Estimating parties' left-right positions: Determinants of voters' perceptions' proximity to party ideology Academic research paper on "Political Science"

Contents lists available at ScienceDirect

Electoral Studies

journal homepage: www.elsevier.com/locate/electstud

Estimating parties' left-right positions: Determinants of voters' perceptions' proximity to party ideology

Kathrin Barbara Busch

GESIS — Leibniz-Institute for the Social Sciences, Unter Sachsenhausen 6-8, 50667, Köln, Germany

CrossMark

ARTICLE INFO

Article history:

Received 27 August 2014

Received in revised form

4 January 2016

Accepted 5 January 2016

Available online 6 January 2016

Keywords: Ideology Issue saliency Party manifesto Party perception Position shift

ABSTRACT

The article aims to explore whether the accuracy of voters' perceptions of party ideology are affected by party position shifts and by the media's turn to non-left-right issues, such as political leadership, during election campaigns. Using data from the Comparative Study of Electoral Systems (CSES) and a country-specific left-right index based on data by the Comparative Manifestos Project (CMP), multilevel analyses reveal that emphasizing leadership issues can lead voters to reflect left-right positions more accurately. A party's left-right position shift between elections does not lead to a significant difference in voters' perceptions, while a shift on the economic sub-dimension of left-right ideology can even lead voters to more position clarity. However, multiple parties' shifting their positions seems to overburden respondents' cognitive capacity.

© 2016 The Author. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND

license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction

Theories of electoral decision making suggest that citizens should vote for the parties with which they identify, which have policies that represent the citizens' own interests best, or which are likely to change policies in their preferred direction (for example Merrill and Grofman, 1999). This decision requires voters to compare their own political stances with those of the parties. Voters must either acquire the relevant information about important political issues or at least have a sense of the parties' ideologies, which can serve as a 'cue to vote choice' (Downs, 1957). Ideological locations must then function as a type of umbrella for a broad range of single issues, which the voter does not necessarily need to study separately.

In election campaigns, party manifestos are the most important documents to determine the parties' ideological positions. According to interviews with party elites, politicians strive to base all of their campaign communications on these party manifestos (Adams et al., 2011, 372). Hence, voters can inform themselves about policy stances directly by reading manifestos or by simply following the election campaign. If the campaign is successful for a party, it communicates the most important messages from the manifestos to the voter and thereby transports party ideology. If a

E-mail address: Kathrin.Busch@gesis.org.

party undergoes an ideological change from one election to the next, a well-informed public should realize this fact. However, recent research has suggested that the mean voter does not perceive changes in party ideologies through party manifestos (Adams et al., 2011, 2014). It is not yet clear what determines how well the individual voter perceives these changes.

The aim of this research is to analyse the determinants that foster or undermine the links between current party locations and voters' perception of these positions. Specifically, it tests determinants related to changes in party positions over time and the saliency of political issues prior to elections to determine how strongly voters' ideological pictures of parties are influenced by current developments in the electoral campaign. Via multilevel analysis, the influence of system and party level characteristics are weighted against individual voter level influence on perceptual accuracy. The basic theoretical argument that inspired this research is that voters' knowledge of parties' current ideological locations is important for their voting decisions. Shifts in party manifesto positions or the emergence of new parties can affect the left-right locations of parties and even the organization of the important parties within the party system. Voters thus must pay some attention to changes to have a clear view of party ideology. If important messages are not communicated to voters, they might base their pictures of parties' ideologies on outdated information and perhaps vote for parties that do not represent them best (anymore).

http://dx.doi.org/10.1016/j.electstud.2016.01.003

0261-3794/© 2016 The Author. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

The research question can be located within the wider body of research on 'political knowledge' or 'political sophistication' (Gordon and Segura, 1997; Granberg and Holmberg, 1988, Neuman, 1981, 1986), for which knowledge of party positions is a central aspect of citizens' education. The innovative aspect of this investigation is the linking of voter estimations to current manifesto information, as well as the question of how strongly party position changes and issue saliency influence voters' perceptions. Apart from individual level determinants, it thus also tests for the influence of party behaviour and the structure of the party and electoral system to communicate manifesto information. It is the first study to use left-right values from the '12-step index' of manifesto data (Franzmann and Kaiser, 2006) as the basis for the calculation accuracy of individual level perception.1

Only if the salient issues within party manifestos form broadly consistent ideological pictures do voters have the opportunity to estimate precisely the parties' locations in this one-dimensional concept of political space, and only then can they choose the party that best reflects their own ideological standpoints. Apart from putting Downs to the test, the research question has several implications.

Firstly, a vast amount of empirical research has connected the ideological congruence of voters' preferences with their political representatives (Blais and Bodet, 2006; Budge and McDonald, 2007; Golder and Stramski, 2010; Huber and Powell, 1994; Klingemann et al., 2006; McDonald et al., 2004; Powell and Vanberg, 2000; Warwick, 2010). This research assumes that voters are able to integrate information about policy into a consistent view of party ideology that is broadly shared by voters and parties.2 If voters did not understand left-right positions, a comparison between parties and voters' ideological stands would be a mere comparison of labels and not of ideological content.

Secondly, some researchers have argued that the policy-based party-voter linkage is normatively preferable to other political ties (Kitschelt, 1995). In an ideal world, well-informed voters could compare their own self-interests with those of qualified political candidates (Gastil, 2000: 32.) or parties. If voters cannot arrive at consistent pictures of party ideologies, they might depend more heavily on other cues to make choices regarding voting, such as party leaders' charisma or clientelist ties. Such choices could be problematic because elections won purely on the basis of charisma or clientelist ties 'invoke the centrality of the electoral process of interest aggregation for the distribution of group and societal

1 Adams et al. (2011, 2014) used the RILE index, which is included in the CMP data, and they did not measure perceptions of party position changes on the individual level, but they compared respondents' perceptions in one year to that of another year. Research that has not examined individual political knowledge has used measurement of voters' party-specific mean estimates (Dahlberg, 2013; Gordon and Segura, 1997; Granberg and Holmberg, 1988), to which single voter's estimates have been compared. However, only the CMP data, to my knowledge, contain information about current and previous party positions. Thus, they are the best starting point to explore how well knowledge about current manifesto positions is conveyed to voters, as well as whether voters notice ideological shifts over time.

2 It should be noted that the understanding of left and right can also vary between countries, but this source of bias will be addressed by using country-specific indices of the CMP data. This process will be explained in greater detail below.

3 Electoral research has also found that some democratic elections have been won mostly due to valence issues, such as which party will manage 'the economy' best or which party 'performs best' (Whiteley et al., 2005; Clarke et al., 2011). However, even so-called valence issues can have relationships with political positions. For example, a respondent might believe that a specific party would handle an economic crisis best, which could be categorized as a 'valence' issue, but this 'best handling' might also be cognitively related to the respondent's knowledge of how the specific party would manage the crisis, i.e., through specific economic policies.

benefits' (Kitschelt, 1995: 450).

However, considering leadership evaluations in elections is not necessarily problematic if it occurs in addition to issue consider-ation.3 My analysis suggests that it can even draw attention to left-right positions. In general, voters' mean estimations of parties on the left-right scale correspond well with the left-right positions of the parties. On the individual level, the precision of a voter's party estimation is best if the voter's own left-right position is close to that of the party but is not flawed by extremely positive affection. The strength of a single party's change on the left-right axis does not significantly lead voters to estimate a party less precisely. If only changes in economic political issues - the economic subdimension of the left-right scale - are analysed, a party's position shift even has a slightly increasing effect on proximity of voters' estimations to the party's manifesto positions. Contrarily, the stronger that multiple parties change their positions on the left-right scale, the less accurately do voters estimate the parties' positions. It is hence the position shifts of the many that clearly decrease position clarity.

In the next section, I will elaborate further on the links to previous research on political knowledge and the precision of voters' party estimations. I will then advance my argument that party positioning behaviour and party system properties, such as the number of parties, account for much of the difference between voters' cognitions about parties' ideological locations. Using data from the Comparative Study of Electoral Systems (CSES, Module 1 and 2, Full Releases) and the Comparative Manifesto Project (CMP) (Klingemann et al., 2006), the multilevel analyses with hierarchical linear models (HLM) in Section 6 will evaluate and compare the determinants of estimation proximity on the individual, party and system levels. This analysis includes 150,642 respondents' party projections in 25 post-election studies of parliamentary elections, located in a variety of older Western and younger Eastern European democracies.4 In the final section, I discuss the findings and their implications for future research.

2. Research on political knowledge and voter perceptions

Knowledge about party ideology is a type of political sophisti-cation5 with a very practical value (Neuman, 1981; Gordon and Segura, 1997). It is a prerequisite for issue voting because only if voters have knowledge of party positions,6 they are able to vote for the parties that represent their own interests best. To test for this knowledge, respondents in election surveys are often asked to estimate parties' positions, most commonly along the political left-right dimension. Researchers into political knowledge draw on these data. They operationalize the precision of the individual voter's estimation by subtracting the voter's estimate of a party's ideological location from the mean of all voters' estimates of the same party's location (Gordon and Segura, 1997; Granberg and Holmberg, 1988; Neuman, 1981,1986). Because not even party experts can know precisely what the 'true' ideological party position is, this approach is legitimate because uses the 'perceptually agreed position' (Dahlberg, 2013) as a valid point of reference.7 Other research, which is also linked to the present study, has analysed whether the mean of the voters indicates perception of ideological

4 The full list of countries and parties in the sample can be found in Appendix .

5 In the following, the terms 'political sophistication' and 'political knowledge' denote the same concept, and I will use them interchangeably.

6 This knowledge is meant generally, both regarding the acknowledgement of single party positions that are important to voters and regarding ideological vote choices as a specific type of 'issue voting', in contrast with a mere party ID vote.

7 A problem with expert data is also that they are usually limited to one point in time and hence do not capture changes in party positions between elections.

change or whether party supporters react to party position changes by adopting positions or changing their party support (Adams et al., 2014; Adams et al., 2012) While this study has a slightly different focus - it seeks determinants that increase or decrease the link between voters' estimations of party ideology and the parties' current self-expressed ideology — it can nevertheless draw on some of these findings from the research on political sophistication and voter reactions to party changes.

2.1. Individual ability, psychological and system effects

Neuman's (1981) ground-breaking study of political knowledge identified two intertwined dimensions of cognitive organization: 'conceptual differentiation' (the 'ability to identify and discriminate among the various political issues, actors, and events') and 'conceptual integration' (the 'explicit organization of ideas and information in terms of abstract or ideological constructs') (Neuman, 1981: 1237). Analysing these cognitive abilities only in relation to socio-demographic determinants, he found that education is most strongly linked to conceptual integration (Neuman, 1981, 1986). Later quantitative studies concentrated specifically on the psychological determinants of individual voters' political knowledge (Granberg and Holmberg, 1988; Merrill et al., 2001; Drummond, 2010). Granberg and Holmberg (1988) found, in their analyses of voters in Sweden and the USA that emotional predispositions towards parties could lead to strong misperceptions. If voters like a party, they tend to exaggerate the party's similarity to their own positions; if they dislike it, they exaggerate the differences and place the party further away from their own political standpoints. These 'assimilation' and 'contrast' effects are found across a variety of political systems (Merrill et al., 2001). However, while contrast effects are stronger in majoritarian systems, assimilation effects are similarly strong under both majoritarian and proportional electoral rules (Drummond, 2010).

Gordon and Segura (1997) showed that system-level impacts influence voters' estimation abilities more strongly than individual voter characteristics. Their explanation was that electoral systems offer more or fewer incentives for voters to perceive information about parties. For example, voters in single-party contexts perceive less political information than voters in multiparty contexts (Gordon and Segura, 1997). The assumption is that, in multiparty systems, more parties have realistic chances of achieving leadership in government. A greater number of parties makes the vote decision more complex and provides an incentive for voters to inform themselves more thoroughly. It should also lead the media to distribute information about a greater number of parties. Further, because the presence of more relevant parties in multiparty systems leads to a potentially greater number of confrontational issues (Sartori, 1976), information about parties' positions should be more detailed as well. In two-party systems, in contrast, voters only need to know whether they roughly incline towards party A or party B to make rational vote decisions. They do not need to consider as many parties as voters in multiparty systems, nor must they know as many position details.

Dahlberg (2013) employed a three-level analysis to determine the level of estimators that is most important for voters' 'estimation precision'. Examining the perceptions of multiple parties within each party system, he found that, on average, party characteristics are more important determinants of voters' party projections than system determinants. In his model, the strongest effect is the result of the divergence of a party's position, compared to all of the positions of other parties in the same country.

The accumulation of previous findings shows that multiple determinants can influence voters' knowledge about party positions. This study follows Gordon and Segura (1997) and Dahlberg (2013)

by emphasizing system-level and party characteristics, but it also controls for the known individual level effects on the political perception. The main contribution of this paper is to test whether party position changes, individual or aggregated over the party system, confuse voters, and if voters' estimation proximity to the current party ideology is lower if the saliency before the election is not related to left-right issues.

