Intra-Party Heterogeneity in Policy Preferences and Its Effect on Issue Salience: Evidence from the Comparative Candidates Survey

Transcription

1 Intra-Party Heterogeneity in Policy Preferences and Its Effect on Issue Salience: Evidence from the Comparative Candidates Survey Nils D. Steiner Department of Political Science Johannes Gutenberg University Mainz Matthias Mader Department of Political Science University of Mannheim Paper prepared for presentation at the 24 th World Congress of Political Science by the International Political Science Association (IPSA) in Poznań, July 2016 Abstract: Quantitative research on the positions of political parties and party competition regularly invokes the assumption that parties are unitary actors with homogenous policy preferences. Drawing on Comparative Candidates Survey (CCS) data from 28 elections in 21 developed democracies, we show that candidates often hold quite heterogeneous issue positions and that the extent of this heterogeneity varies significantly across parties and, most interestingly, even within parties across different issue dimensions. In an effort to explore the implications of such intra-party heterogeneity for party strategy and competition, we argue that intra-party heterogeneity and issue salience go together, because parties will emphasize those issues on which intra-party heterogeneity is low. Empirically, we relate our measures of intra-party heterogeneity from the CSS to data on issue salience from the Chapel Hill Expert Survey and the Manifesto Project. Across different issue dimensions and the two salience measures, we consistently find that parties attach lower salience to issues over which they are internally divided.

2 1. Introduction Quantitative research on the positions of political parties tends to be firmly rooted in the assumption that parties are unitary actors. Making this assumption might be required for some applications, but it limits our understanding of party positioning and voter-party linkages in some important ways. It leads, for example, to the suggestion that parties occupy precisely defined positions in the policy space, when in reality parties are unlikely to be completely united in their preferences. Conceptualizing parties as organizations that consist of actors with possibly divergent preferences allows us to think of parties as having distributions in the policy space with different variances not only across different parties, but also across different issue dimensions. Politicians from one party could occupy relatively homogenous positions on an economic conflict dimension and more heterogeneous positions on a socio-cultural dimension; other parties might be united in their stance towards multiculturalism and European integration, but divided over the regulation of markets, the welfare state, and redistribution. The extent to which parties are unified in their preferences along different dimensions is likely to influence phenomena such as party strategy and competition, coalition building, and voting behavior. Such potential insights are lost to the extent that we treat parties as unitary actors with single positions in the policy space. Comparative research on the repercussions of intra-party policy preference heterogeneity has been hindered by the difficulty of obtaining valid cross-national data on such heterogeneity. 1 In this contribution, we show that party elite survey data are a fruitful, but hitherto largely neglected source for studying intra-party preference heterogeneity and its effects. We draw on the Comparative Candidates Survey (CCS), a cross-national research endeavor which consists of numerous postelection polls of candidates to national legislatures (CCS 2016). The main research interest behind the CCS lies in the issue of the individualization of electoral campaigns (Zittel 2015), but the CSS also collects data on candidates attitudes toward specific issues. We utilize these items to measure the heterogeneity of issue positions within parties across 28 parliamentary elections from 21 developed democracies. In short, we argue and demonstrate that candidate survey data give us direct, pertinent and easy-to-process empirical information on intra-party policy preference heterogeneity, which may be of great value for further illuminating party behavior. 1 We are clearly not the first to make this observation. In a recent review on research on parties and legislators, Saalfeld and Strøm (2014: 392), for example, state that parties still tend to be modeled as unitary actors despite frequent references to the need to relax this assumption. The black box of intra-party heterogeneity, organization, and decision-making [ ] has rarely been opened up. One reason is that measurement is difficult. We thus have to recognize the considerable obstacles caused by the fact that many important variables (such as sincere policy preferences) are non-observable and that attempts to capture them (e.g. recorded plenary votes) suffer from selection bias. There are notable contributions (discussed below) which try to tackle this issue, e.g., on intra-party dissent over European integration using expert ratings (de Vries/van de Wardt 2011; Hellström/Blomgren 2016; Hobolt/de Vries 2015; Netjes/Binnema 2007; Spoon 2012; Steenbergen/Scott 2004, van de Wardt et al. 2014). 1

3 We demonstrate this value by studying the link between intra-party heterogeneity and issue salience. Theories of party competition such as saliency theory (Budge and Farlie 1983) or the related concept of issue ownership (Petrocik 1996) highlight that parties differ over which issues are emphasized by them. For a number of theoretical reasons parties might place more emphasis on issues on which their elites agree. From a strategic-rationalist perspective parties might be electorally incentivized to emphasize internally uncontroversial issues, as appearing more united might be rewarded by voters. The attempt to emphasize issues (or issue dimensions) over which party professionals strongly disagree might moreover jeopardize the parties very survival not only for electoral reasons, but also because it risks mass resignations of professionals and, ultimately, party splits. An even simpler argument is that party leaders might find it easier to agree on a collective party position if they, as individuals, hold the same position on the issue in the first place. Despite its plausible effect on issue salience, the role of intra-party heterogeneity has been somewhat neglected in previous research on the topic. We suspect that this is in part due to the paucity of crossnational data on intra-party preference heterogeneity. While a number of empirical studies do explore the effect of intra-party dissent regarding European integration on the salience of European integration (de Vries/van de Wardt 2011; Hellström/Blomgren 2016; Hobolt/de Vries 2015; Netjes/Binnema 2007; Spoon 2012; Steenbergen/Scott 2004; van de Wardt et al. 2014), we are unaware of studies which link intra-party preference heterogeneity and issue salience in any other issue area. We close this research gap by combining the CCS data on intra-party preference heterogeneity with data on issue salience as measured by expert assessments in the Chapel Hill Expert Survey 2010 (Bakker et al. 2015) and by content analysis of party programs in the Manifesto Project (Volkens et al. 2015). The results strongly suggest that parties place less emphasis on issues over which their candidates hold heterogeneous preferences. While these findings are substantially important in itself, they also illustrate the value of giving up the unitary actor assumption and of considering the role of intra-party preference heterogeneity for party behavior. In this context, our methodological contribution is to show that survey data of party elites are a fruitful, but hitherto neglected source for studying such heterogeneity, its determinants, and implications. The remainder of this paper is structured as follows. In the following section, we describe our concept of intra-party preference heterogeneity and contrast it with related concepts. We discuss previous approaches to the measurement of intra-party preference heterogeneity and how such measures have been used in the past to study the implications of intra-party heterogeneity for issue salience.we go on to propose our main hypotheses on the implications of intra-party preference heterogeneity for issue salience. Given the novelty of our approach to measuring intra-party heterogeneity, the third section describes in detail how the CCS data can be used to derive this measure. We present evidence that candidates often hold quite heterogeneous issue positions and that the extent of this heterogeneity 2

4 varies significantly and meaningfully across parties and within parties across different issue dimensions. Furthermore, construct validity is explored by relating the measure to other instruments used in the field. The fourth section is concerned with testing the hypothesis on the heterogeneitysalience link. After discussing the measurement of salience, control variables, and methods, we present the results. A final section concludes and proposes avenues for future research on intra-party preference heterogeneity. 2. Intra-party heterogeneity concept, measures, and implications This section is divided into three parts. In a first step, we clarify what we mean by intra-party heterogeneity and relate this general definition to our particular focus on candidates for national elections and their issue positions. In a second step, we discuss previous research on intra-party heterogeneity and ways to measure it. Third, we derive our main hypothesis that parties tend to emphasize issues over which their elites hold similar, rather than heterogeneous, preferences and discuss previous research on the link between intra-party heterogeneity and issue salience. Conceptualizing intra-party heterogeneity We understand intra-party elite policy preference heterogeneity, or short intra-party heterogeneity, as the degree to which party elites hold different policy preferences. We are thus interested in what has been labelled (absence of) programmatic cohesion in previous studies, i.e. the general agreement within a party organization on specific issue positions (Kitschelt/Smyth 2002: 1229) as opposed to behavioral unity capturing the observable degree to which members of a group act in unison (Sieberer 2006: 151; see also Bäck/Debus 2016: 5 and Gianetti/Benoit 2009: 5). Two aspects of our definition merit further discussion, relating to the specification of whose preferences and which preferences. Following Norris (1995: 28), who follows May (1973: ), we understand party leaders 2 as members of the government and shadow cabinet, legislators, candidates for elected office, convention delegates and members of the national executive committee. Thus, we focus on the top level of a three-tier party hierarchy, which consists of leaders, sub-leaders, and non-leaders. 3 If the goal is to 2 In the following, we will use the terms elites and leaders synonymously. 3 Where sub-leaders are defined as regional and local party office-holders, other constituency activists, passive grassroots members and voting supporters and non-leaders as occasional and lukewarm party supporters (Norris 1995: 28). 3

