UvA-DARE (Digital Academic Repository) Does politicisation matter for EU representation? A comparison of four European Parliament elections

What is the effect of politicization on EU representation? We empirically test two competing views through a focus on party voter congruence, namely whether parties share the policy preferences of – their voters. The rst perspective expects that the process of politicization either through party or  – media contestation would improve party voter agreement. The second perspective argues that – – politicization has not improved the supply of partisan debate with parties still not offering satisfactory options to voters who are primarily driven by protest based considerations; and as such does  not expect an effect. We analyse congruence on the left right and pro anti EU dimensions, – –  capturing questions related to EU policy and polity, respectively. Our examination of four European Parliament elections (1999 2014) and 341 parties across 53 electoral contexts points – to the limited effect of politicization upon representation. Our ndings have signi cant implica-  tions for the study of EU politicization and representation and open up avenues for future research.


Introduction
Politicization is a key development in European Union (EU) politics (De Wilde et al., 2016).The literature on its consequences has focused on the structure of political con ict in Europe and the process of European integration (for example Hooghe and Marks, 2009, 2018;Hutter et al., 2016).While some posit that politicization contributes to a lasting transformation of the structural basis of European integration (De Wilde and Zürn, 2012;Hooghe and Marks, 2009), others argue that it has not been able to systematically restructure political con ict (for example Börzel and Risse, 2009;Hutter  et al., 2016).This article contributes to this debate through an analysis of the effect of EU politicization on representation in the European Parliament (EP), which forms the direct channel of representation within the EU s constitutional framework.We ask: if ' there is some change permanent or otherwise in the nature and scope of politicization -in Europe, to what extent is this also re ected in the patterns of political representation in  the EU polity?
We empirically test two competing views on the potential effect of politicization on representation.On the one hand, primarily normative oriented research has considered  politicization to be a key force behind the legitimacy of the European project (for example Føllesdal and Hix, 2006;Schmitter, 1969;Statham and Trenz, 2015).The process of politicization either through party or media contestation would entail parties taking --alternative positions on the EU policy and polity dimensions; voters updating their policy preferences and opting for parties that are close(r) to them; thus, resulting in better partyvoter agreement.On the other hand, empirically oriented scholars have shown that  politicizationv has not created a coherent political environment (for example De Wilde et al et al ., 2016;Hutter ., 2016).Politicization has not improved the supply of partisan debate, and EP elections are primarily driven by protest based considerations, which  suggests that politicization has no effect on party voter agreement.
-Our contribution is threefold.First, this article takes a pioneering step at examining the consequences of politicization for the process of European integration through the prism of representation.Understanding whether politicization has an effect on the linkage between political parties and voters is a missing piece in the literature given that ' ' they are key actors and drivers of political contestation.Second, empirically, we focus on party voter ideological congruence, namely the extent to which parties share the preferences of their voters during EP elections on the left right and pro anti EU dimen-- sions.The left right dimension involves policies that are made through the EU in the -Council and the EP (Börzel and Risse, 2009;Braun ., 2016;Hertner, 2015, p. et al 471).The pro anti EU dimension captures debates related to the constitutional frame- work and institutional structures of the EU and thus concerns questions about the EU polity (Norris, 1997, pp. 277 278).We examine congruence in four EP electoral con-tests (1999, 2004, 2009 and 2014).These elections cover a time point prior to the EU s ' enlargement to Central and Eastern Europe and each election prior to, during and after the Eurozone crisis.We focus on the EU 15 as Western European party systems tend to  be more stable and institutionalized compared to their Central and Eastern European counterparts.We analyse congruence on both the left right and pro anti EU -- dimensions for a total of 341 parties, by employing a sophisticated measure of relative congruence (Golder and Stramski, 2010).This allows us to provide a comprehensive picture of representation in the EP on both dimensions across political contexts and over time.Third, our ndings point to the limited effect of politicization upon represen- tation in the EP and thereby support the second, more sceptical, perspective.This nd- ing is important because it serves to qualify the academic debate over the consequences of politicization for European integration.From the lens of representative politics, politicization has neither a positive nor a negative effect on the integration process.This informs claims made by scholars taking a competitive elitism perspective (Føllesdal and Hix, 2006;Hix, 2006) and Habermasians (Statham andTrenz, 2013, 2015;Trenz and Eder, 2004) who view politicization as a positive precondition for further integration.It also quali es the post functionalist approach (Hooghe and   Marks, 2009) that views politicization as potentially aggravating the problems of European integration (see also Hutter ., 2016 for a discussion).et al

