Representation in the European Parliament: Factors affecting the attitude congruence of voters and candidates in the EP elections

Introduction

This article explores factors determining how well voters are represented in the European Parliament (EP). In their quest to explore the quality of representation in democratic elections, many scholars have employed the Responsible Party Model (for example, Thomassen, 1994), which sees political parties as major actors by and through which representation is realized (Sartori, 1968: 471). Studies based on this model have established that the preconditions for effective representation are largely fulfilled at the European level. Voters in European elections have a choice between a considerable number of parties that are ideologically as distinctive and cohesive as national parties (Schmitt and Thomassen, 1999). Voters are able to recognize ideological differences between parties and consider them at the ballot box in European elections (Van der Brug and Van der Eijk, 1999). In addition, the party space in the EP reflects the national political space, with left/right as the major dimension of competition (for example, Schmitt and Thomassen, 2000). When it comes to the left/right dimension, voters, at least those from Western Europe, are well represented in the European Parliament. However, little is known about the quality of representation in the EP beyond left/right, although there are some indications that representation on issues is much poorer than on left/right (Thomassen and Schmitt, 1997, 1999). What has also remained unexplored so far is whether the quality of representation in the EP varies for different groups of citizens, parties and countries. These lacunae in our knowledge of representation at the European level led us to enquire which individual-, party- and system-level characteristics influence the quality of the democratic representation of voters on political issues.

This article is based on the delegate model of representation to analyse the quality of the link between voters and their representatives. In the delegate model, representatives are expected to act as the voice of those who are not present in the representative body (for example, Pitkin, 1967: 133). This link between voters and representatives will be effective if political decision makers enter the policy process with the same issue preferences as those who voted for them (Dalton, 2002: 217). If the attitudes of representatives resemble those of voters, their actions and decisions can be expected to be in line with the political preferences of the citizens they represent (see, for example, Holmberg, 1999; Thomassen, 1994). In order to examine the mass–elite linkage, this article focuses on the congruence of issue attitudes between voters and their representatives in the European Parliament. It focuses on attitude congruence on basic packages of policy opinion in the socioeconomic and libertarian/authoritarian issue domains as well as on issues of immigration and EU integration. In doing so, it contributes to an extensive literature on substantive representation, which investigates whether the issue positions of governments, political parties and actors reflect the attitudes of voters (see, for example, Huber and Powell, 1994). This literature takes a well-established place in research on representation, next to studies that focus on symbolic and descriptive representation (see Pitkin, 1967), which examine whether the presence of representatives with certain ascriptive characteristics, such as race or gender, affects the attitudes of groups with the same characteristics and the extent to which these representatives represent such groups better than other representatives (see, for example, Childs, 2008).

The major aim of this article is to assess which voters are best represented in the European Parliament and which party- and system-level characteristics account for the congruence in positions of voters and parties. So far, our knowledge about determinants of representation in European elections is very limited. The few studies that focus on representation in elections teach us that parties that project clear cues represent their voters better than parties with less clear cues. We also know that representation is better in countries with proportional representation and in countries with a large number of parties (see, for example, Dalton, 1985; Thomassen and Schmitt, 1997). However, little is known about the extent to which individual-level characteristics affect representation. Moreover, conclusions reached so far about representation in European elections are limited in scope. They focus on representation in terms of left/right or on issues related to the EU. There are yet other domains that have not been studied. Our analysis of voter–elite congruence in several issue domains shows that citizens with higher levels of education and political knowledge, as well as those belonging to higher social classes, have their attitudes generally better represented in all issue domains. This article also shows that parties with a clear ideological profile represent their voters better than do other parties, and radical right parties represent their voters better on issues of immigration and EU integration. Further, we offer tentative evidence for better representation in open and ordered ballot systems.