5 consider the impact of intra-party heterogeneity on inter-party politics and party-voter relationships, we need to focus on the subset of party members who matter for intra-party politics and are visible to the public. We would argue that the group of party leaders, as defined above, fit these criteria. Drawing the circle wider, e.g., by including sub-leaders, would include actors without continuous influence on intra-party politics at the national level and who are seldom visible to the general public. Analogously, drawing the circle more narrowly (e.g., by focusing only a few top-leaders) would exclude party actors who fit the two criteria of intra-party influence and visibility to the general public. Below, in the empirical section, we focus on candidates in elections to the national legislature. Arguably, they constitute the core of the group of party leaders, because they perfectly fit the two criteria. First, they play an important role in shaping their party s public image and electoral success (Kriesi 2012; MacAllister 2007). Variation across electoral systems notwithstanding, they usually are highly visible during election campaigns (in their respective constituency) and thus give specific faces to their respective party. Second, looking at decision-making within the party, candidates likely constitute leaders who matter for shaping what the larger party platform looks like (Däubler 2012; Dolezal et al. 2012; Kavanagh 1981). Note that focusing on candidates empirically converges with focusing on party officials 4 : Many of the candidates standing in national elections also hold important roles within the party, be it because they hold official functions in the party s inner hierarchy as well or due to their unofficial influence associated with their role as candidates. The second aspect of the definition of party elites which merits discussion is what we mean by policy preferences. The term may refer to different levels of abstraction. On the most specific level, party elites may disagree over a specific policy issue, e.g., the rate of the value-added tax for a specific product. More generally, there may be disagreement among party elites with regard to different general economic principles, such as economic individualism and humanitarianism. Finally, there may be different beliefs about the ideological position of the party (in a one- or two-dimensional space). Our approach below is to focus, on the one hand, on overarching conflict dimensions which are widely believed to structure political conflict in the developed democracies, namely the economic left-right dimension and the conflict between libertarian and authoritarian socio-cultural policies. On the other hand, we supplement this perspective by a more disaggregated approach that looks at more specific issues that either form a subset of these general conflict dimensions (e.g. multiculturalism) or are not clearly aligned with them (e.g. European integration; protection of the environment). 5 4 One feasible way of focusing on party officials is to survey convention delegates. See the European Political Parties Middle-Level Elites survey (Niedermayer/Schmitt 1987) for an early example of an internationally coordinated project that follows this approach. 5 Analyzing disaggregated policy preferences hedges against the risk of focusing on ideological dimensions where, from a psychological point of view, none exist. We are not aware of research which actually tests the 4

6 Quantitative approaches on intra-party heterogeneity and related concepts Previous research has studied intra-party policy heterogeneity from different perspectives. To this date, however, it has proven difficult to construct measures that allow us to compare the extent of intra-party heterogeneity between parties, cross-nationally and over time. On the one hand, qualitative researchers have been well aware of the fact that parties are not unitary actors but consist of different actors and factions with partially divergent policy preferences (e.g., Daalder 1983; Eldersveld 1964). The qualitative approach to intra-party politics enhances our understanding of the working of political parties but finds its (natural) limits when it comes to detecting general patterns across many cases. To achieve this particular goal, we need to construct quantitative measures of intra-party heterogeneity. We can identify at least five different approaches to this task: (1.) Research on the unity of parties within legislatures, (2.) studies that deal with the ambiguity of parties policy stances, (3.) attempts to study intra-party heterogeneity via quantitative analyses of political texts, (4.) expert assessments of intra-party heterogeneity and (5) the use of party elite surveys. Of these, only the latter three approaches represent a direct attempt to measure intra-party preference heterogeneity. Yet it is worthwhile to discuss the first two strands of literature against the backdrop of our notion of intraparty heterogeneity as well. We discuss these approaches in turn. First, scholars have been studying how united parties are with regard to their voting behavior in legislatures. Data from roll-call voting inform us about the level of intra-party unity in the voting behavior of members of parliament (MPs), potentially even broken down to different issue dimensions. Yet, scholars of legislative politics are of course aware that voting unity is not the same as intra-party preference homogeneity or programmatic cohesion: They are precisely interested in how discipline and the institutions which foster discipline can lead to homogenous voting behavior, i.e. unity, even when the underlying policy preferences of MPs diverge (e.g., Hazan 2003; Sieberer 2006; Stecker 2013). In short, party unity with regard to the voting behavior of MPs is an important phenomenon in itself, but measures of such voting unity can give us at best a distorted picture of the general preference heterogeneity across issue dimensions among party leaders. A second strand of related research departs from the assumption that parties occupy precisely defined positions in the policy space by suggesting that parties deliberately use tactics of blurring (Rovny 2012) and ambiguity (Somer-Topcu 2014) in order to maximize votes. This emerging literature conceives of programmatic clarity as depending on strategic choices made by the party. Thus, ambiguity is not regarded as resulting from intra-party heterogeneity. To the contrary, this approach assumption that specific policy preferences of European party leaders are structured along two dimensions. But see Lupton et al. (2015), who demonstrate for the American case that elites attitudes form a single (liberalconservative) ideological dimension. 5

7 conceptualizes parties as unitary actors, as bodies whose behavior is controlled completely by party leaders. Crucially, there the term party leaders is understood in a much narrower sense than we do here. In a seminal contribution to this literature, Rovny (2012: 271) asserts that parties blur their position on some issues, a strategy which he defines as the deliberate misrepresentation of party positions on the part of party leaders. Accordingly, party leaders employ this strategy with regard to issues that are somehow detrimental to them, on which they do not take outstanding positions [ ] The goal of the strategy is to misrepresent the distance between the party and its potential voters on the critical dimension. By providing a blurring stance, a party may more easily adjust its political message to varied audiences or different segments of the electorate (Rovny 2012: 273). Note that position blurring is clearly conceptually distinct from intra-party dissent (Rovny 2012: 272). Consistently, Rovny stresses that the CHES questions he uses to assess the degree of position blurring asks specifically about the position of party leadership (Rovny 2012: 278, emphasis in the original). Similarly, Somer-Topcu (2014) argues that parties utilize broad-appeal strategies in order to mobilize diverse voters by convincing them that the party holds a position close to their own. Again, sending ambiguous and mixed messages is assumed to be a deliberate strategy of parties conceptualized as unitary actors. Methodologically, both studies draw on the variance in either experts (Rovny 2012) or voters (Somer-Topcu 2014) perceptions of the parties positions to study the alleged consequences of ambiguity rather than measuring party behavior which led to ambiguity in the first place. In yet another interesting contribution to this literature, Bräuninger and Giger (2016) consider the role of party activists vis-à-vis the general electorate and argue that party leaders use programmatic ambiguity to bridge the demands of their two principals, the general electorate and party activists (Bräuninger/Giger 2016: 18). They measure programmatic ambiguity (more) directly as positional ambiguity on a general left-right scale using party manifestos. 6 These contributions treat parties as unitary actors who can and do choose their positions strategically, a process in which the party adjusts the party s political messages to varied audiences. We share with these approaches the initial point that parties do not always have clearly identifiable policy positions. We also agree that this phenomenon has been neglected in research on party behavior and competition and requires further study. Yet, we suspect that such positional ambiguity may not only arise from strategic choices by the party or its very top leadership. It could also be a consequence of a party speaking with different voices, namely the voices of party elites (as defined above) with different policy positions (see also Lo et al. forthcoming). In order to study the role of intra-party heterogeneity 6 See also Giebler et al. (2015) and Lo et al. (forthcoming) for further ambiguity measures derived from the Manifesto database. 6