I. Politicization and EU Representation
Politicization is a multidimensional concept that refers to the expanded role of political con ict in a political system (Hutter andGrande, 2014, p. 1003).In the EU, politicization  tends to be considered a consequence of authority being transferred from the domestic to the EU level (De Wilde and Zürn, 2012;Hooghe and Marks, 2009).Concretely, politicization relates to the extent to which decisions at the EU level are publicly contested and debated; and consists of three key dimensions, including issue salience, actor expansion and polarization (for example De Wilde ., 2016;Hutter ., 2016).Issue et al et al salience hinges upon the visibility of a given issue in public debates.Expansion of actorsv draws attention to the number of actors involved in the con ict with politici- zation being low if only a limited number of actors engage with the issue.Finally, polarization refers to the intensity of con ict over the given issue.For polarization  to occur, actors need to advance opposing stances and offer different solutions to the issue (Grande and Hutter, 2016;Hutter and Kriesi, 2019).Empirically, EU politicization has been studied in the national political arena (Hutter and Grande, 2014;Kriesi, 2016, p. 32) and also during EP elections (Meijers and Rauh, 2016), not least because literature suggests that there are spill over effects of EU politicization from the national  to EP electoral arenas (Ares ., 2017).In this article, we focus on EP elections et al which provide an appropriate context because they are organized at the national level with domestic parties competing for seats in the EP and are known for their second order character (Reif and Schmitt, 1980).
 Prior research on EU representation suggests that party voter agreement on the left -right dimension and related policies has been relatively high (for example Mattila and Raunio, 2006;Vasilopoulou and Gattermann, 2013).Congruence on the EU dimension was also a key empirical nding in the 1980s and 1990s, which was attributed to the  low salience of the EU issue (for example Schmitt and Thomassen, 2000;van der Eijk and Franklin, 1991).Mattila andRaunio (2006, 2012) questioned this proximity in the 2000s, showing that in 2004 parties were closer to their voters on the left right dimension compared to the EU dimension, and that in 2009 voters and parties were drifting further apart on the EU dimension.Dolný and Babo (2015), however, found that in 2009 š congruence on the EU dimension was higher than on the left right dimension.Taken together, although these works offer important insights into the study of EU representation, they do not provide a systematic analysis of EU representation.They examine party voter congruence at one point in time; mostly focus on one dimension (either left right or EU); and employ different operationalizations of representation, which hinders the comparability of their ndings.In this article, we provide a thorough account  of representation in the EP on the pro anti and left right dimensions across countries and  over time.
To what extent does politicization have an impact on congruence?Our aim here is to unpack the key propositions from two leading perspectives in the literature on the potential consequences of politicization on representation.The rst approach has its roots in  early neo functionalist thinkers of European integration (Haas, 2004;Schmitter, 1969). Schmitter (1969, p. 166, italics in the original) puts forward a politicization hypothesis whereby the 'controversiality of joint decision making goes up , which in turn is likely  ' ' to lead to a interested and active in integration .This widening of the audience or clientele ' con ict would result in a rede nition of mutual objectives and ultimately there would be  '  ' a ' ' shift in actor expectations and loyalty toward the new regional center .These authors did not directly address the issue of representation; but essentially suggested that contestation about the EU would lead to higher support for integration, serving to legitimize the EU project.Drawing from this perspective, more recent scholarship suggests that EU politicization can address some of the weaknesses of the system (for example Føllesdal and Hix, 2006;Statham and Trenz, 2015).
Based on the competitive model of representative democracy (Schumpeter, 1943), Føllesdal and Hix (2006) foresaw the following mechanism: contestation fosters political debate and deliberation.Democratic arenas for contestation where issues are deliberated provide incentives to parties to develop and promote alternative and competing policy packages, which in turn becomes a vehicle for public opinion formation.In this -context, voters would have the opportunity to choose between rival candidates and rival policy agendas.The logical sequence of this argument is that given the variety of policy options on offer, voters would be able to opt for parties that are close(r) to their preferences.Hence politicization would improve representation.In their words, contestation would lead to a greater connection between voters preferences and coalitions and ' ' alignments in the EU institutions (Føllesdal and Hix, 2006, p. 553).This logic underpins ' the authors support of open competition for the (s)election of the Commission President.