3. Putting the spotlight on party behaviour and party system structure

This study builds on Gordon and Segura's instrumental-rational view of voters' knowledge and on Sartori's (1976) insights into party competition structures. Parties are the crucial actors because, if they do not manage to communicate their most important information or if they stay deliberately ambiguous during election campaigns, voters cannot make correct issue-based voting decisions. As Dahlberg (2013) found, it can be advantageous for voters' left-right perceptions if parties take distinctive positions: the more divergent that the party's position is to all other parties',8 the higher the voters' perceptual agreement is to a party's left-right position. Despite this potential advantage of parties to be acknowledged for their concerns with specific issues, taking a very divergent position does not generally conform to parties' strategies to win votes. Rather the opposite is the case: usually, parties profit electorally from 'broad appeal' strategies (Somer-Topcu, 2014). Parties then address a broad range of topics, and they sometimes even show inter-party or intra-party personnel differences regarding the same issues. As a result, issues appeal to a broader range of the electorate, and voters' left-right estimations of these parties can also vary more. In the same vein, Dahlberg (2013) argued that it is easier for voters to locate electorally weaker parties than very strong parties: the broader appeal of the latter leads to less clear issue domination than smaller parties usually exhibit.9

Apart from the parties' internal consistency regarding left-right positions, the consistency over time of the political issues that the parties present to the electorate should also be important for voters' perceptions. If voters generally do not perceive parties' ideology changes in party manifestos (Adams et al., 2011, 2014), these changes will lead to misperceptions. Voters' perceptions can be the results of their cognitive capacities, but they can also be dependent on the media's attention to the parties and the saliency of the issues which undergo changes. Further, parties can be more or less effective in the issue domination of the election campaign.10 In the following, hypotheses are mainly drawn regarding party position consistency over time and regarding campaign effects. I go on to explain further which party system characteristics should be important, and then I discuss the specific situation of the younger, post-communist Eastern European democracies. To determine whether voters' abilities, parties' behaviour or party system characteristics determine voters' perceptions of current programmatic information, each of these channels of influence is tested within multilevel linear models.

8 It should be noted here that the divergence measurement that Dahlberg employed is the mean sum of its distances to all other parties weighted by their respective electoral sizes. The distance to an electorally successful party is thus weighted more strongly than the distance to an electorally less successful party.

9 For Dahlberg, this is a confirmation of Kirchheimer's (1966) 'catch-all' hypothesis (Dahlberg, 2013).

10 Researchers into political communication see the media as a powerful actor in 'priming', i.e., the weighting of specific political issues (see Gidengil et al., 2002: 76—77), so parties might only have an indirect influence on this process.

3.1. Position consistency over time

Research into party manifesto data has suggested that parties have quite long-lasting ideological positions (Budge and Klingemann, 2001; Adams, 2001). While they constantly make small movements, major changes are quite rare, at least in the long-established systems of Western Europe.11 Their shifts on the left—right scale are usually undertaken within limited areas of the left—right scale, in a type of zigzagging course.

On average, voters' perceptions of ideological changes in party manifestos is low: if a party changes its left-right position in terms of its manifesto, the mean of respondents does not position the party significantly more to the left or right than the mean of respondents in the previous election (Adams and Somer-Topcu, 2009).12 If voters do not follow the position change of a party, their perceptions of the party's left-right position will be worse, and the stronger the party's position becomes. It is hence hypothesized that.

(H1). The more the position of a party shifts over time, the lower the voter's estimation proximity is to the party's ideological position.

Over time, change sometimes results in party positions overlapping or in parties 'leapfrogging' each other, but by and large, such leapfrogging only occurs between parties that are ideologically close to one another (Budge and Klingemann, 2001). The reason that parties only seldom cross each other's left-right positions seems to be that they frequently adapt their policies in the same direction as — and in response to — rival parties from the same party family (Adams and Somer-Topcu, 2009) and to which they are ideologically very closely positioned (Williams, 2015)13 However, even if parties do not leapfrog each other, multiple party changes might make the estimating of party positions precisely a more difficult task for respondents. A single party's change within a pattern of parties changing in the same direction might be less obvious. It should not be forgotten here that the task of estimating the left-right position of a party is very abstract, while parties' movement on the left—right scale in terms of their manifesto stances might be simply evoked by a specific new left or right issue added to their agendas. Voters of course will not calculate this change; they can only have an approximate picture of parties' left-right positions. Voters might even estimate party positions mainly by comparing one party to the neighbouring parties or by first thinking about which party is in the centre of the scale and then ordering the others around it, as research has suggested (Best, 2013). If multiple parties change in the same direction regarding one or a few issues, and this includes the one which the respondent

11 This statement must be tempered somewhat, because Klingemann et al. (2006) showed that, in Portugal in the first decade after its regime change, parties' left-right positions were also very unstable, similar to the 1990 situation of parties in the new democracies of Eastern Europe.

12 It should be noted that Adams and his colleagues' research was undertaken using longitudinal cross-sectional data and not panel data. The comparison of the party placement of the mean respondents is hence based on different samples. Moreover, the researchers calculated the differences between the means of party placement between years and not with regard to a specific question about party position changes. While the mean placement of a specific party should indeed be different if a party has changed its position over time, this type of measurement is slightly error-prone. For example, sampling problems and outliers can render the mean placements of different samples less reliable than if the same persons were asked about changes in party positions.

13 There is also evidence that niche party behaviour, in terms of position change, is

different than mainstream party behaviour, with niche parties responding less to

public opinion shifts (Adams et al., 2006) and therefore being more stable position keepers. However, because the focus of my analysis is on position change, this research will not further distinguish between mainstream and niche parties.

estimates, voters must ignore that all of these parties are taking more left (more right) positions now and will thus have 'moved' to the left (right) of the scale. However, in this situation, the specific party to estimate does not take a more left (right) position relative to the other parties than before. Its movement will be more difficult to notice and estimate than if its shift were the only shift that occurred. But also in a situation in which multiple parties have moved into multiple directions, the task of estimating a single party's position within that system of manifold party shifts is also likely to be more difficult than within a party system in which there are not many position changes over time. The second hypothesis is thus:

(H2). In a comparison between countries, the stronger the total of all of the parties' positions shifts are except for the estimated,14 the lower the voters' estimation proximity is to the parties' ideological positions.

3.2. Priming of topics in the election campaign

Apart from the hypothesized confusion of voters due to programmatic inconsistency, it is also likely that a large portion of current information does not reach voters at all because most voters do not inform themselves about programmatic positions directly but use the media as an intermediate actor. Because the media usually only cover a sample of current programmatic party issues, those that are 'primed' (Gidengil et al., 2002), this communication means will already create differences between issues in party manifestos and voter perceptions. Further, 'priming' not only leads to an incomplete picture of current party topics but also to an intraparty variation in how much information is presented. This effect does not always depend on the importance or popularity of this specific party, i.e., the 'party logic'. It can also depend on the 'media logic', according to the media's perceived salience of political issues (Hopmann et al., 2011; Mazzoleni, 1987; van Aelst et al., 2008). In any case, the priming of issues means that information about parties' current issue positions is not communicated to voters equitably.

In addition, in most election campaigns, it is more likely that leadership, rather than issues, is primed (Gidengil et al., 2002). According to Norris et al. (1999), the priming of issues within an election campaign only occurs if a very new issue becomes salient. If the political debate revolves around well-known policies, the media focus on political leaders. The question is whether this media concentration on political personalities decreases voters' information about parties' current ideological positions. Although empirical studies have shown that, in neither German (Kaase, 1994; Schoenbach, 1996) nor American elections (Hayes, 2008), 'personalized' voting occurs at the expense of 'issue voting', this type of issue voting could nevertheless be based on insufficient, inaccurate or even outdated information. In other words, the personalization of an election campaign might lead to less accurate or less updated information about current policy positions. The less absolute hypothesis made here is that:

14 It is important to note here that, to differentiate later between the effects of the change in the estimated party itself as hypothesized in H1 and H1b and the effects of all other parties, the total of the parties' change effect in H2 does not include the party change of the specific party, which is estimated in one party-voter dyad. Because both variables will be integrated into the later analysis, the single party change variable controls for this single party's part of the effect within the mean party change variable. In other words, due to the control of the single party's effect, an effect of H2 will be purely the result of the other parties, rather than being the result of the voter's specific party estimation.

(H3). The more a campaign is personalized (i.e., focused on party leaders and personalities), the lower voters' estimation proximity is to current manifesto positions.

Apart from the personalization of politics, other frequent valence issues within election campaigns are parties' speculations about possible governmental coalitions and party performance. These speculations might interfere with the expected negative impact of personalistic topics in an election campaign because discussion of possible political partners and past performance can enhance the clarity of parties' positional similarities and differences. Hence, a control variable will be added that accounts for these party-politics-related topics.

3.3. Age of the political regime as a proxy for party system institutionalization

Party systems in Eastern Europe are very unstable, as indicated by strong party fragmentation, high rates of electoral and legislative volatility and low party anchorage within societies (Rose and Munro, 2009). Some of these characteristics should be interrelated with the hypotheses stated above about potentially stronger party position shifts and the possibly stronger priming of political leaders. Other characteristics, such as assumed greater party knowledge in multiparty systems and potentially more polarized competition in systems with greater numbers of effective parties, will be accounted for with control variables. However, there are specific characteristics of young party systems that might add to voters' confusion. These characteristics roughly all relate to party-voter linkages, which have been found to be rather weak (Mainwaring and Torcal, 2006). For example, frequent changes on the party supply side in the early years after the democratic transition were a cause for, rather than a consequence of, high volatility rates (Tavits, 2008). Similarly, these changes might demotivate disappointed voters for electorally unsuccessful parties from paying attention to elections and party campaigns. Furthermore, in the first years after regime change, Eastern European parties in office often either did not fulfil pre-election promises or, vice versa, did not mention important policy changes before they were elected (Roberts, 2010; Toka et al., 1997). Apart from the possible increase in disappointment or distrust in party positions to which such behaviour could lead, unexpected policies from left (or right) parties might also have added to confusion. An example is that, between 1989 and 2004, leftist party governments in 13 post-communist countries decreased social spending more than rightist party governments (Tavits and Letki, 2009).15 Hence, although voters have general knowledge about the left-right party heuristic (Rudi, 2010b), there are causes for distrust or confusion. While party behaviour differences are diverse and should also be related to differences in systems and societies within Eastern Europe, their unifying aspect still seems to be that they are all related to less experienced parties within less experienced systems of parties, with loser party-voter ties and loser inter-party relationships. When party systems become more stable over time in terms of their competing actors (Rudi, 2010a), interparty competition around political issues and interparty relationships in coalition building (Warwick, 1996), party-voter linkages and position knowledge should increase as well. Because aspects of parties within party systems are too diverse to be modelled completely in this analysis, two control variables, which measure the party age

15 They explain this phenomenon by the leftist parties' more compelling need to demonstrate their belief in democracy and a capitalist market economy, while the rightist parties need to demonstrate responsiveness to the electorate, which suffered hardship due to the economic transition.

and the age of the democracy, will serve as proxies for specific party characteristics in young democracies.

Further control variables are those that have shown to be important in studies of political knowledge (Gordon and Segura, 1997; Dahlberg, 2013). Following these studies, it is expected that multiparty competition leads to clearer party competition lines and hence to more exact voter estimations, while an even greater number of effective political parties (ENPPs) can also lead to less clarity due to 'polarized pluralism' in the party competition structure and cross-cutting of party competition lines (Sartori, 1976). In contrast, the polarization of the party system, the party's mean left-right divergence from all other parties weighted by their vote shares (Dahlberg, 2013) and its affiliation with one of the left-right party families should increase voters' accurate perceptions due to enhancing clarity of left-right differences between parties. The previous vote share of the estimated party in the previous election is used as a proxy for its popularity, and the party's previous participation in government, expressed by the dummy variable 'in government t-1 ', should decrease voters' ability to estimate accurately due to the expected decreases in position focus and clarity. At the individual level, the analysis will control for both the more objective ideological distance and the 'felt' affective closeness to parties, and for sociodemographic characteristics that affect political interest or voting behaviour, such as education status, age (Franklin et al., 2004; Gallego, 2009; Milner, 2002) and gender (Bennett and Bennett, 1989; Inglehart and Norris, 2000).16

4. Data sources, methodology and operationalization

The analysis combines data from eligible persons17 in 25 postlegislative election studies from the Comparative Study of Electoral Systems (CSES Modules 1 and 2, Full Releases) with the left-right party positions of the '12-step index' of parties' left-right positions (Franzmann and Kaiser, 2006),18 which are based on the raw values of the Comparative Manifesto Project (CMP) (Klingemann et al., 2006). Using data from the CSES guarantees a timeframe that is as close as possible to the time of the election, the prior election campaigns and media coverage of the most important political issues and manifesto stances. To ensure comparability in terms of democratic institutions, the data are restricted to election studies from Western and Eastern European democracies19 that were at that time rated 'free' (Freedom House, 2014). Because changes in party positions are of particular interest as an independent variable, the data are further restricted to parties that were also voted into the previous legislature20. The variables are distributed on three analytical levels with a nested data structure: 150,642 individual respondents' estimates are linked to 113

16 The construction of the control variables is presented in Appendix 2.

17 The term 'eligible persons' is more precise than 'voters', because not only persons who cast a ballot are included in the analysis. In the following, when using the term 'voters', I nevertheless refer to this group of 'potential' voters, regardless of whether they actually make use of that right.

