was either supranational, with reports about the ECB and the EFSF, or horizontal, with extensive reports about street riots in Athens, the Bundesverfassungsgericht in Germany, and long-term interest rates for Spain. The prolongation of the euro crisis culminated in disappointment about EU austerity politics and lower levels of trust (Armingeon & Ceka, 2015). EU news related to Ukraine was also highly horizontal. The media covered soundbites from Euromaidan and from the speeches of the Russian president Vladimir Putin. Images from the battleground in the Crimea and the Eastern Ukraine became an integral part of prime time television news. Contentious European issues like the Ukraine conflict are often covered widely even in the tabloid press (Pfetsch, Adam, & Eschner, 2010). The vertical dimension in EU news was once more weak, presumably because political parties were afraid of their voters. The popular mood was against Putin, but popular resistance against EU membership for the Ukraine could be expected to be even stronger than in the case of Poland, Romania and Bulgaria.
Hypothesis 1 is based on the expectation that if a party strongly emphasizes the EU, the euro crisis, or the Ukraine, these issues would turn into wedge issues (Van de Wardt et al., 2014), chasing off a significant part of its voters. A party's emphasis on an issue, or in other words, a party's involvement in an issue, or a party's association with an issue in the media used by a voter will be indicated by the number of news items in which the party and the issue co-occur in the media used by that voter. H1: News coverage of a party's involvement in the EU (H1a), the euro crisis (H1b), or the crisis in the Ukraine (H1c) in the media used by a voter diminishes that voter's likelihood to vote for the party.

Electoral Consequences of Party Emphasis on Old Issue Dimensions
To test whether the vote at EU elections depends on the media portrayal of the involvement of parties in European issues, news on national issue dimensions has to be considered in addition. In line with theories of issue ownership (Petrocik, 1996;Walgrave, Lefevere, & Tresch, 2012) and theories of issue news effects (Kleinnijenhuis, Van Hoof, Oegema, & De Ridder, 2007;Walgrave, Lefevere, & Nuytemans, 2009) it is to be ex-pected that parties who receive media attention for their issue positions on owned issues, either on the left-right dimension (e.g. taxes, social services) or on the cultural dimension (e.g. immigration, Islam) will profit at the elections. Therefore news coverage in the media used by a voter of a party's stances on the leftright dimension is expected to increase that voter's likelihood to vote for that party. The same is expected to hold for news about the cultural GALTAN dimension (Hooghe, Marks, & Wilson, 2002;Kriesi et al., 2006Kriesi et al., , 2008, although the latter is more often debated (Van der Brug & Van Spanje, 2009).
H2: News coverage of a party's involvement with the left-right dimension (H2a) and/or the cultural dimension (H2b) in the media used by a voter increases that voter's likelihood to vote for the party.
We will test whether addressing these common issue dimensions resulted in additional votes, without testing in more detail whether parties emphasized indeed 'their' side of ideological dimensions in line with issue ownership theory (Budge & Farlie, 1983;Petrocik, 1996;Walgrave et al., 2012).

Electoral Impact of Characteristics of Voters and Parties
We now turn from the supply side of news on EU issues in the media to the demand side of voters who select a party also on the basis of structural factors that play a role in second-order elections, in which votes "are determined more by the domestic political cleavages than by alternatives originating in the EU" (Reif & Schmitt, 1980). Voting on the basis of domestic political cleavages leads however to conflicting considerations.
First of all, many voters vote habitually in European elections (Hobolt & Spoon, 2012), based on the party voted for in the last national elections, thus on the basis of prior vote intentions (Van Spanje & De Vreese, 2014).
H3a: Voters tend to vote in European elections for the same party as in the preceding national elections.
Issue voting may however be more prevalent in European elections than in national elections, since strategic considerations about party size and coalition potential matter less in second-order elections (Hobolt & Spoon, 2012;Reif & Schmitt, 1980;Van der Eijk & Franklin, 1996). In EU elections voters tend to cast a sincere vote for the party whose issue positions they like best. The 2014 elections for the European Parliament were held in the mid-term of the national legislative period in the Netherlands. Therefore voters who are disappointed with the party they voted for in the previous national elections, will presumably vote for a nearby party with which they also agree on the issues (Dassonneville & Dejaeghere, 2014;Kleinnijenhuis & Fan, 1999;Van der Eijk, Schmitt, & Binder, 2005; Van der Meer, Van Elsas, Lubbe, & Van der Brug, 2015). In line with the theory of issue ownership we assume that agreement with a party matters especially on issues that are associated with a party by voters (Walgrave et al., 2012).
H3b: Voters tend to vote for the party with which they agree on the issues that they associate with that party.
Dissatisfied voters in second-order elections tend not to vote, and especially not to vote for government coalition parties (Johnston & Pattie, 2001;Reif & Schmitt, 1980): "Parties in national governments do worse in EP elections, especially when the EP elections take place during the middle of the national election cycle" (Hobolt & Spoon, 2012, p. 703).
H3c: Voters are less likely to vote for a government party than for an opposition party in European elections.
In addition to taking part in the national government, subjective satisfaction or dissatisfaction with the national government has been identified as a major determinant of the vote in second-order elections like the EU elections (Hix & Marsh, 2011;Reif & Schmitt, 1980;Van der Eijk & Franklin, 1996).
H3d,e: The greater a voter's dissatisfaction with government performance, the less likely he or she will vote for any party (H3d), and especially for a government party (H3e).
Education and political knowledge will be included as control variables, since a low level of education and a poor political knowledge lead a lack of enthusiasm to vote for any party in elections for the European Parliament (Lefevere & Van Aelst, 2014).