' This process would contribute to contestation about politics , not only , the EU ' in of ' (Føllesdal and Hix, 2006, p. 554, italics in original).It would promote citizens' policy learning and information, which would allow Europeans to better formulate their  preferences on speci c issues (Hix, 2006).
 From a Habermasian public sphere perspective, politicization is also assumed to improve decision making and democratic performance.Democratic functionalism, in par- ticular, suggests that mass mediated public debates and Europeanized public spheres are  essential because they create an iterative process of learning, which enables citizens to update their preferences and make informed decisions.This is where the public gains access to information about party stances and executive decisions.The presence of a public shapes the behaviour of political actors who mobilise their demands with the aim of ' gaining public support for their stances (Statham and Trenz, 2013, p. 969).In other ' words, media contestation over Europe is associated with better informed citizens who  can make better choices.At the same time, the presence of a public provides feedback to political actors who make their positions more transparent (for example Statham and Trenz, 2015;Trenz and Eder, 2004).In short, mediated politicization allows for greater interaction between politicians and the public, which is likely to improve representation.
Taken together, these primarily normative oriented perspectives would expect that the  process of politicization either through party or media contestation would lead to -parties taking alternative positions on the EU policy and polity dimensions; voters updating their policy preferences and opting for parties that are close(r) to them.This suggests that politicization results in better party voter agreement (H1).
-The second approach derives from empirically oriented literature and offers a differen- tiated view on the potential impact of politicization on EU representation.For politicization to result in higher levels of representation, two processes need to be taking place: clarity and affect.Clarity relates to citizens becoming aware of this politicization and affect means that they care about it, and thus react to it in their behaviour.
A number of empirical accounts cast doubt on the question of clarity.First, there is a debate in the literature with regard to the level and intensity of politicization.Whereas some scholars argue that the giant is still sleeping (Green Pedersen, 2012; van der Eijk ' '  and Franklin, 2004), others posit that since Maastricht European integration has entered a new phase of controversial mass politics (Hooghe and Marks, 2009).More recently, ' ' empirical research has shown that politicization is a punctuated and differentiated process, which exhibits strong variation across time and space (De Wilde et al., 2016;Hutter and Kriesi, 2019).In this context, politicization is considered as a set of strategic options for ' political actors (Grande and Kriesi, 2016, p. 291), which implies that parties may not ' consistently put forward rival policy agendas, and as such collective learning processes may not necessarily be enduring.Second, and related, politicization increases uncertainty for mainstream political elites.Their lengthy participation in government suggests that they have been heavily involved in the making of the EU polity (see Hooghe and Marks, 2009, p. 21).If they shifted their EU position, they would risk undermining their own trustworthiness thereby enhancing the credibility of their competitors (Rohrschneider and White eld, 2016).Empirical evidence indeed shows that mainstream parties tend to  stick to their programmatic positions notwithstanding positional shifts among voters (Hooghe and Marks, 2018).Fraught by internal dissent, they generally tend to downplay European issues, which undermines clarity (Braun ., 2016;Mattila and Raunio, 2012, et al p. 590).Along these lines, Meijers and Rauh (2016, p. 83) nd that increased politiciza- tion in EP elections has not been associated with enhanced and more interactive supply ' of partisan debate about Europe .Although mainstream government parties pay some ' attention to the EU issue in national parliamentary debates (Rauh and De Wilde, 2018), they have limited incentives to electorally politicize the EU (Green Pedersen, 2012)  and opt instead for de politicization and technocratic solutions to avoid contestation  (Grande and Kriesi, 2016, p. 290).
In terms of affect, there is also no clear indication that the public responds to politicization.The rst issue here is the quality of information supply, namely the type of informa- tion available that allows citizens to form their preferences.Coverage about the EU s ' involvement in policy making tends to be super cial (Hobolt and Tilley, 2014, p. 143).