18 The 12-step data are available for download from the webpage of the University of Potsdam, http://www.uni-potsdam.de/db/ls_regierungssystem_brd/index.php? article_id=498&clang=0 (last checked July 1, 2015).

19 These election studies have sometimes been divided into different parts of one country for historical reasons (East and West Germany) or because parties are organized along an ethnic division (as in Belgium). Data from the USA, Japan and Belgian Wallonia were excluded because they did not contain some of the important variables in this study. Other Asian and Latin American countries' data were excluded due to data restrictions regarding the index data used (Franzmann and Kaiser, 2006). The full list of election studies and parties included in the analysis can be found in Appendix 1.

20 The restriction to parliamentary parties is due to the CMP data, which mostly

covers only the manifestos of legislative parties, but it also ensures that only parties

with a chance of being known by the bulk of voters are considered.

political parties, which in turn are nested within 25 national countries' election(s) (studies) with specific cultural and electoral system characteristics. Due to this data structure on three levels, the analysis employs a hierarchical linear mixed model, which enables the breaking down of the variation in individual respondents' estimates into 'within- and between- (party and system-) components' (see e.g., Bryk and Raudenbush, 1992). In other words, it assesses simultaneously how strongly individual, party and system determinants account for the variation in estimation proximity.

4.1. The dependent variable

The dependent variable, 'estimation proximity', measures the space between a voter's subjective estimation of a party's left-right position and its position based on party manifesto data. To connect all of the party values per country to the respective voters' estimations, the data are arranged with party-voter dyads: for each respondent, it includes as many party estimates as a respective country's respondents' are asked about and for which data are available in the CSES and the 12-step index database. This approach is similar to Dahlberg's (2013) approach, but in contrast with this and other previous research, the left-right '12-step-index' (Franzmann and Kaiser, 2006) is used here as the reference point for the current left-right manifesto location. This '12-step-index' is based on the 'raw' data of single issue saliencies in party manifestos (Klingemann et al., 2006). The use of these manifesto data is important for my research question because it seeks determinants of how strongly voters' concepts of parties are linked to the current party manifestos. Moreover, to examine whether voters' ideological estimates are influenced by parties' previous changes in position, I need data containing not only the present party positions but also the previous positions.21 Before demonstrating the operationali-zation of the dependent variable in greater detail, I will discuss the appropriateness of CMP data in general and, more specifically, the use of the '12-step index' more thoroughly.

4.1.1. Seeking parties' objective left-right positions: which index is best?

To answer the empirical question of how well a person estimates a party's ideological position, it is first necessary to find a valid measurement of the party's true, objective left-right location, which is a point of scientific debate. It strongly depends on a valid definition of what is meant by left and right. Although different theories are referred to, and used as a base for the construction of left-right indices, for example Marx (Budge et al., 2001) and Bobbio (Jahn, 2011),22 it can be doubted that these theoretic sources are valid across geographical space and time. It is well known that the sub-dimensions of the left-right space, the economic and the non-economic value dimensions, do not cluster everywhere in the same manner (Kitschelt, 1995; Marks et al., 2006). In Western Europe, most parties connect liberal market values with traditional, conservative 'freedoms and rights' values and 'left' economic values with liberal 'freedoms and rights'. However, most Eastern European

21 To my knowledge, there are no survey data that contains voters' placements of both the current and the previous left-right positions of parties. According to Meyer (2013), even political experts' party placement questions on election surveys are often not time-specific, either asking vaguely about 'political parties today' (Huber and Inglehart, 1995) or not mentioning any time dimension (Benoit and Laver, 2007; Castles and Mair, 1984). The estimations thus reflect the experts' expectations about a party's position in a general sense. This lack of exactness is also indicated by the stability of expert judgements in longitudinal comparison (Meyer, 2013: 92).

22 Also, see Franzmann (2013).

parties connect those political sub-dimensions conversely (Hollander, 2003; Kitschelt, 1995; Markowski, 1997; Marks et al., 2006). Also the emphasis on each of the sub-dimensions varies internationally (Rovny and Edwards, 2012): while, in most countries, the economic dimension is dominant, in Austria, Spain, Poland and Slovenia, the 'freedoms and rights' dimension is the more important one. It is hence useful to employ a measurement of ideology that considers the culture-specific meaning of left and right as much as possible.

The CMP raw data are a good starting point because they simply represent the proportion of different political issues that a party mentions in its manifesto, and unlike other databases, they contain position data for a long period of time. It is hence possible to show how positions change. The CMP was confronted with the critique that the data might be prone to procedural errors23 (Benoit et al., 2009). However, other left-right codes of parties' data, such as those compiled by experts, can also have errors, and the coding of left-right positions is then often less transparent than the CMP proceedings. Moreover, the CMP shows that its means for addressing potential coding problems are sufficient24 to produce data that are at least free from systematic, over-time or cross-national bias and that show high validity scores (Klingemann et al., 2006, 103).

What might be of more concern than the reliability of the raw CMP measurements is the question of how best to calculate an index from the data. Researchers have found several methods to accomplish this goal (Budge, 1987; Franzmann and Kaiser, 2006; Gabel and Huber, 2000; Jahn, 2011; Klingemann, 1995; Laver and Budge, 1992). In essence, the approaches are different in terms of how researchers sample the single issues that are important for left-right cleavage. For most indices, researchers rely on a specific theoretical perspective of the left-right dimension and the single issues that belong to each side of the spectrum. Factorial analyses are then run with a different number of left and right 'marker' items, with variations in how many issues are sampled (Budge, 1987; Laver and Budge, 1992; Klingemann, 1995; Gabel and Huber, 2000).25 Most approaches arrive at an index based on the whole country sample that they use. Only the '12-step approach' (Franzmann and Kaiser, 2006) produces country-specific indices. Briefly, Franzmann and Kaiser first distinguished between valence and position issues by country, and they also differentiated between different time spans if the issue structure changed. Using only those position issues for which parties within a party system were really different from one another, they ran regression analyses of party positions on overall party system mean issue positions. This analysis was performed issue by issue, to determine which of the position issues were left and which were right in comparison to a reference party. Only then did they calculate the specific left-right party positions by subtracting the individual party's saliency value from the lowest saliency value of the party system ('base value'). The 'raw values' arrived at were smoothed using the means of the raw values of three election years (the mean raw scores of the previous election with the current election and the next election (Franzmann and Kaiser, 2006; Franzmann, 2013). The choice of single issues for the country-specific indices is hence very pragmatically dependent on the empirically observable left-right

23 Sentences in the manifestos are hand-coded into 114 single issue categories. The relative frequency with which an issue is mentioned is deemed to indicate how salient the party believe that the issue is (Budge and Klingemann, 2001; Klingemann et al., 2006).

24 Potential inter-coder problems are addressed in the training of coders, as well as during the production of the data phase (Lacewell and Werner, 2013).

25 For a more comprehensive comparison, see Dinas and Gemenis (2010).

dichotomy around which country-specific policy debates revolve. This culturally sensitive approach to ideology makes it most suitable for my endeavours.

In the CSES data, respondents' left-right party estimates are captured by the request to rate each party on an 11-point scale. The '12-step index' and the CSES scale run from 0 (leftmost position) to 10 (rightmost position). To calculate how well respondents estimate a party in comparison to the position data, I subtract the respondent's left-right estimates of a party (originated in the CSES data) from the respective party's index value (12-step data). If a respondent's estimated values are the same for all parties, the answers are recoded as missing. All non-response answers — 'don't know', 'refused', etc. — are also coded as missing and are excluded from analysis.26 Negative values are recoded into natural (positive) values because I am interested in the proximity measurement and not in whether someone estimated a party to be further to the left or to the right of its index position. The resulting variable has a skewed kurtosis and is thus logarithmized.27 Because the values so far represent the difference between estimated and manifesto position values, the scale is then inverted. In the resulting variables, higher values hence indicate greater proximity, and lower values indicate lower proximity of the voter's estimate to the index party position.

4.2. Independent variables

The most important independent variables relative to the hypotheses above are individual party change, the mean parties' change and the priming ofleaders within the election campaign. Party change is calculated by subtracting a party's previous position from its present position. Because only the strength of change is important here and not its direction, negative values are recoded as positive values. The mean parties' change is measured as the mean of all parties' position shifts from the previous to the current manifesto's left-right positions. The variable personalization of the election campaign is based on a CSES variable that captures national political experts' open answers to the question '[ ... ] what are the five most salient factors that affected the outcome of the election (e.g., major scandals; economic events; the presence of an independent actor; specific issues)?' For the construction of the variable, the number of personalistic issues mentioned by the national political expert is counted. 'Personalistic' issues are defined here as those issues that refer either to specific politicians without connection to their policies or to more general mentions of lead-ership.28 The variable represents the number of these personalistic campaign issues out of the five possible mentions by political experts. All of these issues are considered to be 'valence' issues because they do not relate to any confrontational political dimension but only to an evaluation of political personalities, which are grounded in shared norms such as 'trustworthiness' or 'being a good leader' in a more general sense.29

26 The imputation of values for 'don't know' answers was not a viable alternative to the pairwise exclusion of all missing values, because the data of most of my sample countries only used the 'missing' code for this question, rather than differentiating among 'don't know', 'refused' and 'missing'.

27 The range of the party-voter means of the original (positive) gap values can be found in Appendix 3.

28 For the present sample, these are 'Personality of Jacques Chirac' and 'Right-Extremist Jean-Marie Le Pen's success at the [preceding, K.B.B.] presidential election' (both France), 'Resignation of the Prime Minister' (Portugal), 'Promises/Trust of leadership' (mentioned in Britain, Hungary, New Zealand, Sweden), 'Leadership' or 'Personalization of the election campaign' (mentioned in Australia, the Czech Republic, Germany, Israel, Netherlands, New Zealand, Switzerland), and 'Images of party leaders' (Finland).

29 For the construction of the control variables, see Appendix 2.

Table 1

Mean correlations of party estimations with party manifesto indices.

Index Mean

12-step index 0.73a

RILE index 0.64a

a a = .05

5. Estimating parties' ideological positions: some descriptive statistics

As an initial overview of how voters' estimations generally fit the party index values, the mean correlations between parties and voter estimations are shown (Table 1). For comparative reasons, the correlation test is performed with both the 12-step index (Franzmann and Kaiser, 2006) and the RILE index (Budge and Klingemann, 2001). Using the values from the '12-step-approach', the correlation value is 0.73 with a 0.05 level of significance. The RILE index values are equally significant, but they correlate according to a lower value of 0.64 with the mean party estimates, lending some support to the view that the '12-step-approach' seems more appropriate30.

In general, there is evidence that party changes on the left—right scale lead to less precise estimations, as Fig. 1 shows.

With a greater mean party change within a country, voters' estimation proximity to parties' left-right ideological positions is lower. Fig. 2 graphs the bivariate relationship between parties' mean changes per country and voters' estimation proximity for the younger and older democracies separately.

This figure shows that in general, the mean proximity of voter estimations to party positions is lower in the younger democracies. Moreover, the impact of mean party change on misplacement is not as strong as in the old systems, as indicated by the less steep slope (Fig. 2).31 This first overview of the data leaves the impression that party and system characteristics can be important determinants of voters' estimation proximity to party positions. To arrive at a reliable answer to why individual voters' estimates are more or less manifesto-oriented, it is, however, necessary to run a systematic analysis that considers all levels of analysis simultaneously.

6. Regression analyses

6.1. Position changes of parties on the left—right scale and issue domination in the election campaign as determinants of estimation proximity

In the following, two three-level hierarchical linear models are run, which contain the individual, party and system level impact factors. Model 1 employs the party left-right position change on the party level and the corresponding mean of all parties' changes on the system level, while model 2 employs the variables for single and mean party changes on the economic sub-dimension of left-right ideology (Table 2). The calculation is meant as a robustness check for model 1, but it also sheds additional light on the way in which position changes can lead to voters' confusion about parties' left-right positions. For means of comparison between the strength of the coefficients within the model, the first model is also run with

30 Due to limitations of space and the word count, a more comprehensive overview of the proximity and the raw 'gap' had to be moved to the appendix; see Appendix 3.