Representation in the European Parliament

The linkage between citizens and political decision makers is one of the most important topics in the study of democratic political systems (for example, Shapiro et al., 2010). Although there are various links in the political chain of democratic command and control, the congruence between the issue attitudes of voters and political elites has been a key topic in studies in representation. In this perspective it is assumed that policy outcomes are likely to reflect citizen preferences if representatives share the opinions of their voters (for example, Miller and Stokes, 1963). The concept of congruence between the issue attitudes of political elites and citizens has been acknowledged as important for analysing the quality of representation (for example, Powell, 2000). In European parliamentary systems, political parties are central to the process of representation (for example, Thomassen, 1994). Therefore, studies of representation in Europe have mostly focused on the voter–party dyad. We follow this approach and examine the congruence between the attitudes of voters and political elites from the party they voted for in the last EP elections. Certainly, policy outcomes may be suboptimal from the perspective of voters even if representatives share issue attitudes with their electorate, because many factors play a role in the process of decision-making. This article will not consider the complex picture of what determines representative policy outcomes, but will simply focus on the congruence between the issue attitudes of voters and political elites as an indicator of representational quality.

We will consider congruence on basic packages of political attitudes that have been important for party competition in all parliamentary systems in both Western and East Central Europe (see, for example, Kriesi et al., 2008). ¹ We distinguish between issues in the socioeconomic and libertarian/authoritarian domains as well as issues related to immigration and EU integration. The socioeconomic domain pertains to topics such as state ownership, the role of private enterprise and state intervention in the economy, and the redistribution of income and wealth, while the libertarian/authoritarian domain includes attitudes towards societal and personal freedoms, such as equal treatment of women and homosexuals. In addition, we conduct separate analyses for attitudes towards immigration and European integration because these issues have gained importance in political discourse in recent years (see, for example, Kriesi et al., 2008).

The congruence of attitudes between voters and representatives may vary across groups of issues. Scholarly literature shows that the attitudes of voters and political elites are structured by at least two dimensions (Thomassen and Schmitt, 1999; Kriesi et al., 2008). Because there is no consensus on whether attitudes towards immigration and European integration are strongly related to these dimensions (Kriesi et al., 2008; Van der Brug and Van Spanje, 2009), we have decided to treat issues of immigration and EU integration separately from the socioeconomic and libertarian/authoritarian domains. Because several dimensions structure public opinion, voters may not be equally well represented in all issue domains by the party of their choice. Our cross-country analysis focuses on factors that generally account for better representation in all issue domains, and also includes variables (mostly at the level of parties) that may account for better representation in some issue domains but not in others. In the following, we examine which individual-, party- and system-level characteristics affect the congruence of attitudes between voters and their party on socioeconomic, libertarian/authoritarian issues, immigration and EU integration.

Operationalization

This study employs the 2009 European Election Study (EES, 2009a; Van Egmond et al., 2010) and the European Election Candidate Survey (EES, 2009b; Giebler et al., 2010) because the two surveys contain identical questions, which tap respondents’ attitudes on a broad range of basic policy issues. The voter study was fielded immediately after the EP elections in June 2009, with independently drawn samples of over 1000 respondents in each of the EU’s 27 member states. The candidate survey was carried out shortly before and after the EP elections and contains information on representatives from 162 parties, which is over 60 percent of the parties that were contacted in the course of the survey. In order to examine the attitude congruence of voters and their representatives, we combine data from the voter and candidate survey, which yields information on 28 political systems, ² 162 parties and over 11,500 respondents in the European Union. ³ We create a multilevel data structure, where voters are nested in parties according to their party choice in the last EP elections, and parties are nested in 28 political systems.