Content Analysis: Data
The tests of the hypotheses on news effects (H1 and H2) are based on an automated content analysis of news attention from December 2013 until the EU elections at May 22 nd 2014, and on a two-wave panel survey shortly before and immediately after the EU elections. Seven national newspapers (Algemeen Dagblad, Het Financieele Dagblad, Metro, NRC Handelsblad, Next, Spits, De Telegraaf, de Volkskrant and Trouw), a free daily (Metro) and NOS television news from the public broadcaster were included in the analysis. The Netherlands is traditionally a nation with a high readership of national newspapers. Even though this has now dropped to about 50%, in combination with public television news the media analyzed here still reach three quarters of the Dutch voters.

Content Analysis: Operationalization
The automated content analysis is conducted using AmCAT (Van Atteveldt, 2008) using search queries for each of the parties, for the left-right dimension, the cultural dimension, the EU, the financial crisis and the Ukraine conflict. These queries are based on a combination of more elementary concepts, such as crime, and immigration in the case of the cultural dimension. The query formulation procedure that was used to optimize both precision (the percentage of found articles that were correct) and recall (the percentage of all correct articles that was found) gave good results, because the media happen to use fairly unique names and labels to denote Ukraine, the euro crisis, and political parties. The resulting search queries are included in the supplementary materials.
Co-occurrences in a single news items were used to assess whether a party addressed an issue. It should be noted that this measure overestimates the frequency with which parties address an issue, since parties and issues may co-occur in a single news item also for other reasons. Co-occurrence in the same news item of a party and an issue can be conceived as a necessary condition for coverage of a party's issue position on that issue.

Panel Survey Data
The authors commissioned a panel survey to the Dutch branch of GfK, an international market research organization. 1806 respondents for the first wave of the panel study in July 2012 were drawn from a GfK database of over 50,000 respondents that had agreed to participate in GfK-research. The sample of 1806 respondents was effectively a stratified sample that guaranteed that the sample would be not only a representative sample with regard to socio-demographic variables (age, sex, education), but also with regard to turnout and party choice in the 2010 elections. Respondents from this sample were asked to participate in a new wave shortly before (n=1233, response rate 68%) to assess their media use and immediately after the EU elections to retrieve their vote (n=1160, response rate 64%). New voters who were not of voting age in 2012 were excluded. The 2014 sample was still a representative sample with regard to almost all demographic and political characteristics, with political knowledge and turnout as notable examples. Panel attrition occurred significantly more often among respondents with a lower political knowledge (as measured in the first wave of the panel survey in July 2012), which explains why turnout according to the post-election sample (65%) is significantly higher than actual turnout (37.3%)-as is the case in almost every panel survey. Since the remaining variance in education, knowledge and turnout is still large, the panel survey data are still perfectly suited to test explanations of party choice at the European Elections 2 since education and political knowledge can be included as control variables.