31 For a full list of respondents' proximity ranges within countries, see Appendix 3, Fig. 3.

Fig. 1. Mean parties' left-right changes and estimation proximity.

• • . .

\ N • •

• •

.6 .8 .2 .4

Mean of all parties' changes per country

MResProx

-----Fitted values

Fig. 2. Mean parties' left-right changes and estimation proximity by 'young'.

z-standardized variables.32

Model 1 confirms Drummond's (2010) finding that respondents who have strong emotions towards a party, as expressed by the dummy variables 'Like Strongly' and 'Don't Like', estimate the party less accurately than those whose emotions fall within the normal range. The unstandardized-coefficients reveal that respondents with strong antipathy towards a party estimate it 0.17 points or approximately 15.5 percentage points33 less accurately than

32 The coefficients of the one- and two-level models are not presented here because they do not add much information to the results. They can, however, be found in Appendix 6.

33 0.17 points here equals approximately 15.5% due to the 11 point left-right scale.

persons with no extraordinary emotions. However, the strongest effect on estimation proximity on the individual level yields the objective ideological distance to the party (Model 1, b-coefficients).

Those respondents who place themselves farther away from the party thus also have more difficulty in positioning it accurately in this case not due to emotions but probably due only to lower levels of electoral interest in it. This outcome and the finding of a negative effect of 'ideological distance' confirm Dahlberg's (2013) analysis of 'perceptual agreement'. It is easier for voters to estimate the left-right positions of parties to the ideologies that they objectively share. In contrast, the exact policy positions of parties, with which voters do not share common ideological ground, are less familiar. In general, a respondent should be best able to estimate a party's

AUS 1996 BELF1999 BGR 2001 CAN~1997 CHE-1999 CZE~1996 DEUOI1998 DEU 1998 DNK~1998 ESP_1996 FIN~2003 FRA~2002 GBR~1997 HUN_1998 IRL~2002 ISL~1999 ISR_1996 NLD~1998 NOR_1997 NZL~1996 POL_1997 PRT~2002 ROLT1996 SVN_1996 SWE~1998

1.5 2 2.5

excludes outside values

Fig. 3. Proximity of respondents' estimations to Left-Right positions.

Table 2

Determinants of respondents' estimation proximity to left-right party positions

Fixed part Model 1 Model 2

13 coefficients (standardized) Beta coefficients (unstandardized) Beta coefficients (unstandardized)

Intercept 3 -0.045 1.509*** 1.384***

Don't Like -0.129*** -0.167*** -0.167***

Like Strongly -0.087*** -0.099*** -0.099***

Ideological Distance -0.127*** -0.037*** -0.037***

Education* 0.056*** 0.016*** 0.016***

Age* -0.014*** -0.000*** -0.000***

Female -0.037*** -0.037*** -0.037***

Party's Change 0.020 0.033

Party's Ecochange 0.078+

Previous Vote Share -0.050 -0.002 -0.004*

Party Age 0.075* 0.001* 0.001*

Left-Right Family 0.013 0.015 0.024

Party Divergence -0.090** -0.140** -0.128**

In Government t-1 -0.045 -0.046 -0.027

Issue: Party Politics 0.130*** 0.092*** 0.052+

Issue: Leader Politics 0.093** 0.069** 0.090**

Mean Party Change -0.146*** -0.544***

Mean Party Eco Change -0.345**

Party System Polar. 0.050+ 0.027+ 0.021

Multiparty System 0.093* 0.119* 0.100+

ENPP -0.113* -0.044* -0.018

Age of Democracy 0.160*** 0.003*** 0.003***

Random effects Model 1 Model 2

Variance comp. df c2 p-value Variance comp. df c2 p-value

Intercept 1, r0

Ideological Distance Slope, r2 Level 3, Intercept1/2 u^ Ideological Distance, Intercept 2, u20

Log Likelihood N (individual/party dyads) N parties N systems

0.164 0.028 0.021 0.011

-887628400.000

194351

79 85 17 24

4633.088 1227.037 23.387 40.378

<0.001 <0.001 0.137 0.019

0.026 0.001 0.002 0.000

-887661700.000

150642

79 85 17 24

4558.296 1226.923 31.658 40.059

<0.001 <0.001 0.017 0.021

***p < .001, two sided, **p < .01, two sided, *p < .05 two-sided, +p < .05 one-sided.

Because the number of level 3 units was too small, the fixed effects were calculated with non-robust standard errors. The differences in estimation of fixed effects with robust standard errors were negligible. All of the effects were fixed except for 'Ideological Distance' on both the party and system levels. Education and Age are group mean centred. The random effects part displayed here is calculated with the unstandardized analysis. The chi-square statistics reported above are based on only 110 of 113 level 2 units, which had sufficient data for computation. Fixed effects and variance components are based on all of the data.

position if he or she occupies a close ideological position to it but is not compromised by strong feelings. The psychological 'assimilation' and 'contrast' effects are the results of overstating the similarities (or differences) of a liked (or disliked) party to one's own positions (Drummond, 2010).

All control variables on the individual level show significant but only moderately strong to almost negligible effects. They reveal that better-than-average educated men estimate party positions more accurately than others.

On the party level of analysis in model 1, only the party's age and its positional divergence from other parties in the party system are significant. The moderately strong negative effect of party divergence on the estimation indicates that, contrary to Dahlberg (2013), parties that occupy more divergent positions than all other parties are estimated less accurately. The reason for this finding might be the same as for the negative effect of respondents' individual distance from the party: more divergent parties' positions might simply be known less accurately due to fewer voters paying attention to them.34 Additionally, older parties are estimated more accurately. These parties might be known to more voters or might be considered more reliable than younger ones. An increase of 10 years in age only leads to an increase of approximately one percentage point in estimation proximity, but moving from the youngest to the oldest party in the sample, the difference leads to a more than 13 percent increase.

A party's change in its left-right position does not lead to a significant effect on voters' estimation proximity. Hence, Hypothesis 1 cannot be confirmed here. In contrast, a party's change on the economic sub-dimension is positively related to estimation proximity (model 2). It increases voters' ability to estimate the party precisely. This difference from model 1 could indicate that, from the voter's perspective, the economic sub-dimension is linked more strongly to the concept of left-right ideology, whereas the non-economic, value-oriented sub-dimension's relationship to left-right ideology is less clear. This assumption is also supported by a closer examination of the individual situations in which a party's strong shift of the left-right position is not connected to a low estimation proximity, or party misplacement, but rather to more precise estimations.35 In these cases mostly 'mainstream parties' like the German SPD, the Portuguese CDU, Iceland's Progressive Party or the Swiss FDP are involved, i.e. electorally favoured, well-known parties whose issue domains are connected rather strongly to the economic left-right dimension.36 All of these parties came in the first four places in the parliamentary elections, in multi-party systems. Their political stances and electoral campaigns should have been mostly well observed both by the media and the electorate. The fact that these parties' shifts on the left—right scale did not produce negative effects in terms of voters' left-

34 In an earlier version of the analysis, in which the divergence variable was not included, the previous vote share showed a negative effect, which was significant at level p < .001.

35 The mean party shift to the left or right equals 0.34 and the standard deviation from this is 0.27. If a shift to the left or right equals or exceeds this mean plus one standard deviation, i.e. the value equals or is bigger than 0.61,1 call this a 'strong shift'. The mean estimation proximity equals 1.39 and the standard deviation isv.5. If the estimation proximity is lower than 1.39 minus half of the standard deviation, i.e. lower than 1.14, I call this a 'low estimation proximity' or 'bad estimation'. Also see Appendix 5 with the full list of parties' left-right change values, mean party proximity values and the qualitative evaluation of these, based on the above mentioned thresholds for strong shifts and low proximity values.

36 Other parties which shifted quite strongly were estimated with about mean precision or only a bit less precise; (within the range of up to the mean plus or minus a half standard deviation). These are Australia's LP, Belgium's CVP, the French UDF, and RPR, Great Britain's Conservatives, Hungary's MSzP, Fidesz-MPP and FKgP, and the Swiss SVP.

right estimations, but rather the opposite, might mean that in these cases, left-right shifting on the economic sub-dimension even drew voters' attentions to the respective political positions, probably due to the parties' popularity, their usual strength on the (economic) left-right issue domain and the saliency of the left-right cleavage in the respective party systems.37 The point to take away, here, is that the effect of party position shifts on voters' estimation precision can really be either positive or negative, most probably depending on the voters' general acknowledgement of the specific party's (economic) issue domination. Position shifts leading to more precise estimations should be rather the exception — and seems to be strongly dependent on the party which shifts.

Turning to the system level determinants of the analysis, the variable for the mean total system change is negatively connected to estimation proximity, confirming Hypothesis 2. It is significant in both models 1 and 2, although the mean parties' total change effect in model 2 is slightly less pronounced than that of mean total change on the left—right scale in model 1. The stronger the mean change is of parties on the left-right axis (model 1) or on the economic sub-dimension of the left—right scale (model 2), the less accurate the respondents' estimations are, or at least the less closely oriented to the manifesto positions they are. While a single party's change does not necessarily lead to confusion of left-right positions or even increase voters' estimation proximity (as in model 2), a system of at least two or more parties' changes adding up seem to be problematic for the voters' estimation capacity.38 In all countries, in which the mean of all parties' position shifts are strong, the reason was medium to strong party positions shifts of at least two or more parties.39 The analysis confirms the hypothesis that in at least some cases, the overall turmoil of party positions within party systems can lead to less precise estimations even of parties which do not shift strongly.

Comparing models 1 and 2, the other variables on the system level all show the same direction of influence, although the coefficients' strength and levels of significance change to some extent. The variable 'age of democracy' controls for possible systematic bias due to the lower level experience of most parties and voters in the younger party systems. Adding to the party age impact, it confirms that, apart from the age of the specific party, the age of the system in which it operates is also important. Where party systems are less institutionalized, as indicated by ever changing coalitions, weaker

37 There are also a few parties whose positions were estimated very imprecisely, even though the respective parties did not change their positions very strongly. These are the Belgium FB, the Bulgarian UDF, the Canadian Reform Party, the Czech ODS, the Finish Green League, Iceland's FSF, the Romanian PDSR, and UDMR, and the Slovenian SD and SKD. The reasons why these parties were not estimated well seem to be more dispersed than in the aformentioned group. Lower levels of popularity and/or non-left-right issue dominated stances might be reasons for less precise estimations for some of them, like the Canadian Reform Party, the Finish Green League, and the Hungarian UDMR. Another reason might be a previous membership in an electoral alliances and/or previous governmental coalitions (the Slovenian Parties, the Finish Green League). More research into individual party characteristics and behaviour would be needed here, to fully explain the voters' misplacements of these parties in a more differentiated way.

38 It should be remembered here that the total party system change is calculated as the sum of all parties' changes within the system. The specifically estimated single party's change is, however, controlled for with the party change variable, because the dependent variable expresses the estimation proximity for specific party-voter dyads on the individual level and not a mean for each voter. A one-party change can hence not be the cause of the effect of total party change decreasing estimation proximity, but it must be a pattern of at least two or more parties changing, leading to voter confusion.

39 See Appendix 5. 'Strong' means that the mean of all party position shifts within a country equalled or was stronger than the mean plus half of the standard deviation (>=0.4). The exception for this accumulation of at least two parties' position shifts leading to strong mean values is Iceland, where only the Progressive Party shifted strongly.

party-voter ties, and lower levels of political trust, party positions are more obscure or might in general even seem arbitrary and interchangeable to voters. Lower levels of attention or more disbelief in party promises and political disenchantment could be causes for less accurate party position estimating in the younger party systems. In that vein, the effect for age of democracy is rather connected to the institutionalization of party positional domination or issue saliency over time. The older the democratic regime is, and the more clear-cut and repeated the positions are, the more voters pay attention to them and acknowledge them.

Similarly, both the debate about possible governmental coalitions and the priming of political leaders seem to increase attention to the left-right positions of the manifestos. On the one hand, this outcome is surprising because the priming of personalities means that there are fewer left-right issues that are highly salient before the election. On the other hand, the finding fits well with the earlier cited studies, which showed that personalized voting does not decrease issue voting (Kaase, 1994; Schoenbach, 1996; Hayes, 2008). The analysis here even shows that leader politics can increase correct ideological placing — likely due to an increased level of attention. If political leadership is primed by the media during election campaigns, increased attention to these is not automatically bad for peoples' awareness of political issues. It is not a zero-sum game. This result can be at least deduced from the analyses here, which control for position change or the parties' change on the economic sub-dimension. In other words, when nothing particularly interesting occurs on the side of parties' left-right positions, and the media then concentrates on leader characteristics, this focus cannot harm existing political (position) knowledge. In contrast, the public's attention might be raised, and knowledge about positions is refreshed.