Both surveys asked respondents to indicate on a scale from 1 to 5 (Likert scale) how much they disagree or agree with a number of statements. These statements tap into attitudes on a number of socioeconomic and libertarian/authoritarian issues, as well as immigration and EU integration. Items in the socioeconomic domain are: ‘Private enterprise is the best way to solve your country’s economic problems’ (we call this item Enterprise), ‘Major public services and industries ought to be in state ownership’ (Ownership), ‘Politics should abstain from intervening in the economy’ (Intervention), ‘Income and wealth should be redistributed towards ordinary people’ (Redistribution). Issue items in the libertarian/authoritarian domain are: ‘Same-sex marriages should be prohibited by law’ (Same-sex marriage), ‘Women should be free to decide on matters of abortion’ (Abortion), ‘People who break the law should be given much harsher sentences than they are these days’ (Law & order), ‘Schools must teach children to obey authority’ (Authority), ‘EU treaty changes should be decided by referendum’ (Referendum), ‘A woman should be prepared to cut down on her paid work for the sake of her family’ (Family). Attitudes towards immigration are tapped by two questions: ‘Immigrants should be required to adapt to the customs of [country of respondent]’ (Adaptation of immigrants) and ‘Immigration to [country of respondent] should be decreased significantly’ (Decrease of immigration). Attitudes towards EU integration are measured with the question of whether EU unification should be pushed further or whether it has already gone too far (EU Integration).

We measured the congruence between attitudes of voters and the party they voted for in three steps. We first determined the position of each party on each issue by computing the mean position of all EP candidates from each of the parties. So the position of party A on an issue is the mean of the positions of all candidates of party A. In the second step we computed for each respondent for each issue the distance between his/her own position and the position of the party. In the third step, we established congruence in each issue domain by computing for each respondent the average distance between them and the party they voted for across all issues belonging to this domain. In the final step, we took the negative value of this score, where high values indicate that the voter is (on average) well represented in a particular domain, whereas low values (that is, large distances) show that the respondent is not well represented. ⁴

Our individual-level explanatory variables are Education, Social class, Political knowledge and Exposure to mobilization efforts. For Education we employ the cross-country measure included in the EES 2009a, which is based on ISCED coding. This measure provides information on the highest level of education completed by respondents and consists of six categories coded as follows: 0 stands for pre-primary education, 1 for primary education or first stage of basic education, 2 for lower secondary or second stage of basic education, 3 for upper secondary education, 4 for post-secondary non-tertiary education, 5 for first-stage of tertiary education and 6 for second-stage of tertiary education. Although the ISCED measure does not capture distinct values of vocational education (Schneider, 2010), it is the best measure in the EES 2009 comparing educational attainment across countries.

Social class is operationalized by occupational categories, which is in line with much recent work in this area. The EES contains a categorization in 11 occupational groups, which match up very closely to categorizations employed in recent work on class voting (for example, Kriesi et al., 2008; Evans and De Graaf, 2012). This is the categorization used in our study.

Political knowledge is measured by the factual correctness of respondents’ answers to seven statements about national and European politics contained in the EES 2009. We coded the correct answers for each question as 1, 0 otherwise. In order to assess whether these questions relate to a single latent dimension, the responses to these items were analysed on a pooled sample of all countries by means of an ordinal IRT (item response theory) model (Mokken, 1971). Constructing a scale on the basis of all countries allows us to adequately construct an index that measures the same concept across countries. We use an ordinal IRT model because it accounts for various levels of difficulty across items and thus constitutes a more valid dimensionality test than other data reduction methods (Van Schuur, 2003). We were able to construct a satisfactory political knowledge scale (with Loevinger H coefficient of 0.33) on the basis of three items: whether Switzerland is a member of the EU, whether the EU has 25 member states and whether every country in the EU elects the same number of representatives to the European Parliament. The additive index for Political knowledge ranges from 0 to 3, where 0 indicates the lowest and 3 the highest level of political knowledge.

In order to construct Exposure to mobilization efforts we performed Mokken scaling for all countries together on questions about how often voters (1) watch a programme about the election on television, (2) read about the election in a newspaper, (3) talk to friends or family about the election, (4) attend a public meeting or rally about the election and (5) visit a website concerned with the election. We have recoded each item so that 1 indicates ‘never’, 2 ‘sometimes’ and 3 ‘always’. A common scale can be constructed for the first four items (with Loevinger H of 0.35). The index we constructed here is the average of responses on all items and ranges from 0 to 3.