8 for positional ambiguity vis-à-vis strategic choices, more direct measures of intra-party preference heterogeneity are needed. The next three approaches attempt to produce such measures. A third strand of research has more directly looked at intra-party heterogeneity by quantitatively analyzing political texts produced by intra-party actors in different arenas (e.g. Bernauer/Bräuninger 2009; Bäck/Debus 2016: chapter 5; Ceron forthcoming; Greene/Haber forthcoming). In a pioneering study, Bernauer and Bräuninger (2009), analyze plenary speeches in the 15 th German Bundestag to study how intra-party preference heterogeneity is linked to faction membership. Greene and Haber (forthcoming) test their theoretical ideas on the determinants of intraparty preference heterogeneity via an automated text analysis of speeches held at party congresses of the two major parties in Germany and France. Specifically, they measure the variance of the estimated position of convention speakers on a left-right dimension. Ceron (forthcoming) leverages motions of party factions presented at party congresses in Italy in order to compute an index of intra-party division, a weighted standard deviation of positions of factions on a left-right scale, and studies how it affects a party s probability of being part of the governing coalition. This line of work is very promising and seems to be developing in such a direction, but for the time being it has not produced cross-national comparable data on intra-party heterogeneity on a larger scale (let alone for different issue dimensions). A fourth approach relies on expert judgements of intra-party heterogeneity or conflict. Like experts can be surveyed on party positions, such surveys can in principle also include items on how unified or divided party elites are in their preferences with regard to specific issues. To our knowledge, the only existing example is a question in the Chapel Hill Expert Survey 2010 (Bakker et al. 2015) and its predecessors asking about internal dissent or conflict within the party with regard to European integration. 7 Note that there is a slight, but important conceptual difference between preference heterogeneity and dissent/conflict. Preference heterogeneity will not always result in open conflict. Open conflict is probably a consequence of preference heterogeneity on an issue, its salience, and the unwillingness to compromise. This touches upon a more general concern with expert assessments of intra-party heterogeneity: Even competent experts might have a hard time assessing pure intra-party preference heterogeneity; as expert assessments are likely to reflect the amount of visible conflict. 8 While surveying experts is, thus, a feasible way to produce measures of intra-party preference heterogeneity, such measures might be noisy and even biased. On a more practical note, expert measures of intra-party preference heterogeneity do hardly exist. The notable exception, i.e. the 7 Specifically, the respective variable EU_dissent in the CHES 2010 records the internal dissent or conflict in the party on European integration in 2010 on a scale from zero ( party was completely united ) to ten ( partly was completely divided ). 8 Hooghe et al. (2010) report that expert assessments of internal party dissent (on European integration) tend to have larger variances than those of party positions lending empirical support to the expectation that rating dissent is generally more difficult. 7

9 Chapel Hill measure of internal dissent on European integration, has allowed scholars to study the implications of intra-party dissent at least with regard to the issue of European integration (de Vries/van de Wardt 2011; Hellström/Blomgren 2016; Hobolt/de Vries 2015; Netjes/Binnema 2007; Spoon 2012; Steenbergen/ Scott 2004, van de Wardt et al. 2014). This work, to which we return below, has examined how intra-party dissent affects the salience of European integration. The extensive use of the CHES EU dissent indicator reveals the interest of the scholarly community in cross-national data on intra-party heterogeneity on an issue-specific basis. A final fifth approach (and the one adopted here) is to measure intra-party heterogeneity via surveys of party elites. 9 In our view, surveys of party elites are a promising source for measuring intra-party heterogeneity not only due to the lack of existing alternatives documented above. Party elite survey data give us direct, pertinent, and easy-to-process empirical information on intra-party policy preference heterogeneity in a standardized way. Given that elites from different parties answered same issue questions, it is straightforward to construct measures that allow us to compare the heterogeneity between parties, potentially even cross-nationally. Moreover, given appropriate items, the approach allows us to measure intra-party heterogeneity on different levels of issue aggregation. Most importantly, we can measure intra-party heterogeneity with regard to specific issue dimensions and thereby go beyond global measures of intra-party heterogeneity. There are only a handful of existing studies (we know of) that derive measures of intra-party preference heterogeneity from surveys of party elites, however. Bailer (2009) utilizes surveys of members of the European Parliament (MEPs) to assess the impact of the EU s Eastern enlargement in 2004 on the general ideological heterogeneity of the EP s parliamentary groups. Specifically, she compares intra-group standard deviations of MEP s left-right self-placements before and after the enlargement. Andeweg and Thomassen (2011) draw on surveys of members of the Dutch Lower House to describe patterns of intra-party heterogeneity on a range of different issue items using van der Eijks s (2001) measure of agreement (see below). Their goal is to show that unity in parliamentary behavior might also arise from preference homogeneity, among other factors. In publications on Eastern Europe (Kitschelt/Smyth 2002) and Latin America (Kitschelt et al. 2010), Kitschelt and colleagues draw on party elite survey data to measure the cohesiveness of parties. They study cohesiveness as a feature of the programmatic structuration of party systems, i.e. the capacity of political parties to compete on the basis of clearly articulated issue-based programs (Kitschelt/Smyth 2002: 1228). Kitschelt et al. s (2010) measure of cohesion is the intra-party standard deviation of legislators positions on a number of issue items. While they are mostly interested in explaining and 9 For a recent general review on interviews and surveys in legislative research including an overview of available recent datasets, see Bailer (2014). 8

10 comparing variation in this metric across countries, the study also looks at general differences between parties, specifically at the determinants of intra-party preference heterogeneity (party size, governing experience, left ideology, extremist parties); however, they do not disaggregate to the level of individual issues or issue dimensions. 10 There are a few more studies that use elite survey data to construct positions of political actors especially of legislators (e.g. Kam 2001; Saiegh 2009; 2015), but also of candidates (e.g. Jankowski et al. forthcomigg). All of these studies demonstrate the empirical value of using party elite survey data. We build on this work below, by using this data to explicitly measure intra-party heterogeneity at the party level. In doing so we go beyond our predecessors in that we compare heterogeneity between parties while differentiating between different issue dimensions in a cross-national study. 11 Despite the general value of party elite survey data, there are also some limitations to consider (also see the discussions by Bailer 2014; Benoit and Gianetti 2009: 232; Laver 2014: 213f. and Saiegh 2009: ). First, an obvious potential problem are low and possibly biased response rates. Valid, comparable measures of intra-party heterogeneity presuppose that the responses of the sample reflect the actual differences in intra-party preference heterogeneity within parties. This may not be the case. E.g., if candidates who dissent with the official party line are less likely to participate, party heterogeneity will be underestimated because party hacks then dominate the sample. The rather low percentages of successful candidates in the sample (Zittel 2015: 290) may be interpreted as evidence for this very situation. A second concern is that it is not clear who actually answers the questionnaires. Is it the party leader or her staff? Assuming that staffers will be more careful to deviate from the party line, this potential problem implies an underestimation of party heterogeneity and could introduce noise and bias to our estimates. Third, even if the questionnaire is filled in by the party professional the recorded responses on issue positions might be seen as possibly not reflecting the party leaders sincere preferences. In the end, it is impossible to know to which extent the revealed preferences actually reflect sincere preferences. Yet, we would contend that the amount of intra-party heterogeneity revealed by elites in such a very low-cost situation of a largely anonymous, academic survey tells us much about the underlying preference heterogeneity within a party that matters for 10 Kitschelt/Smith (2002) measure cohesion in a somewhat different way: Their measure is the standard deviation of where the surveyed middle-level elites place all parties (note: not themselves and not even only their party) on a range of issues. Like Kitschelt et al. (2010) they are mostly interested in cross-country differences and pursue an aggregated approach that doesn t look at cohesiveness of party-issue combinations (as we do below). Somewhat curiously, they do discuss the role of salience: Salience-based theories of party competition posit that parties compete by boosting the salience of issues in an area they own and displaying a clear programmatic profile on these issues (Kitschelt et al. 2010: 1245). Yet, they do not statistically relate cohesiveness and issue salience. 11 More precisely: In contrast to Kitschelt and colleagues (Kitschelt/Smyth 2002; Kitschelt et al. 2010) and similar to Andeweg and Thomassen (2011), we disaggregate to the level of party-issue observations. In contrast to Andeweg and Thomassen and similar to Kitschelt and colleagues, we use cross-national elite survey data that give us enough data points to study the consequences of intra-party heterogeneity via inferential statistics. 9