Turning to the control variables on the system level, in model 1, they are all significant and in the expected directions. Both models show that, in multiparty systems, voters estimate parties more precisely — probably due to the enhanced 'incentive structure' (Gordon and Segura, 1997) to inform oneself if more parties have a chance of entering government. The difference between party systems of less to more than 2.5 parties is a 10 percentage point increase for the larger systems, as the Beta coefficients (model 1), divided by 0.11 indicate. However, an increase in the number of parliamentary parties beyond that multiparty threshold is either negatively connected to estimation proximity, as was expected, or is not significant (model 2). Drummond's (2010) hypothesis that, in multiparty systems, a greater number of parties reduces partisan error due to more 'crowding' on the left—right scale is thus too simplistic because it neglects the nonlinearity of the effect: within multiparty systems, if more parties enter a race, the voters' estimation proximity rather decreases again. The reason for this outcome might be the greater probability of ideological division lines cross-cutting each other and thereby leading to a situation of a 'polarized pluralistic' competition structure (Sartori, 1976) — or even the overload of information, leading voters to lower cognition of single ideological positions. Although the number of parties and the multiparty dummy variable yield quite strong effects, in model

1, 'Polarization of the party system' still has a small but one-sided impact on estimation proximity to the manifesto values; in model

2, the effect is no longer significant. Ideological Distance, the difference between a respondent's self-position estimation and that of a party, has a varying influence on voters' estimation proximity, both on the party level and on the system level. The strength of the effect on estimation proximity thus depends slightly on the party and also on the country. In some party or country contexts, voters' ideological distance from the parties has a greater effect on estimation proximity than in other contexts. However, the variation is only very small, as the variance components in Table 2 show. In the

next section, the question of why single party change does not lead to confusion but rather increases voters' estimation accuracy is addressed again. This third analytical model will advance on differences between more and less consolidated democracies.

6.2. Party change on the economic scale in established and less established party systems

Building on the finding that change in the economic left-right sub-dimension is related differently to voters' estimation positions depending on the age of the democratic system (Fig. 2), a third model of analysis is run, which differentiates between the older, well-institutionalized party systems and the younger, less experienced ones (Table 3). With all else being equal as in model 2, it advances with a dummy variable 'young' for differentiating between young and old party systems and an interaction variable of 'young' times the mean position change on the economic subdimension, to determine whether change really has a different effect in the younger systems.

As it turns out, voters' position estimations are significantly worse in the younger party systems. The direction of influence of the two change variables is the same as in model 2: a single party's change in economic position leads to a slight increase in voter

Table 3

Respondents' estimation proximity to left-right party positions with position changes on the economic sub-dimension and interaction effects of young party systems * changes

Fixed part Model 3 Beta coefficients, (unstandardized)

Intercept 3 1.549***

Don't Like -0.167***

Like Strongly -0.099***

Ideological Distance -0.037***

Education* 0.016***

Age* -0.000***

Female -0.037***

Party's Ecochange 0.073+

Previous Vote Share -0.003*

Party Age 0.001*

Left-Right Family 0.018

Party Divergence -0.107*

In Government t-1 -0.021

Issue: Party Politics 0.038

Issue: Leader Politics 0.079*

Overall/Total Eco Change -0.459**

Multiparty System 0.064

ENPP -0.000

Age of Democracy 0.002

Young -0.230+

Young*Overall Eco Change 0.357+

Random effects Variance comp. Df c2 p-value

Intercept 1, r0 0.026

Ideological Distance Slope, r2 0.00

Level 3, Intercept1/2 u00 0.001

Ideological Distance, Intercept 2, u20 0.011

79 4575.933 <0.001

85 1226.915 <0.001

16 27.115 0.040

24 40.171 0.020

Log Likelihood N (individual/party dyads) N parties N systems

-88764960000.00

194351

***p < .001, two sided, **p < .01, two sided, *p < .05 two-sided, +p < .05 one-sided. Because the number of level 3 units was too small, the fixed effects were calculated with non-robust standard errors. The differences in estimation of fixed effects with robust standard errors were negligible. All of the effects were fixed except for 'Ideological Distance' on both the party and system levels. Education and Age are group mean centred. The random effects part displayed here is calculated with the unstandardized analysis. The chi-square statistics reported above are based on only 110 of 113 level 2 units, which had sufficient data for computation. Fixed effects and variance components are based on all of the data.

estimations, while the aggregate change of party positions on that same economic scale leads to voter confusion. The interaction effect of 'young democracy' * Overall (mean) Eco change is, however, positive, indicating that, in the young systems, the negative effect of mean change on position estimations is much less pronounced than in the older Western democracies. This moderating effect is only found in the model with the change in economic subdimension variables. It is not significant in a model that adds the same type of interaction term to the model with the overall left-right change variables.40 The effect of parties' mean change in left-right positions thus depends on the question of the types of issues that change. If only the mean change on the economic subdimension is tested for, this does not confuse voters in Eastern Europe as much as it does in the older, more established party systems. In contrast, if the full left-right position change is considered additionally, which also consists of the non-economic values dimension, it is in general even more difficult for voters to estimate party positions accurately. This difficulty provides an important insight into the understanding of how the left and right sub-dimensions — the economic and the non-economic values' dimensions — are understood to constitute left-right ideology. To build on Neuman's terminology, to Eastern Europeans, the conceptual integration of the non-economic values' dimension into the ideology paradigm is less clear than the economic values' dimension.

It is interesting that, in situations in which parties are still very much in the process of institutionalization, movements of parties on the economic dimension do not decrease position estimation proximity as strongly as in Western societies. Perhaps in the West, parties' positions are expected to be rather fixed, and voters rely more on their long-term experience. It should also be considered that, in the older systems, the level of party position estimation is better to begin with than in most of the young democracies. The percentage of voters who pay attention to politics is limited or perhaps even almost saturated in the older democracies. If two or more parties deviate from what voters are used to, estimations then can only worsen. In Eastern Europe, in which the overall estimation of positions is less precise, change still provides the opportunity for parties to differentiate themselves more and to set the demarcation line clearer than before. Voters who do attend to politics are then enabled to differentiate the parties better. In summary, the effects of parties' position changes on voters' estimations are not very clear cut. There are differences between established party systems and those in institutionalization stages.

7. Discussion and conclusions

The piece of research has analysed several determinants of voters' perception proximity to parties' left-right positions. From the three-level- analyses, it can be concluded that individual, party, and system level characteristics make a difference in voters' perceptions. One of the most important findings is that much of voters' confusion of party positions is due to the strength of multiple parties changing their left-right positions. To some part, stronger aggregate party position shifts between elections can explain less exact knowledge of party positions in the young, less institutionalized party systems in Eastern Europe.

The individual voter's objective estimation is mostly increased if he or she shares a close political position with the party, but it is strongly harmed by extreme emotional involvement. Voters should not like parties too strongly if they want a realistic view of a party's

40 Due to space restrictions, a table is not displayed here with this model. It can be

found in Appendix 6.

left-right position. Rose-coloured glasses seem to blur the perception of issues that do not fit the beloved picture. Similarly, it does not help if voters do not like a party at all. These psychological effects of both positive and negative affection on estimation proximity, as found in earlier studies of political sophistication (Drummond, 2010), are thus confirmed.

On the party level, it was shown that older parties are known better, but parties that take more divergent positions on the left—right scale are estimated less accurately. Why this party divergence effect does not point in the same direction as in previous research (Dahlberg, 2013) remains unclear. It was assumed that voters do not observe the more divergent parties' positions as well as the more 'mainstream' ones. This assumption also fitted to the observation that in cases in which voters estimated parties precisely despite comparatively strong position shifts, only popular parties who are dominant on the left-right economic subdimension were involved.

System and aggregate characteristics are very important determinants of how well voters perceive parties' left-right positions. There are two reasons for this fact: if voters want to make informed voting decisions, the precise differentiation between parties might be more necessary and useful under specific conditions, such as the multiparty setting. This finding confirms Gordon and Segura's (1997) view that voters seem to adapt to the party system context. Furthermore, the estimation of parties is easier if the party positions are more polarized and if the party competition is clearer due to there being fewer effective parliamentary parties.

The new findings of this study confirm the influence of party behaviour and issue priming within election campaigns. An individual party's position change does not necessarily lead to less estimation proximity to its current ideological position, and change on the economic sub-dimension can lead to even more precise estimations. The reasons for this effect not being very strong and pointing in a different direction should be multifold: It was assumed that the party popularity, the issue domain of the party, and the number of changes which happen simultaneously within the party system should be relevant. Voters' awareness of single party shifts seems to be higher for the more popular parties with traditionally strong economic issue domination.

Contrary to the finding that these single parties' left-right shifts on the economic sub-dimension could even enhance estimation precision, aggregated major change in left-right positions did confuse voters. It had a strong negative effect on voters' perceptions. This finding confirms that of Adams and colleagues' (aggregate data) analyses (2011, 2014), that voters often do not perceive change. If only the change on the economic sub-dimension is considered, it is, however, necessary to also differentiate between the older and younger democratic systems. In the younger party systems, position change on the economic sub-dimension does not decrease voter estimation proximity as strongly as in the older systems, likely due to the mostly, but not always, rightful expectation of voters in the West that the positions are rather fixed. Change on the overall left-right position, however, does not affect voters in the younger systems significantly differently than in the older systems. It was concluded that for voters in the younger democracies of Eastern Europe it is comparatively easier to see how the economic sub-dimension is related to the concept of left-right: When they are asked to place parties on the left-right dimension, they seem to refer stronger to this sub-dimension than to the non-economic one. More in-depth research into data covering a longer time-span would be needed to confirm these interpretations. How individual voters integrate specific single political issues into their understanding of parties' left-right ideologies might also require more in-depth analyses of qualitative data.

Contrary to expectations was the present finding that parties are estimated more closely to their issue positions if a leadership issue is

primed before the election. This insight adds to the research on 'issue voting', which has shown that personalized election campaigns do not result in less 'issue voting' (Kaase, 1994; Schoenbach, 1996; Hayes, 2008). Voters know parties' positions even if the media concentrate more on political leaders and on other valence issues. It is not a zero-sum game that decreases voters' knowledge about current party ideology. In light of the other effects in the analytical model, which considers change on the left-right axis and findings about issue priming (Norris et al., 1999), this result can be interpreted in such a fashion that, if not much change on left-right issues occurs, and the media concentrate on political personalities, voters keep left-right positions in mind. Their knowledge about left-right positions is even advanced, likely due to an overall greater level of attention if leadership is in question. To obtain a deeper understanding and to validate this possible causal chain, more research into the influences of party behaviour and communication is needed.

The present analysis started with some strong claims. Political theory, following Downs (1957), expects voters to know party positions at least roughly to make the best suited voting decisions. While this expectation is very plausible, it is also a strong generalization and a normative claim. A great part of the responsibility is in the hands of the parties who need to take care of strategic position taking which advances on their political scope. Further, it is not yet clear how voters who know party positions less precisely truly behave in the voting booth. More determinants than issue voting cues should be considered. While this paper investigated position knowledge, it remains for future analyses to examine in greater detail how more or less knowledge affects voting behaviour and how factors other than issue vote cues leads voters to different — or even similar — voting decisions.

Acknowledgements

Thanks to Jessica Fortin, Markus Quandt and Stephen Quinlan for their helpful comments in different stages of this research. I would also like to thank two anonymous reviewers of Electoral studies for their constructive advice which helped me to improve the article further.

Appendix 1. List of election studies and parties which are integrated into the analyses

Australia 1996: ALP Australian Labour Party, LPA Liberal Party of Australia, The National Party of Australia41

Belgium Flanders 1999: Agalev (Green), CVP Christian People's Party, SP Socialist Party, VB Flemish Bloc, VLD Flemish Liberals.

Bulgaria 2001: DPS Movement for Rights and Freedom, ODS United Democratic Forces.

Canada 1997: BQQuebecan Bloc, LP Liberal Party of Canada, NDP New Democratic Party, PC Progressive Conservative Party of Canada, R Reform Party.

Czech Republic 1996: Association for the Republic, CSSD Czech Social Democratic Party, ODA Civic Democratic Alliance, ODS Civic Democratic Party, KDU-CSL Christian Democratic Union.

Denmark 1998: CD Centre Democrats, KF Conservative People's Party, SD Social Democratic Party, SF Socialist People's Party, V Liberals.

Finland 2003: Green League, KESK Centre Party, KD Christian Democrats, KOK National Coalition Party, Left Alliance, SDP Finnish

41 For these parties some of the variables were not available, but due to their importance in their respective party systems, and the hypothesized effect of party shifts on voter estimations, their left-right shifts went into the mean calculation of party shifts.

Social Democratic Party, SFP/RKP Swedish People's Party of Finland.