At the party level we have four explanatory variables. Ideological clarity is a party characteristic that denotes the level of agreement among citizens about where a party stands in ideological (left/right) terms. To measure perceptual agreement among voters, we used a question where respondents were asked to place each political party in their party system on the left/right scale ranging from 0 to 10. Following Van der Eijk (2001), we calculated a coefficient of agreement that describes the distribution of these perceptions. The value of this coefficient runs from −1 (maximum disagreement) to +1 (maximum agreement). Radical right parties and green and radical left parties are distinguished by means of dummy variables, constructed on the basis of the classification of the Comparative Manifesto Project (Volkens et al., 2010).

At the country level, we create two dummy variables that distinguish between closed, ordered and open ballot systems. District magnitude is a continuous variable with the number of districts used for EP elections. We adapted the classification of district magnitude and ballot systems from Farrell and Scully (2007). In order to test the hypothesis on the effect of district magnitude in open ballot systems, we created an interaction between District magnitude and the dummy for Open ballot system. Candidate deviation from party position has been operationalized as the standard deviation from the mean party position on an issue (domain), where the mean party position is given by the mean positioning of all candidates belonging to a party. Finally, EU membership in years stands for the number of years since a country joined the EU.

We employ multilevel analysis because our model has a hierarchical structure: individuals are nested in parties according to their choice in the last EP elections, and parties are nested in political systems. In order to facilitate interpretation of the models, we have centred explanatory variables which do not contain the baseline of 0 (Hox, 2010: 59–63). ⁵

Findings

Before we conduct the causal analysis, we will present some descriptive statistics for four outcome variables: congruence in the socioeconomic domain and the libertarian/authoritarian domain, and on the issues of immigration and EU integration (Table 1 in the Web Appendix). The outcome variables range from −5 (the maximum distance between a voter and her/his party on all issues in a domain) and 0 (perfect congruence between party and voter on all issues in the domain). Table 1 reveals that on average the differences in stances of voters and EP representatives are smallest for the socioeconomic domain (mean of −1.14 and minimum of −3.33). In turn, we observe the highest average difference on EU integration (mean of −1.29).

Table 2 in the Web Appendix presents results from multilevel analyses for four outcome variables: congruence in the socioeconomic domain (Model I), the libertarian/authoritarian domain (Model II), immigration (Model III) and EU integration (Model IV). ⁶ These model estimates were obtained in analyses that included the following control variables: place of residence, gender and religion. ⁷ Since we have no theoretical expectations about the effects of these individual-level control variables, we decided not to present these effects in this journal article. The full results are available in the Web Appendix in Table 2 Full Results. In each model we present two versions – Model A and Model B – which include different interactions. We tested these interactions with system-level variables in separate models, because the limited number of observations at this level (N  =  28) does not give the tests enough power when testing them in one joint model. ⁸ In all models, a positive coefficient for an explanatory variable indicates that this factor contributes positively to representation.

With regard to individual-level explanatory variables, we see that representation in the European Parliament improves with levels of education. In each issue domain under study – socioeconomic, libertarian/authoritarian, immigration and EU integration – we see a positive coefficient for education in each of the Models A and B. ⁹ This confirms Hypothesis 1. Table 2 in the Web Appendix reveals that social class matters very little for representation in the socioeconomic and libertarian/authoritarian domains. Yet, on the issues of European integration and especially immigration, workers are less well represented than other groups, even when controlling for education. Since attitude congruence for middle-class voters is higher in two domains than for the working-class, Hypothesis 2 is partially confirmed.

The level of political knowledge matters for attitude congruence as well. Table 2 in the Web Appendix displays a positive effect of knowledge on congruence in the socioeconomic and libertarian/authoritarian domains and for the issue of immigration. This pattern cannot be confirmed for the issue of EU integration, however. With the exception of EU attitudes, our expectation with regard to the effect of political information on the quality of representation (Hypothesis 3) can be confirmed. Surprisingly, voters’ exposure to mobilization efforts does not play a role in attitude congruence. The only significant effects that appear here – those in the libertarian/authoritarian domain – are negative, while effects for other issue domains are positive and non-significant. Here, we have to reject Hypothesis 4.