Linking Media Content to Respondents in the Panel Survey
For each of the media for which automated content analysis data were available a question was asked in the panel survey whether the respondent had made use of them during the last week. To the users who used a specific medium we attributed the content analysis data for that medium with regard to the emphasis of each party on the EU, the euro crisis, the Ukraine, the left-right dimension and the cultural dimension, in line with for example Boomgaarden, Schuck, Elenbaas, and De Vreese (2011) and Kleinnijenhuis, Van Hoof, and Oegema (2006). To users who used more than one medium we attributed the sum of the attention scores for the media used. Additional news items are expected to show diminishing impact, which is often modeled with taking logs or square roots of the number of news items. In line with earlier research we opted for square roots (Van Noije, Kleinnijenhuis, & Oegema, 2008). The content analysis data that were attributed to each respondent provide the best possible measure of the news about a party's involvement in various issue domains in the media of individual voters, although the measure still neglects which respondent skipped how many relevant news items from the media that were used.

Operationalisation of Panel Survey Variables
Party choice in the 2014 EU elections was measured in the 2014 post-election survey wave (n=1160). Respondents were asked whether they had voted, and if so, for which party. Party choice in 2012 was measured as the party one intended to vote for in the first preelection wave before the national elections of 2012 ra-2 Due to panel attrition the percentage of newspaper readers increased from 51% to 55%, and the percentage voters who either read a newspaper or watched public television broadcasts at least once a week from 73% to 77%. The latter percentages are based on the question whether the respondent used these media during the last week, which still overestimates the actual use. The unweighted percentage of 35% abstainers in the post second-order election survey is a good percentage, that is comparable with the weighted percentage of 45% of abstainers in the 2009 EU elections in the Netherlands, which was obtained by reweighting the data on socio-demographic characteristics (cf. Lefevere and Van Aelst, 2014, Table 2). ther than with a 2014 recall measure. Satisfaction with government policy was measured by a single 5-point scale, that was re-scaled to the -1…+1-value range. Incumbency was measured with a -1…+1-scale, in which the maximum score of 1 was assigned to the government coalition parties PvdA and VVD, a zero to "loyal opposition" parties CU/SGP and D66, and -1 to the remaining opposition parties. Last but not least the agreement on issues between a voter and a party was measured with questions about the association between specific parties and specific issues according to a voter. Respondents were asked: 'Which of the issues below comes to your mind first if you think about < party i >? And which issue next?' Respondents could choose from a list of predefined newsworthy issues and were also able to add other issues. Respondents who associated a party with a specific issue were asked: 'To what degree do you agree or disagree with <party i> with regard to <issue j1 | issue j2>' (Kleinnijenhuis & Pennings, 2001). A 5-point scale was used (disagree fully, disagree, neither agree nor disagree, agree, agree fully), which was linearly transformed to a -1…+1 value range. Overall issue agreement of a voter with a party was measured as the average agreement with a party across all issues that were associated with that party. Note that this measurement in terms of associative issue ownership (Walgrave et al., 2012) applies both to position issues and valence issues (Hobolt & Spoon, 2012). The control variable education and political knowledge were measured respectively as the highest education that one finished and as the number of correct answers to twelve factual questions about the recognition of four politicians from photos. The three questions per politician dealt with their name, their party affiliation, and their political function (Cronbach's  = 0.75). Education and political knowledge were linearly transformed to a 0…1-scale to render their logistic regression coefficients comparable.

Data Analysis
A multilevel logistic regression analysis is applied to test the hypotheses on potential causes of the decision whether or not to vote for any single party. 3 Table 1 gives an overview of the mean scores of the variables that will be used to test the hypotheses. Mean scores are presented for abstainers and for the voters of each of the parties in 2014.