11 practical purposes, even though we cannot rule out that party professionals adopt issue positions also for strategic motives even in such a setting. 12 Given these concerns, we turned to the candidate survey data with a healthy dose of skepticism and were careful to check the validity of the measures we can derive from these data first, as detailed below. In conclusion, we thus concur with Michael Laver s (2014: 214) verdict who, focusing on legislators, states that [t]here is an understandable suspicion within the scholarly community that those few legislators who do respond to surveys are atypical and possibly strategic in their responses. Nonetheless, data derived from surveys of legislators policy positions analyzed carefully in a way that takes account of these possible problems are underutilized by scholars who confront theoretical models with positions of real politicians. Intra-party heterogeneity and issue-salience In the last part of this section, we turn to a potential implication of intra-party heterogeneity. Theories of party competition such as the saliency theory (Budge and Farlie 1983) or the related concept of issue ownership (Petrocik 1996) highlight that parties differ not only with regard to their policy positions on issues, but also over which issues are emphasized by them in the first place, i.e. issue salience. Following in the footsteps of existing research on the conflict over EU integration (de Vries/van de Wardt 2011; Hellström/Blomgren 2016; Hobolt/de Vries 2015; Netjes/Binnema 2007; Spoon 2012; Steenbergen/Scott 2004, van de Wardt et al. 2014), below we study how intra-party heterogeneity affects the salience parties attach to multiple issue areas. Before discussing theoretical mechanisms, a conceptual clarification is in order: Given that we have relaxed the assumption of the party as a unitary actor, we consequently do not think of a party s issue salience as being the product of a lone strategic choice made by its very top leader, nor as something inherently monolithic. Rather, what the party as a whole emphasizes in public is the sum of what its individual, publicly visible party elites emphasize. We can think of a party s issue emphasis as an aggregate, i.e. a weighted average of the attention various intra-party actors devote to an issue in their public statements (which can be observed by, e.g., respondents of the Chapel Hill expert survey). Issue emphasis in collective decisions by the party (e.g., what to write into party manifestos) should be seen 12 Beyond these general concerns with party elite survey data, there is an additional concern that is specific to candidate survey data as implemented in the CCS we use below: The universe of candidates is made up of (soon to be) winners and losers of the electoral race. In contrast, research on party behavior is mostly interested in the winners. The broader scope of considering candidates might be a disadvantage; for example when one it interested in the heterogeneity of parties in legislatures. Assuming that party leaders who toe the party line are more likely to be elected, estimates of intra-party heterogeneity based on the universe of all candidates will be too high for these purposes. Yet, for our interest in parties strategies in elections it makes sense to consider all candidates. Of course, it is also possible to integrate information on who was elected to study empirically how elected candidates differ from unelected ones. 10

12 as the result of intra-party decision-making involving party elites that potentially hold divergent preferences. It likely involves considerations of what might be in the strategic interest of the party, but also negotiation and compromise. We thus expect the amount of attention given to different policy areas in parties collective decisions to result from strategic interests of the party as well as from compromises between groups with different issue preferences within the party. Given this conceptualization, we expect that parties place more emphasis on issues (with regard to both: the general public appearance of the party and its collective decisions) on which their elites agree. There are at least two analytically distinct arguments leading to this expectation. First, party elites have an electoral incentive to emphasize internally uncontroversial issues if they believe that voters reward parties which appear united and punish those which appear divided. To prevent this reaction, the candidates of a party may downplay issues they disagree on. This argument, which centers on electoral-rationality or vote-seeking, is emphasized by research on the heterogeneitysalience link with regard to European integration (de Vries/van de Wardt 2011:177; Hobolt and de Vries 2015: 1166). A related argument is that, in addition to running the risk of electoral defeat, the attempt to emphasize internally controversial issues might jeopardize the parties very survival, as such emphasis could lead to mass resignations and even party splits. Against this backdrop, cohesionseeking appears like a prudent goal and downplaying issues on which intra-party heterogeneity is high like an internal-rational strategy to get there (Steenbergen and Scott 2004: 169). 13 There is a second, less rationalistic argument for the heterogeneity-salience link, which seems straightforward to us if the party is conceptualized as a group of individuals without identical policy preferences. Party elites may find it easier to agree on common positions and policy proposals if they hold the same positions on an issue in the first place. As a direct consequence, they are likely to say more about these issues in public party statements (oral as well as written). This mechanism rests on the suspicion that electoral incentives are not the only relevant determinants of elite behavior and thus of issue salience. If party leaders have heterogeneous policy preferences and stick to them out of conviction, the salience of issues in a given party is the result of negotiation and compromise. What makes the agenda is what party leaders can agree on, given the policy preferences of their (leading) members. While it is difficult to distinguish these mechanisms empirically, both lead to the same expectation: The more consensual an issue is among party elites, the higher is its salience (i.e. the more it is 13 This argument mirrors the point frequently made in research on legislative agenda setting that party leaders may set the legislative agenda so as to minimize the salience of disunity [ ] and avoid divisive legislative votes altogether (Giannetti/Benoit 2009: 9). We suspect that a similar mechanism applies also outside the legislative arena, not only reflecting incentives of the top party leaders, but of political candidates and even the party more broadly. 11

13 stressed by the party in its public statements, including its manifesto). Or, equivalently: The higher the intra-party preference heterogeneity on an issue, the lower is its salience (Hypothesis 1). As mentioned above, previous studies on the heterogeneity-salience link have exclusively tested the effect of intra-party dissent on European integration on the salience of European integration using the CHES dissent measure discussed above. Some of the studies report that salience decreases (monotonically) with dissent (Hobolt and de Vries 2015; van de Wardt et al. 2014). Other studies document evidence for a slightly U-shaped relation with salience first decreasing and then, at very high levels of dissent, increasing with intra-party dissent on European integration (de Vries/van de Wardt 2011; Netjes/Binnema 2007; Spoon 2012; Steenbergen/Scott 2004). The U-shaped relation goes back to Steenbergen and Scott who rationalize it in the following way: However, there are limitations to the strategic manipulation of issue salience in the face of internal dissent. While parties that experience modest levels of dissent may be able to deemphasize an issue, parties experiencing major dissent may be unable to do this. Here a significant faction disagrees with the mainstream of the party on an issue. It is likely that such disagreement will inspire extensive debate within the party and, as a consequence, the divisive issue will receive a great deal of emphasis (Steenbergen/Scott 2004: 169). While this might be the case, the U-shaped relation could also be an artifact of the way heterogeneity is measured, with experts observing only high levels of intra-party dissent when preference heterogeneity is made salient through open conflict. In our analysis below, which looks more directly at preference heterogeneity, we therefore do not expect a U-shaped effect. As a caveat, it should be noted that the causal link between intra-party heterogeneity and salience is likely to be a complex one. First, the causal arrow may point in the opposite direction of the one specified above: Candidates may have homogeneous preferences regarding an issue because it is a (chronically) salient issue for the party. Consider that parties typically emerge against the backdrop of societal conflict over a policy issue, such as the social question or concern for the environment. In the following, they are likely to attract and select politicians who have the same position on this issue, but do not necessarily agree on other issues. Second, both phenomena elite preference heterogeneity (in a broad sense) and issue salience on the party level may have a common cause. Top leaders of the party may strategically decide which issues to emphasize in the upcoming electoral campaign and then whip candidates into following this official party line. The complexity and simultaneity of intra-party socialization and communication processes certainly make it difficult to disentangle the direction of causality. In suggesting a causal arrow pointing from intra-party heterogeneity here we follow the prior literature on EU issue salience, yet we see an endogeneity concern. On the one hand, we would argue against that concern that evidence of a systematic association between heterogeneity and salience is sufficiently novel and interesting even when there remains doubt about the exact nature of 12