  We also know that MEP visibility in domestic media depends more on who these MEPs ' are in terms of of ce, seniority and status and less on what they do in the EP in terms '  ' ' of legislative activity (Gattermann and Vasilopoulou, 2015), which does not contribute to citizens better understanding of EU politics.Second, we know that citizens increasingly ' assign responsibility to the EU for policy outcomes; while at the same time being aware of its weak accountability structures (Hobolt and Tilley, 2014).Citizens cannot directly hold EU executive politicians to account.Some citizens respond to this limited political choice by defecting from mainstream to challenger parties, especially if they are more Eurosceptic than the mainstream parties (Hobolt and Tilley, 2016).This could be partly driven by ideology; however, literature also shows that protest based considerations regarding  government approval are crucial to explaining vote switching in EP elections, which  undermines the scope for improved representation (Magalhaes, 2016).In fact, EP elections remain second order.Although the salience of the elections in the media has increased,  citizens have not become more engaged and knowledgeable about the EU and thus are not better at judging what parties are offering (van der Brug and de Vreese, 2016).Likewise, the EP s lead candidate ( ) initiative has not substantively ' Spitzenkandidaten altered the second order nature of EP elections (for example Gattermann ., 2016). et al This discussion points to the following mechanism: sporadic, punctuated and differentiated politicization does not necessarily create a stable and coherent political environment that would lead to better party voter agreement.Politicization has not improved the supply of partisan debate, voters are not necessarily better informed, and EP elections are primarily driven by protest based considerations.This suggests that politicization has  no effect on party-voter agreement (H2).

II. Data and Methods
To evaluate the theoretical framework presented above, we designed a new dataset in which our unit of analysis party voter congruence is nested in political contexts, --which vary across time and space, namely four EP elections in 15 EU member states.There are two important considerations, namely comparability and coherence in the observations, which we accounted for in our approach.To address the rst issue, we  operationalize our main dependent and independent variables with relative measures at the party system election level.Given that the interpretation of the left right and pro anti  - EU dimensions may differ between elections and countries, relative measures allow us to compare patterns of congruence and politicization across political contexts.To address the second issue, the calculations of our main variables are based on a single data source, namely the post election voter surveys of the European Election Studies (EES) of the  years 1999, 2004, 2009 and 2014

Dependent Variables and Model
EES data allow us to identify the political parties individual voters had voted for, the individual self placement of these voters on the left right dimension and their attitudes  towards European uni cation, as well as the placement by the respondents of all parties  on the same scales.We calculate our dependent variables as relative voter congruence (RVC) on the left right and pro anti EU dimensions.This measure is based on Golder -- and Stramski s (2010, p. 96) relative citizen congruence, which is operationalized as ' '   ' many to one relationship.Contrary to a simple mean difference between party and voter positions, RVC considers the distance between the ideological position of the individual voter and her party, in relation to the dispersion of the preferences of all voters having supported the same party.It is thus suitable to compare levels of congruence across various country contexts at different points in time: where V ij is the ideal point of the th voter of party , and MV i j j is the ideological position of the median voter position of all voters ( ) of party N j.P j is the party s ideological ' position, which we calculated by taking the means of party evaluations per country and election year.All respondents were asked to evaluate all parties regardless of their party political preferences. 1 However, we consider only the 40 per cent most educated respondents across the whole political spectrum as opposed to the entire electorate to ensure the validity and reliability of our dependent variable.This is because less educated respondents are more inclined to opt for the middle position on an evaluation scale rather than admitting to not knowing the answer if this is the case (see also Alvarez and Nagler, 2004, p. 50;Golder and Stramski, 2010, p. 98).This tends to result in more centrist evaluations of the mean placement of each party.Mean comparisons between the two different operationalizations indeed show signi cant differences, with the measures based on all  voters party evaluations being closer to the mean position of the standardized scales ' (Table A4).As a robustness check, we calculated a second measure of our dependent variables based on all voters party evaluations.
' 2 We report the respective models in Online Appendix D. In these models, we also calculated those independent variables, which take party positions into account (see below) based on all voters party evaluations.
' The results are similar to those that we report below; any differences are reported in the footnotes.
While we acknowledge that there are many ways to measure party positions, including expert surveys and party elite surveys (for example Belchior, 2013;Rohrschneider and White eld, 2016;Schmitt and Thomassen, 2000), we agree with Dalton and  McAllister (2015, p. 767) who argue that the public s perceptions of the parties create ' ' the basis of their voting choices even if these perceptions are imperfect , which is -' important for our argument that relates to congruence (see also Mattila andRaunio, 2006, 2012;van der Eijk and Franklin, 1991).This would be different if we studied responsiveness among parties and voters as the latter do not necessarily adapt their evaluations of party positions when these change (for example Adams ., 2011).An additional ad-et al vantage of calculating the mean party position in this way is that voters use the same scale as for their self placements and at the same point in time (Golder and Stramski, 2010,  p. 99).This would be different had we relied on expert surveys or party manifestos.