Among party-level variables, ideological clarity plays a role in congruence of issue attitudes. A clearer ideological stance of a party in left/right terms leads to more congruence with its voters in the socioeconomic and libertarian/authoritarian domains. This is precisely what we expected in our Hypothesis 5. Furthermore, radical right parties represent their voters better on immigration and EU integration than do other parties, which confirms Hypothesis 6a. However, there is no evidence that green and radical left parties represent their voters better on socioeconomic issues. The effect for radical left parties is negative in several cases. The significant effect for green parties in Model I A is not enough to draw any conclusions on this matter. Hypothesis 6b should thus be rejected.

With regard to system-level variables, there is some evidence for more congruence in open and ordered ballot systems than in closed ballot systems in the libertarian/authoritarian domain, where three of the four effects are positive and significant. In all other cases the effects are insignificant, and they are sometimes positive and in other cases negative. These results thus provide only limited evidence for Hypothesis 7a. ¹⁰ Because the effect of an open ballot system may be conditional on district magnitude, we have introduced an interaction between a dummy for open ballot system and a measure of district magnitude (Model A presents this interaction for every issue domain). The obtained coefficients are positive, suggesting that the higher the district magnitude in an open ballot system, the higher the opinion congruence between voters and EP candidates from the party of their choice. However, these effects do not reach conventional statistical significance. This points us to our Hypothesis 8a, but because of the low number of cases we have to be cautious in our conclusions. We remain similarly cautious about how candidates’ deviations from their parties’ positions affect representation in an open ballot system. In three domains the effects are negative and insignificant, but in the case of European integration the effects are positive and significant. Last, but not least, how many years countries have been members of the EU does not play a role in the quality of representation of their citizens in European elections.

The last observations concern the strength of the effects and the variance explained at each level of analysis. The effects of the individual-level variables are not very strong. When comparing the predicted values on the dependent variable between the minimum and maximum scores on each of the independent variables, the difference never exceeds 0.20 on scales with a range of about 3.50 points (see also Table 2 in the Web Appendix). For instance, education ranges between 0 and 6, and the size of the parameter is .03 in the case of immigration (Model III). So, the predicted difference in attitude congruence between the highest- and the lowest-educated citizen is 0.18. This difference is meaningful but rather limited. On EU integration, Model IV B tells us that the predicted difference between professionals (parameter is .03) and semi-skilled workers (parameter is −.13) is .16 points. Other individual-level effects are weaker.

Judging by the size of the parameters, the effects of party- and system-level variables are somewhat larger than those of individual-level variables. Model III B predicts that radical right parties represent their voters much better on the issue of immigration than other parties. The difference is .48, which is a substantial difference on a scale with a range of 4.50 points. Also, the effect of the ideological clarity of parties is substantial. When agreement about the left/right position of a party increases from .25 (low agreement) to .75 (high agreement), the representation of their voters in the socioeconomic and libertarian domains will increase by .12 to .17. The predicted effect of an ordered ballot system versus a closed ballot system is even slightly larger (0.20), but only in the libertarian/authoritarian domain.

Moving to the explained variances, we first of all see that our models are least able to explain individual-level variance. The explained variance ranges from 2 percent to 10 percent depending on the issue domain. Party-level variance can be much better explained. The R²’s range from 1 percent to 60 percent. Variance at the system-level can be explained most accurately, with explained variance ranging from 18 percent to 90 percent. Although there are substantial differences in the variance explained across issue domains, there is a general pattern showing that the system-level factors explain most of the variance, whereas individual-level factors do not explain much. So, the extent to which individuals are represented does not depend very much upon their social class, education or political knowledge.

Conclusion

This article has analysed which individual-, party- and system-level characteristics account for the quality of representation in European elections.