14 the causal processes linking the two phenomena. On the other hand, we implement empirical robustness checks below that try to mitigate the endogeneity concern. 3. A measure of intra-party heterogeneity in candidates issue positions Given the novelty of our measure of intra-party heterogeneity, it seems prudent to devote some space to describing it. We proceed in three steps: First, we explicate in detail how the CCS data can be used to derive the measure(s) of intra-party heterogeneity based on candidate s issue positions. Second, we take a descriptive look at these data and show that candidates often hold quite heterogeneous issue positions. The extent of heterogeneity varies significantly and meaningfully across parties and within parties across different issue dimensions. Third, we relate our measure to other instruments in the field to assess its validity. To measure intra-party heterogeneity, we use the Comparative Candidates Survey (CCS 2016). The CCS is an international collaboration which integrates a number of post-election 14 polls of candidates for national legislatures. While the surveys are locally organized, they draw on a common core questionnaire to allow for international comparison. Although the CCS is mainly concerned with electoral campaigning, specifically whether candidates run individualized vs. party-based campaigns (Zittel 2015), it also includes a battery on candidates attitudes toward specific issues, which is most pertinent for our purposes. Module 1 of the CCS was carried out in 36 elections between 2005 and A second module runs from 2013 to 2018 and continues the issue items from Module 1. The current version of the dataset (version 3.1; April 2016) includes data for 31 national legislative elections in 23 developed democracies. Unfortunately, the issue items were not included in their entirety in every national survey. This is why the number of available cases (elections and countries) varies somewhat over the different issue areas we consider below. Moreover, we had to drop three elections entirely from the analysis due to almost completely missing issue position data. This leaves us with data for 28 elections from 21 different countries This implies that party heterogeneity is measured after the respective election. While this does not lower the validity of the measurement per se, it renders this data suboptimal for testing the impact of party heterogeneity on issue salience in the election which the candidates competed in. While we would expect stability in candidate s policy preferences, at least over the short-run, this problem exacerbates the endogeneity concern a little (as far as the analyses with the Manifesto data are concerned that measure salience in the election campaign). We return to this problem in a robustness check below. 15 These are Australia (2007, 2010), Austria (2008), Belgium (2007, 2010), Czech Republic (2006), Denmark (2011), Estonia (2011), Finland (2007), Finland (2011), Germany (2005, 2009), Greece (2007, 2009, 2012), 13

15 While the items are always the same (or rather: functionally equivalent), several other features of the national candidate surveys differ somewhat due to the decentralized nature of the project: The definition of the universe/sampling procedures, the modes of data collection and the response rates. While most surveys opted to survey all candidates for the respective parliamentary elections, some of the surveys were limited to specific parties with less (e.g. only those represented in parliament) and occasionally more restrictive inclusion rules (e.g. only the two largest parties). The overwhelming majority of surveys used self-completion mail-back questionnaires. The mean response rate (including only cases for which detailed figures on response rates are available) is a respectable 38.6% with a maximum of 66.7% (Iceland 2009) and a minimum of 16.2% (Czech Republic 2006). The issue battery in the CCS asked for respondents agreement with fourteen different statements on five point scales. Additionally, the survey included four separate questions on European integration (see Table A 1 in the appendix for a complete list of issue items included). As mentioned, our approach is to both look at overarching conflict dimensions which are widely believed to structure political conflict in the developed democracies as well as more specific issues. As for the general conflict dimensions, we follow the widely accepted notion of a two-dimensional conflict space. One is an economic dimension encompassing positions on social welfare, redistribution and market intervention, the other refers to socio-cultural issues contrasting authoritarian and libertarian positions (such as law and order vs. civil liberties) and traditional moral views vs. libertarian views on social lifestyle issues (such as same sex marriage and abortion). Focusing on those items which could be clearly mapped to only one of these two dimensions on theoretical grounds, we constructed simple additive indices for these two dimensions from three (economic dimension) respectively five (sociocultural) single items rescaled to range from zero to one. 16 As for the specific issue, we focus on European integration, the cultural assimilation of migrants, and environmental protection. These were selected for the following two reasons: We expect them to vary in salience across parties; we were able to identify equivalent categories in the datasets we use to measure salience. While positions on European integration are again measured with an additive index of three single items (rescaled to the same range), our measures on cultural assimilation of migrants and environmental protection are single items (see Table 1 for a full list of all items used to measure issue positions). - - Table 1 about here - - Hungary (2010), Iceland (2009), Ireland (2007), Italy (2013), Netherlands (2006), Norway (2009), Malta (2013), Portugal (2009, 2011), Romania (2012), Sweden (2010), Switzerland (2007, 2011), and United Kingdom (2013). Canada 2008, Finland 2007 and New Zealand 2011 were dropped from the dataset. 16 Table A 4 in the appendix documents the pairwise correlations of all items. The pattern shows that, overall, items of a common dimension correlate positively with one another and that these associations are stronger than between items of different dimensions. 14

16 Given these position measures, it is relatively straightforward to calculate measures of intra-party heterogeneity. For our three additive, quasi-interval level indices, this is as simple as calculating the standard deviations of candidate positions within parties. For the two single-items, we follow van der Eijk s (2001) proposal on how best to measure agreement on an ordered rating scale instead. In order to measure heterogeneity rather than homogeneity (as does the standard deviation), we reverse the sign of his index to produce a measure of disagreement. 17 One potential complication is that our measures of intra-party heterogeneity are likely to become very noisy once we observe only a small number of candidates for a specific party. As an easy fix, we include only parties for which at least ten valid candidate observation on the issue (dimension) in question are available in all analyses discussed below. Having described the construction of the measures, we turn to some descriptive empirical findings. In order to convey a general impression, we only focus on the two overarching conflict dimensions for now. Figure 1 shows the distribution of candidates issue positions in the two-dimensional policy space across the 22 elections for which we have complete data on the relevant issue items. In addition to the position of single candidates, the figure also shows the mean positions of the candidates of a particular party. Moreover, the stroke length of the respective crosses is set to the standard deviation of the positions of the respective parties candidates on the relevant dimension. E.g., look at the German Bundestag election of 2009 (second row, last panel): Comparing the Left party and the Greens, we can see that the Greens are relatively homogenous on the socio-cultural dimension and relatively heterogeneous on the economic dimensions; the reverse is true for the Left party. As experts of German politics may confirm, this is exactly the kind of pattern we would expect a valid measure of intra-party heterogeneity to uncover. More generally, there are four impressions to take away from Figure 1: The (relative) party positions seem reasonable; we generally observe a high amount of intraparty heterogeneity; candidates of different parties often have overlapping positions in the policy space; the patterns we observe with regard to intra-party heterogeneity across dimensions and parties appear sensible as well insofar as we have some prior expectations on these Figure 1 about here - - To go beyond the impressions we can draw from descriptive inspections of Figure 1, we opted to further validate our measures by comparing them with existing alternatives. Such a check for criterion 17 Note that even in these latter cases these measures of disagreement and the simple standard deviations are closely related with correlations of r=0.87 and Another observation is that in those instances where we have observations at different points in time we observe a high degree of stability. This could be interpreted as hinting towards a satisfactory reliability of the measures. 15