We have two measures for RVC; one for the left right dimension (RVC -LR ), and a second for the pro anti EU dimension (RVC  EU ).Our unit of analysis is ideological congruence per party in each country and each election for the EU 15 member states  ( = 341).Both variables range from 0 to 1 (RVC N LR : = 0.09; SD = 0.11; RVC M EU : M = 0.10; SD = 0.10).0 stands for perfect ideological congruence between voters and the party they voted for.Higher values indicate that their ideological positions are further apart from each other.As the low means and the standard deviations of our dependent variable suggest, the distributions are skewed right.A linear regression model is not appropriate as it may yield tted values for the variable of interest that exceed its lower '  and upper bounds (Ferrari and Cribari Neto, 2004, p. 799).We thus choose a beta regres-'  sion model, which is suitable for the structure of our dependent variable (0 1) and < < y can also accommodate skewed distributions. 3 Our data do not have a precise panel structure.The number of cases increases with each election year, ranging from 80 parties in 1999, over 79 in 2004 and 84 in 2009, to 98 parties in 2014 covering a total of 53 party systems in all election years (see Table A1).Given this variation, we cluster our robust standard errors by election year and party systems.
2 T tests reveal no differences between the two operationalizations on the pro anti EU dimension.However, our variable on  - the left right dimension is signi cantly smaller ( - M difference = 0.01, 0.001) compared to the measure based on all voters p < (Table A5). 3 In accordance to the model speci cation, we have transformed 0 values to 0.000001. Does politicization matter for EU representation?
Furthermore, we are unable to include the same parties and countries consecutively each year (for an overview see Tables A2 and A3).This relates, rstly, to the calculation  of our dependent variables.Some of the original variables concerning self placements,  party placements and education have missing values.Our data thus excludes the entire samples of Belgium and Sweden in 2004.Secondly, our case selection is also determined by one major independent variable reported below, namely party polarization.Since this variable accounts for party positions of all coded parties in a given party system/year, we require as much information as possible on these other party positions.However, our dataset includes only few parties for some election contexts.We therefore decided to only include party systems per election if the coded parties together make up at least 80 per cent of the total vote share.This has led to the exclusion of seven additional party systems/years (see Table A3).Thirdly, in some countries electoral coalitions have changed over time, such as in Italy, or they have not been treated consistently as coalitions or individual parties in different EES waves, such as in Germany or the Netherlands.In these cases, we have values for their various coalitions or for individual parties at different points in time.Lastly, we only consider parties for which there were at least ten respondents in the EES.Despite the missing values, our data consist of a comparable number of parties and party systems for each election year (see Table A1).

Independent Variables
Our measures of politicization are directly linked to the de nition of De Wilde  et al. (2016, p. 4) that focuses on three dimensions, namely polarization of opinionwhich according to the authors is often operationalized as polarization at the party system level (De Wilde ., 2016, p. 7), salience and the expansion of actors and audiences.et al Likewise, Hutter and Grande (2014, p. 1003) conceptualize politicization as being manifest in increased actor polarization, rising issue salience and expansion to non governmental actors.We operationalize politicization as follows: rst, we consider   all three components separately and measure (actor) polarization by party system dispersion; salience by media visibility; and actor expansion as focus on EU level actors  in news coverage.We do this to examine whether any of these variables has a distinct effect that may drive the effect of the composite measures (see below).
Party polarization uses the measure of weighted party system dispersion by Schuck et al. (2011, p. 45), who themselves rely on Alvarez and Nagler (2004, p. 50): where VS jk is the vote share and P jk is the position of party in country either on the EU j k dimension, and P k is the weighted mean of all party positions in country .To measure salience and expansion of actors to non domestic and nongovernmental  actors, respectively, we include two variables from De Vreese and Azrout (2019) who conducted quantitative content analyses of newspapers during the three weeks leading up to Election Day in each country.( = 0.09; = 0.06) is the proportion EU visibility M SD ' of articles [on the front page] that mention the EU, or one of its institutions, at least once' (De Vreese and Azrout, 2019, p. 8).Europeanness of EU news M SD n ( = 0.32; = 0.20; = 328) measures the proportion of articles that have an EU actor as main actor on front ' ' pages (De Vreese and Azrout, 2019, p. 8).The latter variable has missing values for Belgium and Luxembourg in 1999.
A second way of operationalizing politicization is to relate the three different components to one another.On their own, the indicators may not explain much.Rather, polarization may have an effect when EU issues are salient and/or EU actors become visible.We rely on Hutter . ( 2016 Table A6 compares the mean values of these system level variables over time.While  there are no considerable and linear differences with respect to party polarization, the mean EU visibility scores are signi cantly higher in 2014 compared to the previous  years, and the effect is rather sizable with η 2 = 0.484.Similarly, the Europeanness of news is highest in 1999 and lowest in 2004, after which mean scores are slightly increasing (η 2 = 0.264).