The first main finding of our investigation is that citizens with higher levels of education and more political knowledge, and who belong to higher social classes have their views generally better represented than the lesser educated, those with less knowledge and those belonging to the working-class. These results are not new or surprising, but they do show that, like national parliaments, the EP suffers from a bias towards representing the more-educated middle-class citizens (for example, Schattschneider, 1975). The reasons behind these inequalities in representation can be multiple. Better representation of more-educated middle-class voters can be a natural consequence of the fact that EP candidates are likely to be highly-educated, and themselves belong to the middle-or upper-classes. There may also be strategic reasons involved: EP candidates may ignore less-educated, lower-class voters because these voters are less likely to turn out in EP elections. Finally, the cognitive skills of voters may play a role. More-educated and more-knowledgeable voters may be more aware of the parties’ policies, and hence be in a better position to choose those parties that they agree with the most.

The inequalities in representation of different groups of voters have implications for the democratic character of the EU. With the recent extension of powers to the EP, which took place through the Treaty of Lisbon, an attempt has been made to alleviate the democratic deficit in the EU. The findings of this article highlight the fact that, even if some problems of a lack of democratic accountability are addressed, the European Parliament still suffers from inequalities in representation of various groups of citizens. Inequalities in representation imply that not all EU citizens are able to influence the direction of the policy agenda to the same extent, which is likely to have repercussions for the direction of policies formulated at the EU level. At the same time, we should also emphasize that variations at the individual level cannot be explained very well with these socio-structural variables. This means that the differences within these groups are much larger than the differences between the groups.

The second main finding of this article is that important differences exist between the different domains in how well groups of voters are represented. In the socioeconomic and libertarian/authoritarian domains, better-educated and more knowledgeable voters are better represented, but, when controlling for these variables, social class does not have an additional effect. Yet in the issue areas of immigration and European integration, the working-class is significantly less well represented than other groups. A possible reason could be that working-class citizens tend to vote for left-wing parties, which they agree with on socioeconomic policies but disagree with on the issues of migration and European integration. Van der Brug and Van Spanje (2009) pointed to the representation gap at the national level, which stems from the fact that the ideological profile of the ‘authoritarian worker’ (Lipset, 1960) is not represented in the national party space. Our findings are in line with their idea.

Thirdly, parties with clear ideological profiles present their voters best in the socioeconomic and the libertarian/authoritarian domains. Controlling for ideological clarity, radical left-wing parties represent their voters less well in these domains. Thus, the traditional established parties with clear ideological profiles (mainly social democratic, Christian democratic, conservative and liberal parties) represent their voters best in these domains. However, radical right parties represent their voters on average much better in the fields of immigration and European integration. This speaks to recent literature, which has demonstrated that the issue of further European integration has increasingly become subject to political competition among political parties and increasingly affects voters’ electoral choices (for example, De Vries, 2007). Our findings show that, even if voters more frequently rely on their attitudes towards EU integration in casting a ballot, the extent to which the party of their choice will represent them on this issue will vary. So, even if considerations regarding EU integration have begun to guide voters in their party choice, putting an end to the period of ‘permissive consensus’, significant differences can be observed with regard to how voters are represented on this issue and, ultimately, in the extent to which their vote can influence the direction of integration.

The last finding of this article suggests that in open and ordered ballot systems there is more attitude congruence between voters and their party in the libertarian/authoritarian domain. Several scholars have hypothesized that representation would work best in open ballot systems, especially when there are large districts (for example, Farrell and Scully, 2007). Hobolt and Høyland (2011) also expected the quality of candidates to be positively affected by these characteristics of electoral systems, but did not find significant effects. The results from our analyses are mixed, but provide some tentative evidence that attitude congruence increases with district magnitude. In such systems, a candidate’s deviation from his/her party’s position positively affects representation, but only in the field of European integration, where the largest representation gap was noted. These findings extend existing knowledge about ballot systems, since their recent introduction for the EP elections. Our findings offer some tentative evidence for a better voter–candidate congruence of opinions in more open ballot systems. Dependent on their further investigation, these findings have potential policy implications for bringing voter preferences closer to representatives at the European level.