17 validity is complicated by the fact that there are not many, if any, measures of intra-party heterogeneity available which could claim higher validity themselves. Against this background, we decided to validate the estimated parties mean issue positions in a first step by comparing them to the CHES and Manifesto positions widely used in comparative politics research. In order to do this, we merged our party-level data derived from the CCS with data from the CHES 2010 (Bakker et al. 2015) and the Manifesto Project (Volkens et al. 2015). 19 We focus on the CHES data here as they include ready-to-use issue specific position measures for political parties that match our five issue dimensions well. Figure 2 plots the mean party positions as estimated from the CCS with their respective counterparts from the CHES and shows a strong correlation in all cases. - - Figure 2 about here - - While this supports the general validity of the position data, it doesn t speak directly to the validity of our intra-party heterogeneity measure. In a second step, we thus wondered whether our intra-party heterogeneity metric correlates with those measures that were proposed in the literature as measures for programmatic ambiguity. We focus on Rovny s (2012) blurring measure, i.e. the standard deviation of CHES expert judgements on party placements, as it can be straightforwardly computed for specific issue dimensions. We find weak positive correlations of 0.30 in the case of European integration and multiculturalism, but for the other three dimensions the correlations are close to zero (see Table A 3 in the appendix). As a third way to validate the CCS data, we can compare our estimate of intra-party heterogeneity on European integration with experts perceptions of party dissent regarding European integration. We obtain a weak positive correlation of This is the most direct validity check we can offer, yet it is limited to one issue dimension and, as previously discussed, open dissent or conflict is not identical 19 A few technical details on this data merging are in order: The CHES 2010 was administered in the autumn of 2010 and generally constitutes a good, though not perfect match with the CCS data that includes data on elections between 2005 and 2013 with 18 out of the 28 elections we use having taken place between 2009 and In case of the Manifesto that is similarly structured along elections, the time dimension posed no problems. There are a few instances in which institutionalized party collaborations (such as between the CDU and CSU in Germany or fixed electoral alliances like the Portuguese CDU alliance between the PEV and PCP) are treated differently across the datasets. In these cases we consistently aggregated the data to the alliance level if necessary using weighted means in order to be able to merge the data. For example: We created a combined entry for the CDU/CSU from the individual entries for CDU and CSU in the CHES which reflects the CDU s values to 80% and the CSU s values to 20% (which roughly corresponds to their respective seat shares in their joint fractions in the 16 th and 17 th German Bundestag). Of course, the CHES and the Manifesto data do not include information on every party included in the CCS (and also vice versa). Overall, we could successfully merge information on 158 parties with CHES 2010 and 172 with the Manifesto data. 144 observations are common to all three datasets. 20 Note that the correlation between blurring on European integration and EU intra-party dissent is

18 with pure preference heterogeneity. 21 In light of these problems, it is actually reassuring that we do find a meaningful, if only weak, correlation. Clearly, there is still a long way to go in the development, exploration, and validation of measures for intra-party phenomena such as heterogeneity, dissent, and conflict. 4. Intra-party heterogeneity and issue salience To relate our measures of intra-party heterogeneity to issue salience, we need relatively fine-grained information on the importance various issues have for parties in our sample of developed democracies. Like for the positional data, we can draw on the Manifesto data and the CHES data. The former gives us a manifesto-based estimate of issue salience: As Klingemann et al. write, the main information measured in the coding of manifesto data is the relative emphasis parties give to the different messages they wish to transmit to electors (Klingemann et al. 2006: 116; emphasis in the original). The CHES data contain experts assessments of the salience various issues have in the party s public appearance. Drawing on the conceptual distinction from above, the Manifesto measure taps into the salience of issues in collective party decisions. The expert assessments are broader summary judgements of a party s publicly observable issue emphasis, which likely involve multiple considerations of the expert. As such, they represent some aggregation of individual elite and collective party communication and behavior. In accordance with our theoretical reasoning above, we thus expected intra-party heterogeneity to be negatively related to both measures and conceive of measuring salience in two very different ways as robustness check. How did we match our five intra-party heterogeneity measures to the salience data? As to the Manifesto data, this required computing the fraction of statements on a particular issue dimension by summing up the relevant categories belonging to this dimension. We thereby follow Tavits and Potter (2015:749) in operationalizing the salience of the general economic and socio-cultural ideological dimensions. Like them, 21 We have argued theoretically that open conflict is observed when preference heterogeneity on an issue is high and the issue is also salient. This reasoning is supported by our data. This further validates our measure. We ran a RE-model with EU dissent predicted by an interaction between EU preference heterogeneity and EU salience (as measured by the CHES) and obtained a statistically significant interaction coefficient. At low levels of salience, preference heterogeneity is essentially unrelated to dissent; at high levels of salience, preference heterogeneity is closely related to dissent. In other words, preference heterogeneity does not lead to dissent when parties remain largely silent on an issue (and, as we argued above, this is therefore a wise strategy). This is exactly the pattern we would expect. This finding supports our conceptual differentiation between preference heterogeneity and salience plus our argument that the expert measure is not an unbiased measure of pure preference heterogeneity. 17

19 as we are interested in the frequency with which a party makes reference to either the interest or the values dimension rather than its actual left or right position on these dimensions we do not distinguish between left-leaning and right-leaning positions (as is typical in manifesto-based research). Therefore, in order to measure Emphasis on each of the two dimensions, we record, for each party in each election, the total sum of all CMP components on that dimension (Tavits and Potter 2015: 749; emphasis in the original). In addition to measures for these two dimensions, which draw on their coding scheme, we computed analogous measures for European integration, environmental protection, and multiculturalism (see Table A 2 in the appendix with details on our coding schemes and correlations between the Manifesto and CHES salience measures). 22 As for the CHES, we have direct measures for the salience of European integration, environmental protection and multiculturalism. For the general economic and socio-cultural conflict dimension, there are no overall salience scores readily available. We therefore took the mean of more disaggregated issues which belong to one of the two dimensions, respectively. In analyzing the link between intra-party heterogeneity and issue salience, we first look at simple bivariate scatterplots and then go on to regress salience on heterogeneity. Note that our data exhibit a hierarchical two-level structure: Our observations are on political parties which are nested in elections (or, equivalently, country-years). We therefore estimated both fixed effect(fe)-models and random effect(re)-models. The FE-models leverage within-variance only and thereby implicitly control for all possible confounders which affect issue salience at the country-year level. The RE-models attribute some of the between-variance to election-level intercepts modeled as error terms. Given that our models implicitly control for confounders at the country-year level, we included only a control variable at the party-level (for now). The most important potential confounder at the party level is the extremity of the party position on the issue dimension in question. We measure this via the absolute distance from the mid-point of the issue dimension according to the CHES position measure. In line with previous research (de Vries/van de Wardt 2011:176, Wagner 2012), we expected parties with an extreme position on an issue dimension to attach more salience to this dimension. At the same time, candidates of parties with extreme positions on a particular issue are more likely to hold homogenous positions on this issue. This partly reflects a methodological problem: Candidates from a party which is located close to one of the endpoints of the scale have simply less room to move than those of parties with moderate positions In these cases, we took the natural logarithm of the resulting sum (i.e. the percentage share of all statements in a manifesto devoted to the respective issue dimension) as the distribution is otherwise strongly right-skewed. 23 Note that this is a thorny issue indeed: From an intra-party heterogeneity point of view, an extreme position of a party as a collective entity can also be considered a consequence of its elites having uniformly the same extreme positions on an issue dimension. Moderate collective positions could result from heterogeneous positions of party elites. 18