Controls
Politicization is a strategic opportunity for EU issue entrepreneurs (De Vries and Hobolt, 2012).We created a dichotomous variable which differentiates between challenger parties (0) that have not been in government and mainstream parties (1) that have (De Vries and Hobolt, 2012).Most of our parties are mainstream parties (60.7%) (Table A7).Drawing upon the literature on party voter congruence (for example Belchior, 2013;Mattila andRaunio, 2006, 2012), we also add the following party level variables:  Vote share is operationalized by the party s vote share (or that ' of an electoral coalition) in the respective EP election in a given country.We expect smaller parties to be more representative as they tend to be more ideologically homogenous (Mattila and Raunio, 2006).
is calculated from the year of Party age establishment of the party.Congruence between voters and older parties is likely to be better compared to newly established parties because voters are able to rely on historical information when assessing party positions in order to decide which party to vote for.
At the system level we have to be parsimonious given the relatively small N. We calculated the measured by the number of seats they received effective number of parties in the EP following the respective election on the basis of Gallagher s (2015) index.Party ' system fragmentation is likely to result in higher voter party congruence.Furthermore, we account for variation in over time and across country.Lastly, as politicization turnout patterns might differ across regions, we include a dummy for North Western Europe, which is 0 for the remaining South European countries plus Ireland (Hutter and Kriesi, 2019).Table A7 includes all descriptive statistics.

III. Results
Mean comparisons indicate that aggregate levels of party voter congruence on both dimensions range between 0.078 and 0.113 (Table A8).Higher values indicate less congruence; and RVC LR does not exceed 0.1 (1999) and has lowest values in 2009 (0.078); RVC EU is highest in 1999 (0.113) and lowest in 2014 (0.082).It is dif cult  to assess the extent to which these levels of congruence can be considered satisfactory, as the literature does not provide any thresholds that would indicate whether congruence is suf cient or even good.Golder and Stramski (2010, p. 101) analysed data from 36  countries between 1996 and 2005 from the Comparative Study of Electoral Systems project.They found higher values for relative congruence between governments and citizens, that is the same measure that we applied for party voter congruence, in both proportional ( = 0.17, = 0.13) and majoritarian systems ( = 0.17, = 0.14).In M SD M SD other words, our results indicate that EP parties and their voters are comparatively closer on both the left right and EU dimensions.
-To investigate these patterns we turn to our explanatory analysis.The results from the beta regressions are reported in the following gures.Figure 1 contains the average  marginal effects on congruence on the left right dimension (Model 1) and congruence on the pro anti EU dimension (Model 2).They include the individual measures of  politicization, namely party polarization, EU visibility and Europeanness of news.The subsequent gures visualise similar models, but include the composite politicization  measures A and B that account for all three components (Figure 2) or only salience and polarization (Figure 3), respectively.
Figure 1 shows that party polarization over left right or EU issues does not have any effects on congruence on either dimension.Actor expansion, which we operationalized as Europeanness of news, does not seem to matter either for better or worse levels of congruence be that over policy issues or matters related to the polity.-Likewise, the salience of EU issues (EU visibility) has no effect on party voter congruence on either dimension.There are no differences between mainstream and challenger parties.Moreover, as Figure 2 demonstrates, politicization if measured by multiplying salience with the sum of actor expansion and party polarization (Hutter et al., 2016) does not have any effect on party voter congruence on either -dimension.Likewise, our second operationalization of politicization which multiplies salience with party polarization (Hutter and Kriesi, 2019) also has no effect (Figure 3).The control variables do not often reach conventional levels of signi cance.However,  the effects of party age, turnout and effective number of parties point to the expected direction.

Conclusion
In this article, we sought to uncover the relationship between politicization and representation through an empirical analysis of party voter agreement in four EP elections (1999, -2004, 2009 and 2014) in the EU 15 member states.We tested two opposing views on this  relationship.The rst perspective expects that the process of politicization either  through party or media contestation would improve party voter agreement (H1).The -second perspective, on the other hand, does not expect an effect (H2).Our ndings are  in line with the second hypothesis.This has implications for the study of EU politicization.It points to the potential limitations of approaches that view politicization as a positive precondition for further integration (for example Føllesdal and Hix, 2006;Hix, 2006;Statham andTrenz, 2013, 2015).It also informs the perspective that considers politicization as potentially aggravating the problems of European integration (Hooghe and Marks, 2009).