France 2002: FN National Front, LV Greens, PCF Communists, PS Socialist Party, RPR/UMP Rally for the Republic, UDF Union for French Democracy.

Germany West 1998: Alliance '90/Greens, FDP Free Democratic Party, PDS Party of Democratic Socialism, SPD Social Democratic Party.

Germany East 1998: Alliance '90/Greens, FDP Free Democratic Party, PDS Party of Democratic Socialism, SPD Social Democratic Party.

Great Britain42 1997: Conservative Party, Labour Party, LDP Liberal Democrats, SNP Scottish National Party43

Hungary 1998: FIDESZ-MPP Alliance of Young Democrats — Hungarian Civic Party, FKgP Independent Smallholders' Party, MSzP Hungarian Socialist Party, SzDSz Alliance of Free Democrats.

Iceland 1999: FSF Progressive Party, Sj Independence Party.

Ireland 2002: Fianna Fail, Fine Gael, Green Party, LP Labour Party, PD Progressive Democrats, Sinn Fein41

Israel 1996: Ha'avoda (Labour), Likud, Mafdal, Meretz, Shas.

Netherlands 1998: CDA Christian Democratic Appeal, D'66 Democrats 66, GL Green Left, PvdA Labour Party, SP Socialist Party, VVD People's Party for Freedom and Democracy.

New Zealand 1996: LP Labour Party, Alliance, NP National Party, NZFP New Zealand First Party.

Norway 1997: DNA Labour Party, FrP Progress Party, H Conservative Party, KrF Christian Democratic Party, SP Centre Party, SV Socialist Left Party.

Poland 1997: PSL Polish People's Party, SLD Democratic Left Alliance, UP Union of Labour.

Portugal 2002: CDU Unified Democratic Coalition, PP Popular Party, PS Socialist Party, PSD Social Democratic Party.

Romania 1996: PDSR Party of Social Democracy in Romania, PUNR Party of Romanian National Unity, UDMR Hungarian Democratic Alliance.

Slovenia 1996: LDS Liberal Democracy, SD Social Democratic Party, SKD Slovene Christian Democrats, SLS People's Party.

Spain 1996: Convergence and Unity CiU, EAJ-PNV Basque Nationalist Party, PP Popular Party, PSOE Socialist Workers' Party, 1U United Left.

Sweden 1998: CP Centre Party, FP Liberal People's Party, KD Christian Democrats, MS Moderate Coalition Party (The Alliance Manifest), VP Left Party, SAP Social Democratic Labour Party.

Switzerland 1999: CVP/PDC Christian Democrats, FDP/PLR Freethinking Democrats, GLP/PVL Green Party, SPS/PSS Social Democrats, SVP/UDC Swiss People's Party.

Appendix 2. Operationalization of the control variables

Note that if no other source is mentioned, the below variables origin from or are based on CSES data.

Appendix 3. The distribution of voters' proximities to parties within countries and summary statistics for the original variable "Estimation gaps" between parties and manifesto positions

Legend: The boxplot shows the distribution of voters' proximities to parties within countries.44 Denmark, Sweden and the

42 Great Britain without Northern Ireland.

43 Only Scottish respondents evaluated the Scottish National Party.

44 A table with the mean and the minimum to maximum mean ranges of the original 'gaps' between persons' estimations and party positions can be found below.

"Don't Like", Like two dummy variables which are coded 'one' if the respondent has a strong positive (negative) affection towards the party. The variable, from Strongly": which the dummies are calculated, is an 11-point metric scale question. It ranges from zero for 'strongly disliking' to ten for 'strongly liking' the

party. Respondents were asked to place every one of the five most important parties from the current election on the scale. An extraordinary negative (positive) affection is represented by values of 'x' which are below (above) the range of the country specific mean minus (plus) one (country specific) standard deviation

"Education": measured with an eight-point scale from "no schooling" to "university undergraduate degree completed" (CSES)

"Age": a metric variable from 18 (eligibility age) onwards

"Female": A dummy variable coded one if the person is female

"Previous Vote share": The party's vote share in the last election

"Party Age": a metric variable representing the age of the party in the current year of election.

"Left-Right Family": a dummy variable which builds on the information of party families in the CSES. It is coded one if the party's party-family is one of the core left-right ones.

"Exgovernmental": a dummy variable which is coded one if the party was in the current election's preceding government. The information on this was collected from parline.

"Issue: Party Politics": represents the number of party politics issues within the five most important salient issues of the election campaign, mentioned by national political experts (CSES data). These concern either pre-election discussions about 'Party Alliances, Dynamics, Re-Alignment' (Belgium Flanders, Czech Rep., Hungary, Netherlands, Norway, Poland, Slovenia, Spain, Switzerland) or valence issues such as the evaluation of 'Party Performance' (Australia, Belgium Flanders, Canada, Germany, Great Britain), or 'Party Scandals' (e.g. Norway)

"Party System measures were taken from Dalton (2008). For Flanders I calculated the value, using Dalton's formula.

Polarization":

"Multiparty System": a dummy variable coded one if there are more than 2.5 parties within the party system.

"ENPP": the effective number of parliamentary parties, as calculated by Gallagher (2013)

"Age of Democracy": the origin of this is a variable called "system tenure" in the "Political Institutions dataset (Beck et al., 2001). It represents the age of the current political regime.

"young": a dummy variable with the value one, if the party system is one of the young Eastern European ones of the sample. East Germany is coded as zero,

here, since the party system was inherited to a large part from the West

Netherlands are the countries with the highest median proximity — voters tend to estimate parties more accurately than in other countries. All of these are Western European countries with a long democratic history and the Netherlands can be even seen as a prototype of a 'consociational democracy', characterized by very proportional elections and strong party ties with civil society (Lijphart, 1999). Poland sticks out with the lowest median voters' estimation proximity, followed by Romania, Slovenia, Hungary and Iceland. Apparently, four of these countries are young communist successor countries, which in the early years after their regime changes had particularly volatile party systems (Rudi, 2010a) and a strong "non-economic or cultural, new-politics dimension" (Evans and Whitefield, 1993: 157). The way how strongly respondents also think of these issues when they have to place parties on the left—right scale might vary. The outstandingly poor Polish median

proximity seems to be mostly due to the very big estimation gap of voters' estimations of the Polish Democratic Left Alliance (SLD), a communist successor party. In that case the party's left-right position might have been specifically hard to estimate. Also in Iceland in 1999, the low median proximity of voters' estimations to party positions is mainly due to voters' misplacements of one party, the Progressive party. A possible reason might be that voters missed the centrist Progressive Party's position change to the left because of other, more spectacular position takings on the left and right taking place at the same time, due to the formation of a new right and a new left party, the 'Liberal Party', and the 'Left-greens' and a new party union, the 'Left Alliance' (see Hardarson and Kristinsson, 2001).

Mean estimation gaps per country, and minimum + maximum mean gaps of parties within countries.

Election study Party with min and max difference (between voters' estimations and party position) Country ranges s.e. Mean of all parties per country

Australia 1996 ALP Labour Party 1.92 (Min value) 0.11 2.06

NPA National Party 2.17 (Max value)

Belgium Flanders 1999 VLD Flemish Liberals 1.91 (Min value) 0.58 2.29

VB Flemish Bloc 3.43 (Max value)

Bulgaria 2001 DPS Movement for Rights and Freedom 1.55 (Min value) 0.88 2.43

ODS United Democratic Party 3.31 (Max value)

Canada 1997 LP Liberal Party of Canada 1.66 (Min value) 0.86 2.48

Reform Party 3.64 (Max value)

Czech Republic KDU-CSL Christian Democratic Union 1.57 (Min value) 0.67 2.33

ODS: Civic Democratic Party 3.27 (Max value)

Denmark 1998 SF Socialist People's Party 1.68 (Min value) 0.29 2.13

Centre Democrats (CD) 2.13 (Max value)

Finland 2003 The Left Alliance 1.38 (Min value) 0.45 2.23

The Greens of Finland 2.99 (Max value)

France 2002 Union for French Democracy 1.69 (Min value) 0.33 1.99

FN National Front 2.63 (Max value)

Germany 1998 FDP Free Democratic Party 1.53 (Min value) 0.43 1.81

PDS Left Party 2.56 (Max value)

Germany East 1998 FDP Free Democratic Party 1.52 (Min value) 0.12 1.65

Alliance '90/Greens 1.79 (Max value)

(continued )

Election study Party with min and max difference (between voters' estimations and party position) Country ranges s.e. Mean of all parties per country

Great Britain 1997 LDP Liberal Democrats 1.23 (Min value) 0.40 1.89

Conservative Party 2.33 (Max value)

Hungary 1998 FIDESZ MPP: Fidesz - Hungarian Civic Party 1.82 (Min value) 0.43 2.46

MSzP Hungarian Socialist Party 3.00 (Max value)

Iceland 1999 F Progressive Party 1.56 (Min value) 0.84 2.42

Sj Independence Party 3.27 (Max value)

Ireland 2002 LP Labour Party 1.54 (Min value) 0.42 2.05

Fianna Fail 2.54 (Max value)

Israel 1996 LIKUD 1.49 (Min value) 0.41 2.05

AVODA (Labour) 2.65 (Max value)

Netherlands 1998 GL Green Left 1.18 (Min value) 0.13 1.39

VVD People's Party for Freedom and Democracy 1.55 (Max value)

New Zealand 1996 NZFP New Zealand First 1.44 (Min value) 0.20 1.67

NP National Party 2.00 (Max value)

Norway 1997 SV Socialist Left Party 1.12 (Min value) 0.37 1.54

DNA Labour Party 2.21 (Max value)

Poland 1997 UP Union of Labour 2.09 (Min value) 0.82 3.25

SLD Democratic Left Alliance 3.88 (Max value)

Portugal 2002 PS Socialist Party 1.21 (Min value) 0.45 1.68

PP Popular Party 2.42 (Max value)

Romania 1996 Party of National Unity 2.71 (Min value) 0.11 2.82

UDMR Hungarian Democratic Party 2.97 (Max value)

Slovenia 1996 LDS Liberal Democracy 2.18 (Min value) 0.53 2.7

Christian Democrats 3.52 (Max value)

Spain 1996 Convergence and Unity 1.60 (Min value) 0.30 2.01

PSOE Socialist Worker 2.99 (Max value)

Sweden 1998 CP Centre Party 0.92 (Min value) 0.33 1.44

KdS Christian Democratic Party 2.03 (Max value)

Switzerland 1999 CVP-PDC Christian Democrats 1.26 (Min value) 0.24 1.6

Swiss People's Party 1.97 (Max value)

All Parties Minimum: 0.92 (Min value) 0.34 1.52

Mean: 2.00

Maximum: 3.88 (Max value) 0.63 2.58

Appendix 4. Variable summary statistics

Variable N Mean SD Min Max

Estimation proximity 150642 1.39 0.51 -0.00 2.34

Don't like 194351 0.18 0.39 0.00 1.00

Like strongly 194351 0.27 0.44 0.00 1.00

Ideological distance 165183 2.29 1.74 0.00 9.42

Education 192324 4.89 1.72 1.00 8.00

Age 193651 46.07 16.84 18.00 101.00

Female 194208 0.52 0.50 0.00 1.00

Party's change 113 0.33 0.27 0.00 1.32

Party's ecochange 113 0.40 0.39 0.01 2.40

Previous vote share 113 16.26 12.68 0.00 52.24

Party age 113 49.04 42.26 3.00 156.00

Left-right family 113 0.73 0.44 0.00 1.00

Party divergence 113 0.82 0.33 0.21 1.83

In Government t-1 113 0.38 0.49 0.00 1.00

Age of Democracy 25 47.20 26.73 4.00 73.00

Party system polar. 25 3.40 1.03 1.83 5.44

Overall/total PS change 25 0.35 0.16 0.14 0.76

Overall/total eco change 25 0.40 0.19 0.05 0.78

Young 25 0.14 0.35 0.00 1.00

ENPP 25 3.81 1.23 2.16 6.61

Issue: party politics 25 0.80 0.65 0.00 2.00

Multiparty system 25 0.80 0.41 0.00 1.00

Appendix 5. Evaluation of Parties' Left-Right Shifts and Voters' Mean Proximity values

Country

LR-Shift Prox. Evaluation

Mean: Mean: Strong Shift (>=0.61) -

0.34; 1.39; around average/good

Std: Std: 0.5 estimation (Proximity 0.27 >=1.14)

Low to Medium Shift, Rather strong to strong (<0.61), bad estimation shift (>=0.61) bad (Proximity <1.14) estimation (Proximity <1.14)

Low to medium shift (<0.61), average to good estimation (Proximity>=1.14)

Mean LR-Shifts

Mean of parties' LR-Shifts: 0.34

Australia LPA 0.66 1.40 X 0.58

(1996) ALP 0.50 1.32 X

NPA 0.59 1.48 X

Belgium VLD 0.39 1.40 X 0.51

Flanders CVP 1.00 1.32 X

(1999) SP 0.23 1.37 X

VB 0.49 0.93 X

Agalev 0.44 1.34 X

Bulgaria ODS 0.06 0.95 X 0.17

(2001) DPS 0.27 1.5 X

Canada LP 0.31 1.48 X 0.35

(1997) R 0.02 1.06 X

PC 0.43 1.38 X

BQ 0.69 0.98 X

Czech ODS 0.01 0.94 x 0.28

Republic CSSD 0.27 1.45 X

(1996) KDU-CSL 0.07 1.5 X

Association 1.01 1.08 X

for the Rep.