20 Figure 3 shows bivariate scatterplots of intra-party heterogeneity and the two types of issue salience across the five dimensions. In line with our expectation, we find a negative correlation in all cases. With the exception of the socio-cultural dimension, the correlation is always statistically significant and meaningful in size. Furthermore, this visual inspection does not point to the inverted U-shaped relation that has been hypothesized with regard to intra-party dissent on European integration and the salience of European integration. Thus, the overall pattern of the results strongly supports our expectation. But what about the divergent finding for the socio-cultural dimension? It might be due to the amalgam character of this dimension: It summarizes a number of related, but still somewhat distinct issues. It is possible that parties are not similarly united or divided over these distinct issues and that emphasis on these different issues is not well aligned with each other either. In line with this interpretation, we find a strong negative correlation between intra-party heterogeneity and issue salience when we focus on multiculturalism, which is part of the socio-cultural issue dimension. Overall, the simple bivariate analysis lends initial support to a negative relation between intra-party heterogeneity and salience. - - Figure 3 about here - - The pattern we already observed in the scatterplots holds up in the regression analysis. Table 2 reports estimation results for the CHES salience measures from a total of ten models, i.e. an RE-model and an FE-model for each of the five issue dimensions. There is a negative and statistically significant coefficient for intra-party heterogeneity with regard to the economic dimension, European integration, protection of the environment, and multiculturalism. The socio-cultural dimension deviates from this pattern with a coefficient that is indistinguishable from zero. As expected, we find that parties that are farther from the center of an issue attach more salience to this issue. This is true for all five issue dimensions considered here. - - Table 2 about here - - We now turn to the Manifesto salience measures as dependent variables. The results, which are shown in Table 3, are analogous to those reported above: Intra-party heterogeneity has a negative and statistically significant effect on salience of the economic dimension, European integration, protection of the environment, and multiculturalism; again, the nil hypothesis cannot be rejected in the case of the socio-cultural dimension. While the findings for issue extremity are not as clear cut as with the CHES salience measure, the coefficients always point in the expected direction and are often statistically significant. - - Table 3 about here

21 These findings generally support our argument of a close negative association between intra-party heterogeneity and issue salience. Yet, as was noted above, this association might to some extent reflect reverse causality with salience having an effect on intra-party heterogeneity. These endogeneity concerns arise inter alia from the fact that the candidate surveys are conducted after elections, while party manifestos are agreed upon before the election. One may argue that a straightforward solution to this problem is analyzing the effect of party heterogeneity at time point t on issue salience at time point t Conducting this analysis yields largely the same results as those reported above mitigating the endogeneity concern at least to some extent Table 4 about here Results and implications Quantitative research on the positions of political parties and party competition regularly invokes the assumption that parties are unitary actors with homogenous policy preferences. Many have the suspicion that this assumption does not reflect reality, but data availability has hindered researchers to subject it to empirical tests on a larger scale. In turn, it remains unclear whether relaxing this assumption by explicitly allowing for such heterogeneity will lead to different conclusions about party politics. In this paper, we have made an attempt to address this research gap. The publicly available data of the Comparative Candidates Survey (CCS) were used to look at intra-party elite policy preference heterogeneity, or short intra-party heterogeneity. Relying on this data from 28 national elections in 21 developed countries, we showed that candidates often hold quite heterogeneous issue positions and that the extent of this heterogeneity varies significantly across parties as well as within parties across different issue dimensions. These results once more underline the importance of carefully considering the implications of intraparty heterogeneity when thinking about party competition. They echo what, e.g., Daalder observed more than 30 years ago that in the real world of politics it is hardly defensible to regard the party as a unitary actor (Daalder 1983: 21). Party elite surveys potentially yield a direct, pertinent, and easyto-process empirical information on intra-party heterogeneity in a standardized way. Given appropriate 24 However, it should be noted that our argument for the causal direction heterogeneity -> salience states that party leaders should stress issues on which they can agree. If those whose preferences are measured and those who are involved in drafting the manifestos are not the same, our argument does not apply. Due to candidate replacements between elections, this situation arises when estimating the effect of candidate preferences measured in post-election surveys on issue salience in the following election (manifesto). In this sense, this analysis is not a perfect research design either. 25 Due to (as of now) missing data on issue salience, about half the 100-or-so cases from the analysis presented above have to be dropped. 20

22 items, the approach even allows us to measure intra-party heterogeneity on different levels of issue aggregation. We may thus go beyond the diagnosis of inappropriateness and relax the assumption of parties as unitary actors. In a second step, we have looked at one implication of such intra-party heterogeneity for party strategy and competition. It was argued that intra-party heterogeneity and issue salience go together because parties will emphasize those issues on which intra-party heterogeneity is low. Empirically, we related measures of candidate heterogeneity based on CSS data to data on issue salience measures from the Chapel Hill Expert Survey and the Manifesto Project. Across different issue dimensions and the two salience measures, we consistently found that parties attach lower salience to issues over which they are internally divided. These findings contribute to existing research from the EU context. First, these studies on EU dissent and salience cannot tell us whether the findings are generalizable to other issue areas we have shown that they are (in the sense of a linear association between heterogeneity and salience). Second, we have employed a measure of intra-party heterogeneity which does not run the risk of confounding heterogeneity and salience on the measurement level. Given that the quantitative study of intra-party heterogeneity leads into largely unknown territory, it may be prudent to explore the nature of this heterogeneity further. A promising direction of future research starts with the finding that parties which belong to different party families emphasize different issues. Given the link between intra-party heterogeneity and salience, we would suspect that there is systematic variation in heterogeneity over party families for specific issues as well. That is because the images of parties from a given party family should make its members stick to certain policy issues due to electoral incentives; also, there may be selection and socialization effects which may lead to greater homogeneity of positions on symbolically meaningful issues. Examples include the Green parties, which should be more united on the socio-cultural dimension, specifically on environmental issues, than on the economic dimension; the Christian Democrats should be more united on the socio-cultural dimension, especially on issues concerning traditional morality and order; the Social Democrats and left socialists should be more united on the economic dimension. We would like to close with a cautionary note. Throughout the paper, we have pointed to the methodological problems of using the CCS for measuring intra-party heterogeneity. Technical differences in conducting these surveys (regarding, e.g., the sampling frame and the field time) and possibly lacking cross-cultural equivalence of the issue questions and dimensions may lower comparability. More fundamentally, one may wonder whether elite surveys can ever yield a valid measure of intra-party heterogeneity if we cannot be sure who fills in the answers in mail or online survey and given the skewedness of the realized samples toward unsuccessful candidates (in the CCS; see above). These are important questions, which we have raised but were unable to answer satisfactorily. Clearly, the answers will determine the validity and thus usefulness of utilizing elite 21

23 survey data in the way proposed here. We interpret the fact that our measure of intra-party heterogeneity is correlated with conceptually similar constructs and exhibits theoretically expected effects on different measures of salience as evidence that elite surveys are indeed a prominsing avenue for studying the nature, causes, and consequences of intra-party heterogeneity. Yet, this paper is no more than a first step down this road and further research is needed. 22

24 References Andeweg, Rudy B., and Jacques Thomassen Pathways to party unity: Sanctions, loyalty, homogeneity and division of labour in the Dutch parliament. Party Politics 17(5): Bäck, Hanna and Marc Debus Political Parties, Parliaments, and Legislative Speechmaking. New York: Palgrave Macmillan. Bailer, Stefanie The puzzle of continuing party group cohesion in the European Parliament after Eastern enlargement. In: Gianetti, Daniela and Kenneth Benoit: Intra-party politics and coalition governments. London: Routledge Bailer, Stefanie Interviews and surveys in legislative research. In: Martin, Shane, Thomas Saalfeld and Kaare W. Strøm: The Oxford Handbook of Legislative Studies. Oxford: Oxford University Press Bakker, Ryan, Catherine de Vries, Erica Edwards, Liesbet Hooghe, Seth Jolly, Gary Marks, Jonathan Polk, Jan Rovny, Marco Steenbergen and Milada Anna Vachudova Measuring party positions in Europe: The Chapel Hill expert survey trend file, Party Politics 21(1): Benoit, Kenneth and Daniela Gianetti Intra-party politics and coalition governments: Concluding remarks. In: Gianetti, Daniela and Kenneth Benoit: Intra-party politics and coalition governments. London: Routledge Bernauer, Julian and Thomas Bräuninger Intra-party preference heterogeneity and faction membership in the 15th German Bundestag: A computational text analysis of parliamentary speeches. German Politics 18(3): Budge, Ian and Dennis Farlie Explaining and Predicting Elections: Issue Effects and Party Strategies in Twenty-Three Democracies. London: Allen & Unwin. Bräuninger, Thomas and Nathalie Giger. Forthcoming. Strategic Ambiguity of Party Positions in Multi-Party Competition. Political Science Research and Methods. CCS Comparative Candidates Survey Module I [Dataset - cumulative file]. Distributed by FORS, Lausanne, Ceron, Andrea. Forthcoming. Inter-factional conflicts and government formation: Do party leaders sort out ideological heterogeneity? Party Politics. Daalder, Hans The Comparative Study of European Parties and Party Systems. In: Daalder, Hans und Peter Mair: Western European Party Systems: Continuity & Change. Beverly Hills: Sage Däubler, Thomas The Preparation and Use of Election Manifestos: Learning from the Irish Case. Irish Political Studies 27(1): de Vries, Catherine E., and Marc van de Wardt EU issue salience and domestic party competition. In: Oppermann, Kai and Henrike Vierig: Issue Salience in International Politics. London: Routledge Dolezal, Martin, Laurenz Ennser-Jedenastik, Wolfgang C. Müller und Anna Katharina Winkler The Life Cycle of Party Manifestos: The Austrian Case. West European Politics 35(4): Eldersveld, Samuel J Political Parties: A Behavioral Analysis. Chicago: Rand McNally. Gianetti, Daniela and Kenneth Benoit Intra-party politics and coalition governments in parliamentary democracies. In: Gianetti, Daniela and Kenneth Benoit: Intra-party politics and coalition governments. London: Routledge Giebler, Heiko, Onawa Promise Lacewell, Sven Regel, and Annika Werner Niedergang oder Wandel? Parteitypen und die Krise der repräsentativen Demokratie. In Merkel, Wolfgang: 23