Parties have been slow at adjusting to the politicization of EU affairs and have not been successful at communicating this to their voters.If anything, the sorting of parties and voters into like minded alliances is transitory and very likely to change.Maier  et al. (2019) show that the media rather than parties set the agenda on EU related -- issues even during campaigns, and that there is no linear relationship between an EU issue s media visibility and the extent to which political parties react to it.It is ' conceivable that parties require more time to adjust to new political realities in order to be able to communicate this to their voters, who in turn would need to update their preferences and re align.Scholarship should thus examine the effect of EU politicization  on representation in future EP elections.While we agree that mainstream parties are programmatically in exible (Hooghe and Marks, 2018) and our ndings support this  - to a large extent our results also demonstrate that challenger parties are not closer to their voters compared to mainstream parties.Although both sanctioning and ideological selection processes matter when casting a vote for challenger parties (Hobolt and Tilley, 2016), our ndings on congruence in contexts of politicization also lend support  to theories of protest voting, and more broadly to the second order EP elections model  ( Reif and Schmitt, 1980).This suggests that the new transnational cleavage (Hooghe and Marks, 2018), which has been in the making during the last decades, is shaped by both ideological and protest based considerations.
 We are con dent about our results for a number of reasons.First, and regarding our  measure of the dependent variable, the aggregate analysis suggests that levels of congruence are relatively high ranging between 0.078 and 0.113.Werner, (2020, p. 133) demonstrates that one needs to be cautious when interpreting levels of relative congruence as the distribution tends to be slightly biased towards lower values, namely better congruence.According to Werner (2020) this would also partially explain why the literature nds rel- atively high levels of congruence between representatives and represented (for example Belchior, 2013;Mattila andRaunio, 2006, 2012;Vasilopoulou and Gattermann, 2013).Nonetheless, we are able to compare our results to existing studies that rely on the same measure.Our values are slightly lower (namely better) than congruence between citizens and governments in the 36 countries studied by Golder and Stramski (2010, p. 101) between 1996and 2005. This resonates with Dalton (2017, p. 618) who argues that at the aggregate level whether partisans as a collective are close to their chosen parties  ' positions little has changed over time in EP elections.He nds that the basic pattern   ' is that like minded voters and parties are able to connect, which is an essential aspect  of democratic representation .' Second, high levels of congruence entail limited variation to be explained by party and party system level theoretical propositions.Nonetheless, our measures of politicization  are in line with the literature.We relied on the conceptual de nitions of politicization  provided by De Wilde . (2016, p. 3) and Hutter and Grande (2014, p. 1003) and ap-et al plied two operationalizations of the composite measures (Hutter ., 2016;Hutter and et al Kriesi, 2019) alongside the individual items.While the latter authors also measure salience and actor expansion by a content analysis of European news outlets, their precise operationalization differs from De Vreese and Azrout (2019).We opted for De Vreese and Azrout (2019) given that their coherent and comprehensive database covers the EU 15  member states in the 1999 to 2014 EU elections.Our measure of polarization (Alvarez and Nagler, 2004) is well established and has previously been applied in the context of  EU studies (for example Gattermann and Hefftler, 2015;Schuck ., 2011).Similar et al to existing scholarship (De Wilde ., 2016;Hutter and Kriesi, 2019;Hutter et al et al., 2016), our measures of politicization do not increase linearly over time, but are context dependent, which supports the reliability of our measures and the validity of  our inferences.Third, we have tested the relationship between politicization and representation over four EP elections.Hence the results are not driven by debate speci cities during a given EP election. Does politicization matter for EU representation?