Voter Characteristics, Exposure to Media Content, and Party Choice
Average scores for the abstainers are included in the first row of Table 1 3 A multilevel logistic regression analyses on "stacked" data with combinations of parties and respondents as the units of analysis is to be preferred over a multinomial regression analysis on "wide" data with respondents as the units of analysis. The independent news variables about the emphasis that a party puts on an issue should predict only the dichotomous choice for that party, and should not be allowed to exert all types of side-effects on the decision to make a choice between other parties. This is guaranteed with a multilevel logistic regression analysis, and not with a multinomial regression analysis. The parties are roughly ordered from left (SP) to right (PVV). Habitual turnout as measured by the percentage of a party's 2014 EU voters who voted for the same party at the EU elections of 2009 is lowest for the parties at the extremes, thus for the SP and PVV. Education, knowledge, government satisfaction and news exposure are also relatively low for voters of the SP and PVV. Education is highest for GroenLinks (0.76), but voters for the PvdA exhibit on average the highest news exposure (0.79) and the highest political knowledge (0.92). The final five columns in Table 1 show average exposure to political issues of voters in the EU 2014 elections for each party based on their media use. They are not based on all voters. Table 1 shows, for example, that PVV voters, given their media use, could have encountered on average 37 news items in which the PVV played the drum of cultural issues-e.g. the Islam-from December 1st 2013 until the elections on May 22nd 2014. A comparison per row shows that the PVV addressed the cultural dimension more often than any other issue according to the media that were consumed by PVV voters. A comparison per column shows that the government coalition parties PvdA and VVD addressed the cultural dimension even more often according to the media that were followed by PvdA-voters, respectively VVD-voters. The left-wing PvdA focuses slightly more on the left-right dimension than on the cultural dimension (43 vs 40) whereas the reverse holds for the VVD (40 vs 41). The government parties take the lead also in addressing EU issues. Table 1 shows that the media that were used by voters of the opposition parties SP, GroenLinks, and ChristenUnie do not pay a lot of atten-tion to the new euro crisis or the Ukraine. Table 2 shows logistic regression coefficients that represent the effects of news about the issue positions of parties on the vote for a party. Model 1 is the empty model that is included to enable a comparison of goodness-of-fit measures AIC and DIC. Model 2 includes only voter and party characteristics (hypothesis 3). Model 3 includes also the effects of the emphasis of parties on the left-right dimension and the cultural dimension according to the media (hypothesis 2) and the effects of a party's emphasis on the euro crisis and the Ukraine according to the media (hypothesis 1).

Assessing the Effect of EU Related News Controlled for Other Factors
The decreasing AIC and DIC scores show that model 3 fits the data better than model 2, which in turn fits the data better than model 1. This implies that news about party positions on the EU partially explain EP vote choice, also when controlling for news about party positions on the left-right dimension and the cultural dimension, and for structural characteristics of voters and parties.

Controls for Structural Determinants of the Vote in Second-Order Elections
The direction of the significant regression coefficients in both model 2 and model 3 in Table 2 confirm expectations that voters consider national political cleavages when voting in second-order elections. The decision to vote for a specific party is influenced strongly by habitual voting (Franklin & Hobolt, 2011), as measured by whether one voted already for the same party in the 2012 national elections (H3a). Voting by heart on the basis of agreement on political issues with the party to be voted for matters strongly in second-order elections (H3b). Voters who are disappointed with their previous party will not vote for an arbitrary other party but for a party with which they agree on political issues (Dassonneville & Dejaeghere, 2014;Kleinnijenhuis & Fan, 1999;Van der Meer et al., 2015). Satisfaction with government policy reveals the straightforward interaction effect that especially parties that take part in the coalition part will be rewarded in case of satisfaction and punished in case of dissatisfaction, in line with the literature on retrospective voting (Johnston & Pattie, 2001;Van der Brug, Van der Eijk, & Franklin, 2007) (H3c, H3d and H3e). The controls for levels of education and knowledge show that these important predictors of turnout in second-order elections (Lefevere & Van Aelst, 2014) increase the likelihood to vote for each of the parties.

Effects of Political News
Model 3 shows that news matters along with these structural determinants of party choice in second-order elections. The significant logistic regression coefficients show that reports in the media used by voter about party positioning on the left-right dimension (H2a) and on the cultural dimension (H2b) increase the likelihood to vote for these parties. The most likely underlying mechanism is that voters will reward parties who succeed in getting media coverage for their owned issues, which are usually either left or right (Budge & Farlie, 1983), or either Green, Alternative and Libertarian or Traditional, Authoritarian and Nationalist (Hooghe et al., 2002). Because national parties are not portrayed as powerful players in EU news, we expected a negative effect on the vote of the news about parties addressing the EU, the euro crisis, or the Ukraine conflict. No effect shows up for addressing the EU (H1a). Negative boomerang effects show up for addressing the euro crisis (H1b) and the Ukraine conflict (H1c), as is indicated by the significantly negative logistic regression coefficients. Thus, hypothesis H1 is confirmed, and the puzzle why parties did not mobilize their voters on the new EU issues in spite of an explosion of EU news is solved. In the news effects model 3 the direct negative effect of incumbency on the vote in second-order elections (H3c) becomes insignificant, which suggests that the negative effect of incumbency on the vote is mediated by involvement of the governing parties in news about EU related issues, which makes them unpopular.
We now turn to the random part to assess the variance of regression coefficients. The variation in the random intercept across respondents is remarkably small. We tested also a model with random intercepts across parties and a random slope model with partyspecific habitual voting, which showed the same positive and negative signs for the regression coefficients. 4