25 Demokratie und Krise. Zum schwierigen Verhältnis von Theorie und Empirie. Wiesbaden: Springer VS Greene, Zachary, and Matthias Haber. Forthcoming. Leadership competition and disagreement at party national congresses. British Journal of Political Science. Hazan, Reuven Y Does Cohesion Equal Discipline? Towards a Conceptual Delineation. The Journal of Legislative Studies 9(4): Hellström, Johan and Magnus Blomgren (2016). Party debate over Europe in national election campaigns: Electoral disunity and party cohesion. European Journal of Political Research 55(2): Hobolt, Sara Binzer and Catherine E. de Vries Issue Entrepreneurship and Multiparty Competition. European Union Politics 48(9): Hooghe, Liesbet, Ryan Bakker, Anna Brigevich, Catherine de Vries, Erica Edwards, Gary Marks, Jan Rovny, Marco Steenbergen, Milada Vachudova (2010). Reliability and validity of the 2002 and 2006 Chapel Hill expert surveys on party positioning. European Journal of Political Research 49(5): Jankowski, Michael, Sebastian Schneider and Markus Tepe. Forthcoming. Ideological alternative? Analyzing Alternative für Deutschland candidates ideal points via black box scaling. Party Politics. Kam, Christopher Do ideological preferences explain parliamentary behaviour? Evidence from Great Britain and Canada. Journal of Legislative Studies 7(4): Kavanagh, Dennis The Politics of Manifestos. Parliamentary Affairs 34(1): Klingemann, Hans-Dieter, Andrea Volkens, Judith Bara, Ian Budge, and Michael D. McDonald Mapping Policy Preferences II. Estimates for Parties, Electors and Governments in Central and Eastern Europe, European Union and OECD Oxford: Oxford University Press. Kitschelt, Herbert, and Regina Smyth Programmatic Party Cohesion in Emerging Postcommunist Democracies: Russia in Comparative Context. Comparative Political Studies 35(10): Kitschelt, Herbert, Kirk A. Hawkins, Juan Pablo Luna, Guillermo Rosas and Elisabeth J. Zechmeister Latin American party systems. New York: Cambridge University Press. Kriesi, Hanspeter Personalization of National Election Campaigns. Party Politics 18(6): Laver, Michael Measuring policy positions in political space. Annual Review of Political Science 17: Lo, James, Sven-Oliver Proksch and Jonathan B. Slapin. Forthcoming. Ideological clarity in multiparty competition: A new measure and test using election manifestos. British Journal of Political Science. Lupton, Robert N., William M. Myers und Judd R. Thornton Political Sophistication and the Dimensionality of Elite and Mass Attitudes, The Journal of Politics 77(2): MacAllister, Ian The Personalization of Politics. In: Dalton, Russell J. und Hans-Dieter Klingemann: The Oxford Handbook of Political Behavior. Oxford: Oxford University Press May, John D Opinion Structure of Political Parties: The Special Law of Curvilinear Disparity. Political Studies 21(2): Netjes, Catherine E. and Harmen A. Binnema The Salience of the European Integration Issue: Three Data Sources Compared. Electoral Studies 26(1):

26 Niedermayer, Oskar and Hermann Schmitt (1987): European Political Parties' Middle-Level Elites. GESIS Data Archive, Cologne. ZA1598 Data file Version 1.0.0, doi: / Norris, Pippa May's Law of Curvilinear Disparity Revisited: Leaders, Officers, Members and Voters in British Political Parties. Party Politics 1(1): Petrocik, John R Issue Ownership in Presidential Elections, with a 1980 Case Study. American Journal of Political Science 40(3): Rovny, Jan Who Emphasizes and Who Blurs? Party Strategies in Multidimensional Competition. European Union Politics 13(2): Saalfeld, Thomas and Kaare W. Strøm Political Parties and Legislators. In: Martin, Shane, Thomas Saalfeld and Kaare W. Strøm: The Oxford Handbook of Legislative Studies. Oxford: Oxford University Press Saiegh, Sebastián M Recovering a basic space from elite surveys: Evidence from Latin America. Legislative Studies Quarterly 34(1): Saiegh, Sebastián M Using joint scaling methods to study ideology and representation: Evidence from Latin America. Political Analysis 23(3): Sieberer, Ulrich Party Unity in Parliamentary Democracies: A Comparative Analysis. The Journal of Legislative Studies 12(2): Somer-Topcu, Zeynep Everything to Everyone: The Electoral Consequences of the Broad- Appeal Strategy in Europe. American Journal of Political Science 59(4): Spoon, Jae-Jae How Salient Is Europe? An Analysis of European Election Manifestos, European Union Politics 13(4): Stecker, Christian How Effects on Party Unity Vary across Votes. Party Politics 21(5): Steenbergen, Marco R. and David J. Scott Contesting Europe? The Salience of European Integration as a Party Issue. In: Marks, Gary and Marco R. Steenbergen: European Integration and Political Conflict. New York: Cambridge University Press Tavits, Margit and Joshua D. Potter The Effect of Inequality and Social Identity on Party Strategies. American Journal of Political Science 59(3): van der Eijk, Cees Measuring Agreement in Ordered Rating Scales. In: Quality and Quantity 35: van de Wardt, Marc, Catherine E. de Vries, and Sara B. Hobolt Exploiting the cracks: Wedge issues in multiparty competition. The Journal of Politics 76(4): Volkens, Andrea, Pola Lehmann, Theres Matthieß, Nicolas Merz, Sven Regel and Annika Werner (2015): The Manifesto Data Collection. Manifesto Project (MRG/CMP/MARPOR). Version 2015a. Berlin: Wissenschaftszentrum Berlin für Sozialforschung (WZB). Wagner, Markus When do parties emphasise extreme positions? How strategic incentives for policy differentiation influence issue importance. European Journal of Political Research 51(1): Zittel, Thomas Constituency Candidates in Comparative Perspective How Personalized Are Constituency Campaigns, Why, and Does It Matter? Electoral Studies 39:

27 Tables and Figurees Figure 1: Candidate positions p in a two-dimensiional policy space s Note: Daata from the Comparativee Candidate Survey (CCS S); x-signs in ndicate the ppositions of individual candidatees (jitter added to prevent overlaying); o tthe crosses indicates the mean m positionss for all candiidates of a specific pparty with the stroke length h being set to tthe standard deviation d of th he positions o f the respectiv ve parties candidatees on the relevvant dimension n. Very small parties and in ndependent caandidates are eexcluded. 26