Our aim was to provide a rst, yet thorough, account of the relationship between EU  politicization and representation.We recommend the following steps to further investigate this relationship.First, our analysis distinguished between challenger and mainstream parties to account for potential strategic behaviour of political parties (for example De Vries and Hobolt, 2012).However, Eurosceptic (challenger) parties are considered the main driver of politicization of EU affairs (for example Hooghe and Marks, 2009).Furthermore, traditional ideologies along the left right spectrum may also explain varying levels of congruence across different types of political parties (for example Costello et al., 2012;Lefkofridi and Casado Asensio, 2013;McEvoy, 2012).We thus recommend  that, theoretically, future research takes ideological party positions into account to assess the extent to which these moderate the effects of politicization on representation.Such a distinction between different party types would also allow researchers to study the extent to which party responsiveness is affected by either growing or varying levels of EU politicization.Although research has started addressing questions pertaining to the effect of politicization on the relationship parties (for example Maier ., 2019; Meijers between et al and Rauh, 2016), we know little about the potential effect of EU politicization on party responsiveness to voters.Mainstream parties deserve particular attention since they tend to either ignore (for example Hooghe andMarks, 2009, 2018) or downplay politicization of EU affairs (for example Braun ., 2016;Grande and Kriesi, 2016).Here, research et al could also examine the extent to which EU politicization has an effect on voter perceptions of political parties (see Adams ., 2011).et al Second, we examined representation through the prism of party voter congruence.
-Our unit of analysis is at the party level and follows extant literature that focuses on how well representatives are performing in representing their voters (for example Mattila andRaunio, 2006, 2012).Future research could further disaggregate the empirically relationship between EU politicization and representation by studying data at the level of individual voters (for example Dalton, 2017;McEvoy, 2012).Such an analysis would also allow for a precise individual level test of clarity and affect, by for example examin- ing the extent to which information gains through individual exposure to politicization in uence voters receptiveness to party cues.It could also assess whether EU issue  ' salience at the individual level corresponds to EU issue salience at the party level.
Third, we studied congruence on the left right and pro anti EU dimensions.These re-- spectively align with questions pertaining to EU policy and polity, and constitute crucial dimensions in the study of politicization (De Wilde and Zürn 2,102, p. 140).However, congruence has also been studied with respect to distinct policy areas (for example Lefkofridi and Casado Asensio, 2013;Vasilopoulou and Gattermann, 2013).It therefore  remains to be seen whether EU politicization has a differentiated effect on representation that is conditional upon .Furthermore, EP elections are not the specific policy dimensions only arena of EU politicization (for example Hutter and Grande, 2014;Kriesi, 2016;Statham and Trenz, 2013).We thus recommend that researchers examine whether representation at the has been affected by increasing levels of politicization of national arena European issues, if at all (see Ares ., 2017, for a focus on spill over effects).Examples et al of such contestation include the recurring Eurozone crisis (for example Lobo and Karremans, 2018) as well as the impact of the global health crisis resulting from Covid 19.In sum, we hope that our article inspires scholarship and sets an extensive  research agenda of conceptual and empirical questions that future work can address.
The vote share k is based on rst preference votes in the case of Ireland.Since the placement scales slightly  differed in each EES we standardized the party positions for all election years to compare party positions over time.As with the dependent variables, we have two measures for WPSD, one that measures the extent to which the left right dimension is contested by political parties in the domestic context (Party polarization LR , = 0.23; = 0.05), M SD So a Vasilopoulou and Katjana Gattermann and a second that measures polarization on the pro anti EU dimension ( ) and calculate politicization by (a) multiplying et al salience with the sum of actor expansion and party polarization, and following Hutter and Kriesi (2019, p. 1005) who amended their own measure of politicization by (b) multiplying salience with party polarization only.Considering these two alternative measures provided by the extant literature allows us to address possible spurious effects derived from the operationalization of politicization.In other words, it adds validity to our ndings.As before, we have two composite measures for each politicization  variable: PoliticizationA LR ( = 0.06; = 0.04) and M ( = 0.02; = 0.01) for the second one.We estimate the respective M SD models separately.

Figure 1 :
Figure 1: Average marginal effects on levels of congruence on the left right dimension (Model 1) and the pro anti EU dimension (Model 2), = 328.-N Note: Dots represent average marginal effects and lines 95% confidence intervals.Positive effects denote decreasing levels of congruence.

Figure 2 :
Figure 2: Average marginal effects on levels of congruence on the left right dimension (Model 3) and the pro anti EU dimension (Model 4), = 328.-n Note: Dots represent average marginal effects and lines 95% confidence intervals.Positive effects denote decreasing levels of congruence.

Figure 3 :
Figure 3: Average marginal effects on levels of congruence on the left right dimension (Model 5) and the pro anti EU dimension (Model 6), = 341.-n Note: Dots represent average marginal effects and lines 95% confidence intervals.Positive effects denote decreasing levels of congruence.
for the EU 15 member states (Popa and Schmitt, 2015; One advantage of this approach is that since we rely on voters placement for party -' positions as we explain below we avoid missing values for those parties which are not included in other data sources, such as expert surveys or manifesto data.