ODA 0.05 1.26 X

Denmark SD 0.60 1.39 X 0.33

(1998) V 0.05 1.57 X

KF 0.44 1.55 X

SF 0.37 1.62 X

CD 0.21 1.33 X

Finland KESK 0.47 1.23 X 0.34

(2003) SDP 0.43 1.37 X

KOK 0.53 1.22 X

Left Alliance 0.07 1.59 X

Green 0.09 1.04 X

League

KD 0.59 1.32 X

SFP/RKP 0.22 1.24 X

France RPR/UMP 0.87 1.29 X 0.39

(2002) FN 0.14 1.18 X

PS 0.10 1.40 X

UDF 0.79 1.47 X

LV 0.08 1.33 X

PCF 0.37 1.42 X

Germany SPD 0.81 1.52 X 0.29

East Alliance/ 0.05 1.45 X

(1998) Greens

FDP 0.24 1.53 X

PDS 0.06 1.45 X

Germany SPD 0.81 1.54 X 0.29

West Alliance/ 0.05 1.51 X

(1998) Greens

FDP 0.24 1.52 X

PDS 0.06 1.24 X

Great Britain Labour 0.56 1.35 X 0.47

(1997) Conservative 1.07 1.24 X

LDP 0.20 1.63 X

SNP 0.02 1.41 X

Hungary MSzP 0.69 1.23 X 0.68

(1998) FIDESZ-MPP 0.68 1.30 X

FKgP 0.67 1.15 X

SzDSz 0.67 1.06 X

Iceland SJ 0.19 0.96 X 0.76

(1999) FSF 1.32 1.49 X

Ireland Fianna Fail 0.11 1.46 X 0.21

(2002) Fine Gael 0.46 1.47 X

Labour 0.10 1.53 X

Sinn Fein 0.13 1.46 X

(continued )

Country Party LR-Shift Prox. Evaluation Mean LR-

Shifts

Mean: Mean: Strong Shift (>=0.61) - Low to Medium Shift, Rather strong to strong Low to medium shift Mean of

0.34; 1.39; around average/good (<0.61), bad estimation shift (>=0.61) bad (<0.61), average to good parties' LR-

Std: Std: 0.5 estimation (Proximity (Proximity <1.14) estimation (Proximity estimation Shifts: 0.34

0.27 >=1.14) <1.14) (Proximity>=1.14)

Green 0.24 1.25 X

Israel (1999) Ha'avoda 0.44 1.16 X 0.14

Likud 0.08 1.55 X

Shas 0.0 1.29 X

Mafdal 0.18 1.45 X

Meretz 0.00 1.24 X

Netherlands PvdA 0.14 1.58 X 0.25

(1998) VVD 0.18 1.48 X

CDA 0.09 1.62 X

D' 66 0.29 1.54 X

GL 0.59 1.71 X

SP 0.18 1.52 X

New NP 0.46 1.35 X 0.31

Zealand LP 0.16 1.49 X

(1996) NZFP 0.17 1.57 X

Alliance 0.46 1.48 X

Norway DNA 0.19 1.31 X 0.31

(1997) FrP 0.05 1.52 X

H 0.56 1.53 X

KrF 0.35 1.62 X

SP 0.35 1.58 X

SV 0.32 1.66 X

Poland SLD 0.54 1.64 X 0.55

(1997) PSL 0.69 0.87 X

UP 0.42 1.35

Portugal PSD 0.03 1.50 X 0.40

(2002) PS 0.57 1.70 X

PP 0.11 1.21 X

CDU 0.87 1.50 X

Romania PDSR 0.45 1.12 X 0.18

(1996) UDMR 0.04 1.08 X

PUNR 0.14 1.15 X

Spain (1996) PP 0.29 1.38 X 0.22

PSOE 0.15 1.50 X

IU 0.24 1.56 X

CiU 0.43 1.54 X

EAJ-PNV 0.06 1.47 X

Slovenia LDS 0.22 1.31 X 0.16

(1996)

SLS 0.26 1.26 X

SD 0.14 1.13 X

SKD 0.01 0.95 X

Sweden SAP 0.24 1.55 X 0.28

(1998) MS 0.02 1.53 X

VP 0.55 1.51 X

KD 0.29 1.33 X

CP 0.16 1.78 X

FP 0.45 1.57 X

Switzerland SVP/UDC 0.78 1.35 X 0.37

(1999) SPS/PSS 0.03 1.46 X

FDP/PLR 0.81 1.54 X

CVP/PDC 0.47 1.67 X

GLP/PVL 0.11 1.51 X

The list of party abbreviations can be found in Appendix 1.

Appendix 6. Model 1 with separate introduction of levels 1 -3.

Fixed part

1-Level Model

2-Level Model

3-Level Model

ß coefficients, unstandardized

Fixed effects (with robust std. errors)

Fixed effects (with non- robust std. errors)#

Intercept 3 1.514*** 1.624*** 1.509

Don't Like -0.167*** -0.167*** -0.167

Like Strongly -0.105*** -0.099*** -0.099

Ideological Distance -0.032*** -0.037*** -0.037

Education 0.016*** 0.016*** 0.016

Age -0.000* -0.000* -0.000

Female -0.037*** -0.037*** -0.037

Party's Change -0.001 0.033

Previous Vote Share -0.003* -0.002

Party Age 0.001** 0.001

Left-Right Family 0.020 0.015

Party Divergence -0.139** -0.140

In Government t-1 -0.052 -0.046

Issue: Party Politics 0.092

Issue: Leader Politics 0.069

Overall/Total PS Change -0.544

Party System Polar. 0.027

Multiparty System 0.119

ENPP -0.044

Age of Democracy 0.003

Random effects Variance Components (V.C.) and degrees of freedom (d.f.) V.C. d.f. V.C d.f. V.C. d.f.

Intercept 1 r0 0.023*** 85 0.028*** 79 0.027*** 79

Ideological Distance Slope, r1 0.001*** 109 0.001*** 85

Level 3, Intercept 1/2 u00 0.116*** 24 0.010*** 24 0.000 17

Ideological Distance, Intercept 2 u20 - 0.000* 24

Log likelihood -89470470000 -88782150000 -88766170000

N (individual/party dyads) 1150642 150642 150642

N parties 113 113 113

N systems 25 25 25

***p < .001, two sided, **p < .01, two sided, *p < .05 two-sided, +p < .05 one-sided.

# Since the number of level 3 units was too small, the fixed effects were calculated with non-robust standard errors.The differences to the estimation of fixed effects with robust standard errors were negligible. All effects fixed except for 'Ideological Distance' both on party and system level. Education and Age are group mean centred. The random effectspart displayed, here, is calculated with the unstandardized analysis. The chi-square statistics reported above are based on only 110 of 113 level 2 units, which had sufficient data for computation. Fixed effects and variance components are based on all the data.

Appendix 7. Respondents' estimation proximity to left-right party positions with left-right position change and an Interaction effect of young party systems * change

Fixed part Beta coefficients, (unstandardized)

Intercept 3 1.673***

Don't Like -0.167***

Like Strongly -0.099***

Ideological Distance -0.037***

Education* 0.016***

Age* -0.000***

Female -0.037***

Party's Change 0.032

PrVote Share -0.003+

Party Age 0.001*

Left-Right Family 0.017

Party Divergence -0.120**

In Government t-1 -0.037

Issue: Party Politics 0.079**

Issue: Leader Politics 0.071*

Overall/Total L-R Change -0.543***

Multiparty System 0.086

ENPP -0.030

Age of Democracy 0.002

Young -0.133

Young*Overall Change 0.091

Random effects

Variance Comp.

Intercept 1, r0 Ideological distance

slope, r2 Level 3,

Intercept1/2 u0o Ideological distance,

Intercept 2, u20 Log likelihood N (individual/party

dyads)

N parties N systems

0.026 0.001

4602.649 1227.106

20.097 40.314

<0.001 <0.001

0.215 0.020

-88763350000.00 194351

113 25

***p < .001, two sided, **p < .01, two sided, *p < .05 two-sided, +p < .05 one-sided. Since the number of level 3 units was too small, the fixed effects were calculated with non-robust standard errors. The differences to the estimation of fixed effects with robust standard errors were negligible. All effects fixed except for 'Ideological Distance', which is random both on party and system level. Education and Age are group mean centred. The random effects part displayed, here, is calculated with the unstandardized analysis. The chi-square statistics reported above are based on only 110 of 113 level 2 units, which had sufficient data for computation. Fixed effects and variance components are based on all the data.

References

Adams, J., 2001. A theory of spatial competition with biased voters: party policies viewed temporally and comparatively. Br. J. Polit. Sci. 31 (01), 121—158.

Adams, J., Clark, M., Ezrow, L., Glasgow, G., 2006. Are niche parties fundamentally different from mainstream parties? the causes and the electoral consequences of Western european parties' policy shifts, 1976—1998. Am. J. Polit. Sci. 50 (3), 513—529.

Adams, J., Ezrow, L., Leiter, D., 2012. Partisan sorting and niche parties in europe. West Eur. Polit. 35 (6), 1272—1294.

Adams, J., Somer-Topcu, Z., 2009. Policy adjustment by parties in response to rival parties' policy shifts: spatial theory and the dynamics of party competition in twenty-five post-war democracies. Br. J. Polit. Sci. 39 (4), 825—846.

Adams, J., Ezrow, L., Somer-Topcu, Z., 2011. Is anybody listening? evidence that voters do not respond to european parties' policy statements during elections. Am. J. Polit. Sci. 55 (2), 370—382.

Adams, J., Ezrow, L., Somer-Topcu, Z., 2014. Do voters respond to party manifestos or to a wider information environment? an analysis of mass-elite linkages on european integration. Am. J. Polit. Sci. 58 (4), 967—978.

Bennett, L.L., Bennett, S.E., 1989. Enduring gender differences in political interest: the impact of socialization and political dispositions. Am. Polit. Res. 17 (1), 105—122.

Benoit, K., Laver, M., 2007. Estimating party policy positions: comparing expert surveys and hand-coded content analysis. Elect. Stud. 26, 90—107.

Benoit, K., et al., 2009. Treating words as data with error: uncertainty in text statements of policy positions'. Am. J. Polit. Sci. 53 (2), 495—513.

Best, R., 2013. Using the manifesto estimates to correct systematic 'centring' error in expert and electoral positioning of parties. In: Volkens, A., et al. (Eds.), Mapping Policy Preferences from Texts. Statistical Solutions for Manifesto Analysts. Oxford University Press, Oxford, pp. 33—48.

Blais, A., Bodet, M.A., 2006. Does proportional representation Foster closer congruence between citizens and policy makers? Comp. Polit. Stud. 39 (10), 1243—1262.

Bryk, A.S., Raudenbush, S.W., 1992. Hierarchical Linear Models: Applications and Data Analysis Methods. Sage Publications, London.

Budge, I., 1987. The internal analysis of election programmes. In: Budge, I., et al. (Eds.), Ideology, Strategy and Party Change. Cambridge University Press, Cambridge, pp. 15—38.

Budge, I., et al. (Eds.), 2001. Mapping Policy Preferences. Estimates for Parties, Electors, and Governments 1945-1998. Oxford University Press, Oxford, p. 274.

Budge, I., Klingemann, H.-D., 2001. Finally! comparative over-time mapping of party policy movement. In: Budge, I., Klingemann, H.-D., Volkens, A., Bara, J., Tanenbaum, E. (Eds.), Mapping Policy Preferences. Estimates for Parties, Electors, and Governments 1945-1998. Oxford University Press, Oxford, pp. 19—50.

Budge, I., McDonald, M.D., 2007. Election and party system effects on policy representation: bringing time into a comparative perspective. Elect. Stud. 26 (1), 168—179. http://dx.doi.org/10.1016/j.electstud.2006.02.001.

Castles, F., Mair, P., 1984. Left-right political scales: some expert judgements. Eur. J. Polit. Res. 12 (1), 73—88.

Clarke, H., Sanders, D., Stewart, M., Whiteley, P., 2011. Valence politics and electoral choice in britain, 2010. J. Elect. Public Opin. Part. 21 (2), 237—253.