Conditional Effect of News about a Party's Stance on the Ukraine
Multilevel logistic regression estimates often give a poor impression of the marginal effects of separate variables in the model, even when different explanatory variables have the value range as in Table 2. To illustrate effect size, Figure 2 presents a linear plot of the effect of a party's emphasis on the Ukraine on the probability to vote for that party. The X-axis shows the association between a party and the Ukraine in the news followed by voters of that party, while the Y-axis shows the logarithmic transformation of the probability to vote for that party, with all other variables set to their means. The distribution of news about the profiles of parties is shown on the x-axis by means of small, more or less densely plotted, vertical bars.
Even when controlled for other variables the probability to vote for a party decreases from 20% for a theoretical party that did not address the Ukraine at all, down to far less than 1% for a theoretical party that addresses the Ukraine at every occasion.
For a further interpretation of Figure 2 it is worthwhile to consider the average emphasis of individual parties according to the news media that were followed by their voters on the x-axis, as presented in the last column in Table 1. Given average values on other variables, VVD voters would have been assigned a probability of roughly 1% only to vote for the governing VVD given the high amount of Ukraine related EU news (n=22 news items on average) about the VVD. The likelihood to vote for the Christian-Democrats (CDA), the Socialist Party (SP) or GroenLinks would amount to almost 5%, since these opposition parties were more able to avoid the Ukraine (n=8, 9 news items, respectively). All in all Figure  2 shows that the negative effect of party-related news on the Ukraine on the EU vote is quite substantial. Figure 2. Effect of number of news items about a party and the Ukraine in one's media on the probability to vote for party (logarithmic scale). Note: The effects are conditional on mean values for the remaining independent variables and for the random intercept in the multilevel regression model of Table 2. The grey area denotes the 95% confidence interval.

Discussion
One can wonder why the explosion of EU related news on the euro crisis and the Ukraine conflict before the 2014 EU elections hardly resulted in a higher enthusiasm for the EU positions of political parties, as shown by the very low increase of turnout (cf. Figure 1). The puzzle why this did not occur in 2014 can be solved by looking at the micro-level effects of media coverage on the individual vote.
The current study confirms that issue voting in a European context matters (e.g. Hobolt & Spoon, 2012;Van de Wardt et al., 2014). The study shows news effects, in line with the literature on effects of the visibility and the tone of EU related news (e.g. Azrout et al., 2012;Van Spanje & De Vreese, 2014). The unique contribution of this study is to provide empirical evidence that amplification of parties' issue positions on EU related issues in the news media actually diminished electoral support. The emphasis of parties on the euro crisis and the Ukraine did not motivate voters but scared them off, when controlled for structural characteristics of voters and parties, and for addressing the traditional left-right dimension and the cultural dimension. The result that a party's emphasis on the left-right dimension and the cultural dimension in the media motivates voters to vote for that party is in line with survey research that established the importance of the left-right dimension in EU elections (Hobolt & Spoon, 2012). The result that the EU's relation with Ukraine and the euro crisis can't be addressed by parties in the news media without losing voters is in line with survey research which showed that the EU is a wedge issue in multiparty systems (Van de Wardt et al., 2014).
A limitation of this study that we focused on issue news and on retrospective voting based on satisfaction with government policy, but not on news about support and criticism, and cooperation and conflict, or about success and failures, losses and benefits, and the horse race, which also exerts effects on the vote (Kleinnijenhuis et al., 2007;Schuck et al., 2016;Van Spanje & De Vreese, 2014).
Political parties were not portrayed in the media as relevant players in the Euro crisis and the Ukraine conflict: the 'vertical' dimension of Europeanization was weak. This points in the direction of a vicious circle between a low visibility of national parties in EU related news and electoral losses for parties who relatively strongly emphasize EU related issues, most often incumbent government parties. Months of prolonged news about path breaking party stances on EU related issues, such as the 2015 EU immigrant crisis, can possibly offer an escape from such a downward cycle, and create the momentum to break the vicious circle.