Dahlberg, S., 2013. Does context matter - the impact of electoral systems, political parties and the individual characteristics on voters' perceptions of party positions. Elect. Stud. 32 (4), 670—683.

Dalton, R.J., 2008. The quantity and quality of party Systems: Party system polarization, its measurement, and its consequences. Comp. Polit. Stud. 41, 899—920.

Dinas, Elias, Gemenis, K., 2010. Measuring parties' ideological positions with manifesto data: a critical evaluation of the competing methods. Party Polit. 16, 427—450.

Downs, A., 1957. An Economic Theory of Democracy. Harper and Row, New York.

Drummond, A., 2010. Assimilation, contrast and voter projections of parties in left-right space: does the electoral system matter? Party Polit. 17 (6), 711—743.

Evans, G., Whitefield, S., 1993. Identifying the bases of party competition in eastern Europe. Br. J. Polit. Sci. 23, 521—548.

Franklin, M.N., Lyons, P., Marsh, M., 2004. Generational basis of turnout decline in established democracies. Acta Polit. 39,115—151.

Franzmann, S.T., 2013. Towards a Real Comparison of Left-right Indices: A Comment to Jahn. Party Politics. http://dx.doi.org/10.1177/1354068813499865.

Franzmann, S.T., Kaiser, A., 2006. Locating political parties in policy space: a rean-alysis of party manifesto data. Party Polit. 163 (12), 163—188.

Gabel, M., Huber, J.D., 2000. Putting parties in their place. Am. J. Polit. Sci. 44, 94—103.

Gallego, A., 2009. Where else does turnout decline come from? education, age, generation and period effects in three european countries. Scand. Polit. Stud. 32 (1), 23—44. http://dx.doi.org/10.1111/).1467-9477.2008.00212.x.

Gastil, J., 2000. By Popular Demand: Revitalizing Representative Democracy through Deliberative Elections. University of California Press, Berkeley, p. 276.

Gidengil, E., Blais, A., Nevitte, N., Nadeau, R., 2002. Priming and campaign context: evidence from recent Canadian elections. In: Farrell, D.M., Schmitt-Beck, R. (Eds.), Do Political Campaigns Matter? Campaign Effects in Elections and Referendums. Routledge, London, New York, pp. 76—91.

Golder, M., Stramski, J., 2010. Ideological congruence and electoral institutions. Am. J. Polit. Sci. 54 (1), 90—106.

Gordon, S.B., Segura, G.M., 1997. Cross-national variation in the political sophistication of individuals: capability or choice? J. Polit. 59 (1), 126-146.

Granberg, D., Holmberg, S., 1988. The Political System Matters. Social Psychology and Voting Behavior in Sweden and the United States. Cambridge University Press, Cambridge.

Hardarson, O. Th, Kristinsson, G.H., 2001. The 1999 parliamentary election in Iceland. Elect. Stud. 20, 325-331.

Hayes, D., 2008. Has television personalized voting behavior? Polit. Behav. 31, 231-260.

Holländer, M., 2003. Konfliktlinien und Konfigurationen der Parteiensysteme in Ostmitteleuropa 1988-2002. Dissertation Tübingen 2003. Books on Demand, Norderstedt, p. 560.

Hopmann, D.N., van Aelst, P., Legnante, G., 2011. Political balance in the news: a review of concepts, operationalizations and key findings. Journalism 13 (2), 240-257.

Huber, J.D., Inglehart, R., 1995. Expert interpretations of party space and party locations in 42 societies. Party Polit. 1, 73-111.

Huber, J.D., Powell, G.B., 1994. Congruence between citizens and policymakers in two visions of Liberal democracy. World Pol. 46 (03), 291-326, 10.2307/ 2950684.

Inglehart, R., Norris, P., 2000. The developmental theory of the gender Gap: Women's and Men's voting behavior in global perspective. Int. Polit. Sci. Rev. 21 (4), 441-463. http://dx.doi.org/10.1177/0192512100214007.

Jahn, D., 2011. Conceptualizing left and right in comparative politics: towards a deductive approach. Party Polit. 17, 745-765.

Kaase, M., 1994. Is there personalization in politics? candidates and voting behavior in Germany. Int. Polit. Sci. Rev. 15, 211-230.

Kirchheimer, O., 1966. The transformation of west european party systems. In: La Palombara, J., Weiner, M. (Eds.), Political Parties and Political Development. Princeton University Press, Princeton, pp. 177-200.

Kitschelt, H., 1995. Formation of party cleavages in Post-Communist democracies. Party Polit. 1 (4), 447-472.

Klingemann, H.-D., 1995. Party positions and voter orientations. In: Klingemann, H.-D., Fuchs, D. (Eds.), Citizens and the State. Oxford University Press, Oxford, pp. 183-206.

Klingemann, H.-D., et al. (Eds.), 2006. Mapping Policy Preferences II: Estimates for Parties, Electors, and Governments in Eastern Europe, European Union, and OECD 1990-2003. Oxford University Press, Oxford, New York, p. 267.

Lacewell, O., Werner, A., 2013. Coder training: key to enhancing reliability and validity. In: Volkens, et al. (Eds.), Mapping Policy Preferences from Texts. Statistical Solutions for Manifesto Analysts. Oxford University Press, Oxford, pp. 169-194.

Laver, M., Budge, I., 1992. In: first ed.Macmillan (Ed.), Party Policy and Government Coalitions. XXII, Basingstoke, p. 448.

Lijphart, A., 1999. In: Patterns of Democracy: Government Forms and Performance in Thirty-six Countries, vol. XIV. Yale University Press, New Haven, p. 351.

Mainwaring, S., Torcal, M., 2006. Party system institutionalization and party system theory after the third wave of Democratization. In: Katz, R.S., Crotty, W. (Eds.), Handbook of Party Politics. Sage, London, pp. 204-22 .

Markowski, R., 1997. Political parties and ideological spaces in east Central Europe. Communist Post Communist Stud. 30 (3), 221-254. http://dx.doi.org/10.1016/ S0967-067X(97)00006-8.

Marks, G., Hooghe, L., Nelson, M., Edwards, E., 2006. Party competition and european integration in the east and west: different structure, same causality. Comp. Polit. Stud. 39 (2), 155-175.

Mazzoleni, G., 1987. Media logic and party logic in campaign coverage: the Italian general election of 1983. Eur. J. Commun. 2, 81-103.

McDonald, M.D., Mendes, S.M., Budge, I., 2004. What are elections for? Conferring the median mandate. Brit. J. Polit. Sci. 34 (01), 1. http://dx.doi.org/10.1017/ S0007123403000322.

Merrill, S., Grofman, B., 1999. A Unified Theory of Voting: Directional and Proximity Spatial Models. Cambridge University Press, Cambridge, UK, New York, p. 213.

Merrill, S., Grofman, B., Adams, J., 2001. Assimilation and contrast effects in voter projections of party locations: evidence from Norway, France, and the USA. Eur. J. Polit. Res. 40 (2), 199-221. http://dx.doi.org/10.1111/1475-6765.00594.

Meyer, T.M., 2013. Constraints on Party Policy Change. ECPR Press, Essex.

Milner, H., 2002. Civic Literacy: How Informed Citizens Make Democracy Work, vol. IX. University Press of New England, Hanover, NH, p. 293.

Neuman, R.W., 1981. Differentiation and integration: two dimensions of political thinking. Am. J. Sociol. 86 (6), 1236-1268.

Neuman, R.W., 1986. The Paradox of Mass Politics: Knowledge and Opinion in the American Electorate. Harvard University Press, Cambridge, Mass (etc).

Norris, P., et al., 1999. On Message: Communicating the Campaign. Sage, London, Thousand Oaks, Calif, p. 218.

Powell, G.B., Vanberg, G.S., 2000. Election Laws, disproportionality and median correspondence: implications for two visions of democracy. Br. J. Polit. Sci. 30 (3), 383-411.

Roberts, A., 2010. The Quality of Democracy in Eastern Europe. Public Preferences and Policy Reforms. Cambridge University Press, Cambridge, p. 252.

Rose, R., Munro, N. (Eds.), 2009. Parties and Elections in New European Democracies. ECPR Press, Colchester, p. 275.

Rovny, J., Edwards, E.E., 2012. Struggle over Dimensionality: Party competition in western and eastern europe. East Eur. Polit. Soc. 26 (1), 56-74.

Rudi, T., 2010a. Wahlentscheidungen in postsozialistischen Demokratien in Mittel-und Osteuropa: Eine vergleichende Untersuchung, first ed. Nomos, Baden-

Baden, p. 310.

Rudi, T., 2010b. Die Links-Rechts-Dimension in Mittel- und Osteuropa: „Super-Issue" oder bedeutungslos? In: Faas, T., Arzheimer, K., Roßteutscher, S. (Eds.), Information — Wahrnehmung — Emotion. VS Verlag für Sozialwissenschaften, pp. 169—189.

Sartori, G., 1976. Parties and Party Systems: a Framework for Analysis. Cambridge University Press, Cambridge.

Schoenbach, K., 1996. The "Americanization" of german election campaigns: any impact on the voters? In: Swanson, D.L., Mancini, P. (Eds.), Politics, Media, and Modern Democracy. An International Study of Innovations in Electoral Campaigning and Their Consequences. Praeger Publishers, Westport, pp. 91—105.

Somer-Topcu, Z., 2014. Everything to Everyone: The Electoral Consequences of the Broad-appeal Strategy in Europe. American Journal of Political Science, p. 14. Online Publication Version.

Tavits, M., 2008. On the linkage between electoral volatility and Party System instability in Central and Eastern Europe. Eur. J. Polit. Sci. 47 (5), 537—555.

Tavits, M., Letki, N., 2009. When left is Right: Party ideology and policy in Post-Communist europe. Am. Polit. Sci. Rev. 103 (04), 555. http://dx.doi.org/ 10.1017/S0003055409990220.

Toka, G., 1997. Political parities and democratic consolidation in east Central Europe. In: Diamond, L.J., et al. (Eds.), Consolidating the Third Wave Democracies: Themes and Perspectives, Baltimore, pp. 93—134.

Van Aelst, P., et al., 2008. Politicians in the news: media or party logic?: media attention and electoral Success in the belgian election campaign of 2003. Eur. J. Commun. 23, 193—210.

Warwick, P.V., 1996. Coalition government membership in West european parliamentary democracies. Br. J. Polit. Sci. 26 (04), 471 —499.

Warwick, P.V., 2010. Bilateralism or the median mandate?: an examination of rival perspectives on democratic governance. Eur. J. Polit. Res. 49 (1), 1—24.

Whiteley, P., Stewart, Marianne C., Sanders, David, Clarke, H.D., 2005. The issue agenda and voting in 2005. Parliam. Aff. 58 (4), 802—817, 10.1093/pa/gsi062.

Williams, Laron K., 2015. It's all relative: spatial positioning of parties and ideological shifts. Eur. J. Polit. Res. 54,141-159.

Data sources

Beck, Thorsten, et al., 2001. New Tools in Comparative Political Economy. The Database of Political Institutions, 15:1,165—176 (September), World Bank Economic Review. The 2012 Version, (updated January 2014), report number WPS2283 can be found at. http://econ.worldbank.org/WBSITE/EXTERNAL/ EXTDEC/EXTRESEARCH/0,contentMDK:20649465~pagePK:64214825~piPK: 64214943~theSitePK:469382,00.html (July 7, 2015).

Freedom House, 2014. Freedom in the World. Country Ratings and Status by Region. FIW 1973-2015 (EXCEL). http://freedomhouse.org/report-types/freedom-world (July 7, 2015).

Franzmann, S., Kaiser, A., 2006. Locating political parties in policy space: a rean-alysis of party manifesto data. Party Polit. 163 (12), 163—188. The data can be downloaded from. http://www.uni-potsdam.de/db/ls_regierungssystem_brd/ index.php?article_id=498&clang=0 (July 7, 2015).

Gallagher, Michael, 2013. Election Indices Dataset. http://www.tcd.ie/Political_ Science/staff/michael_gallagher/ElSystems/index.php (July 7, 2015).

Klingemann, Hans-Dieter, et al., 2006. Mapping Policy Preferences II. Estimates for Parties, Electors, and Governments in Eastern Europe, the European Union and the OECD, 1990-2003. Oxford University Press, Oxford (Data on CD).

The Comparative Study of Electoral Systems, August 4, 2003. CSES MODULE 1 FULL RELEASE [dataset]. University of Michigan, Center for Political Studies [producer and distributor], Ann Arbor, MI. http://dx.doi.org/10.7804/cses.module1.2003-08-04 version. www.cses.org.

The Comparative Study of Electoral Systems, June 27, 2007. CSES MODULE 2 FULL RELEASE [dataset]. http://dx.doi.org/10.7804/cses.module2.2007-06-27 version. www.cses.org.