Under Which Conditions Do Parties Attract Voters’ Reactions to Issues? Party-Varying Issue Voting in German Elections 1987–2009

Are voters’ choices influenced by parties’ position-taking and communication efforts on issues during a campaign? And if so, do voters’ reactions to issues differ across parties? This article outlines a research design for the statistical identification of party-varying issue reactions within the established paradigm of the Spatial Theory of Voting. Using a special feature of conditional logit and probit models – i.e. the estimation of alternative-specific coefficients instead of fixed ‘generic’ issue distance effects – it is possible to detect asymmetrically attached issue saliencies at the level of the voters, and hence at the demand-side of politics. This strategy opens a new way to systematically combine insights obtained by saliency approaches with the Spatial Theory of Voting. An application to the German parliamentary elections from 1987 to 2009 demonstrates that it is predominantly parties taking polar positions – and, more specifically, niche parties taking polar positions – that induce such asymmetric issue voting.

The extensive literature on issue voting based on the Spatial Theory of Voting has time and again shown that issues do matter (see e.g. Adams et al. 2005;Thurner 2000). But does every party attract identical electoral attention to one and the same issue, given obviously quite different issue-related campaign strategies? Surprisingly, the literature on spatial models has been rather tacit on this topic. A recently growing research community in the area of issue competition highlights that issue-related campaigning includes not only positiontaking but also asymmetric intensity and emphasis in presenting such positions (see De Vries and Hobolt 2012;Green-Pedersen 2007;Guinaudeau and Persico 2014;Rovny 2012;Wagner 2012b). This latter aspect captures the saliency attached by a party to an issue. Previous research on spatial issue voting has exclusively focused on the implications of the differences in parties' position-taking for their market shares and for voters' choices. In contrast, parties' asymmetric emphasis on issues has been at the centre of the Saliency Theory of Voting Farlie 1983a, 1983b;Robertson 1976) and the Issue Ownership approach (Petrocik 1996).
This article contributes to the discussion by offering a new way to systematically integrate the saliency aspect of issue competition with the Spatial Theory of Voting. The Spatial Theory of Voting assumes that the saliencies voters attach to issues are identical across parties. In contrast, the saliency approaches explicitly expect varying issue saliencies at the level of the parties. We propose to relax the Spatial Theory's, mostly implicit, assumption with regard to the equality of voters' reactions towards parties on issues. The empirical examination of this assumption will allow us to assess whether the saliencies voters attach to issues are actually identical across parties. Following the saliency approaches, we agree with the observation that parties strategically attempt to shift voters' attention to those issues that have a higher relevance in their portfolio, and to downplay the saliency of those issues that are of minor importance for their electoral segments. This results in varying issue saliencies at the level of the parties, representing what we call supply-side issue saliency. As a consequence of this, we expect that the saliency attributed by voters to a particular issuewhich we name demand-side issue saliencyalso varies across parties. Therefore, supply-side asymmetry in issue saliency should be reflected in asymmetric demand-side issue saliency, and vice versa.
Based on this argument, the article outlines a straightforward and simple research design for the statistical identification of such party-varying issue saliencies within the established paradigm of the Spatial Theory of Voting and Random Utility Maximisation. To achieve this objective, we will use a special feature of conditional logit and probit modelsi.e. the possibility to estimate party-specific coefficients instead of fixed 'generic' spatial issue distance effects (see Thurner 2000). This will allow us to estimate issue saliencies at the level of the demand-side that potentially differ between parties. It is this feature that enables us to provide a systematic theoretical and empirical bridge between the saliency approaches on the one hand and the Spatial Theory of Voting on the other. Therefore our objective is in line with recent efforts to improve our understanding of parties' issue-related campaign strategies beyond pure position-taking. Additionally, our suggested approach follows the claim that an appropriate model of issue voting should always integrate party strategies and individual vote choice (see e.g. Bélanger and Meguid 2008;Green and Hobolt 2008;van der Brug 2004) and to compare various equilibrium outcomes.
We apply our model of spatial issue voting with party-varying saliencies to the German parliamentary elections from 1987 to 2009. Our results provide strong empirical evidence that issue saliencies at the level of the voters vary considerably among the parties of this multi-party system, indicating that the impact of issues on party choice is different across parties. Therefore, the resulting equilibrium positions clearly deviate from those identified under the current commonly used assumption of identical reactions to issues. In addition, the results support our expectation that it is predominantly parties taking polar positionsand, more specifically, niche parties adopting polar positionsthat attract voters' reactions to issues. This finding clearly demonstrates that the type of party (mainstream vs. niche) moderates the parties' success when engaging in a strategy of issue polarisation in order to become the object of issue voting.
The next section introduces the debate on the role of saliency in models of issue voting. We then derive hypotheses on the occurrence of party-specific issue voting. Subsequently, we illustrate the statistical approach to the estimation of such party-varying issue saliencies at the level of the voters. Finally, we introduce the data and present our findings.

The Role of Saliency in Models of Issue Voting
According to the Downsian perspective of issue voting, formalised in the Spatial Theory of Voting, voters evaluate parties with regard to their differential position-taking on manifold issues and cast a ballot for the party that guarantees the highest expected utility (Downs 1957). As parties in modern complex societies decreasingly succeed in offering consistent one-dimensional ideological packages covered by, for example, the left-right dimension, the identification of the relative impact of separate controversial 'position issues' on electoral success is highly important Alvarez and Nagler 1998;Thurner 2000).
Determining the relative statistical weight assigned to issues in the individual vote decision (demand-side issue saliency) in spatial issue voting models follows established techniques: the electoral researcher calculates distances between each voter's ideal point and the perceived position of each party on an issue. The subsequent estimation of coefficients for these distances yields saliencies expressing the weight of a particular issue for voters' electoral decisions. Alvarez and Nagler (1998) and Thurner (2000) demonstrated that spatial issue voting in multi-party systems can be properly estimated by using socalled conditional logit and probit models. These statistical models allow us to take into account the issue distances for each of the parties. In contrast to the attributes of voters, these variables constitute attributes of choice objects. Based on these models and their analyses of issue voting in the US and European contexts, Adams et al. (2005) are able to show, for example, that there is non-convergent position-taking even in two-party systemscontrary to the Downsian expectation of convergent party platforms. In this approach, voter loyalties (party identification) as well as candidate evaluations are theoretically introduced as non-policy or non-issue factors. They capture segment-specific biases requiring that parties take divergent issue positions.
A careful reading of the contributions by Alvarez and Nagler (1998) and Adams et al. (2005) uncovers a striking assumption, which is actually highly restrictive but has not so far been discussed in detail (for an exception, see Thurner 2000). As a rule, this literature assumes that all voters react identically to the issue positions of all partiesi.e. voters attach issue saliencies that are supposed to be identical for every party. For example, Glasgow (2001: 124) states the following: 'The impact of issue distances is assumed to be constant across all three parties included in the model'. Despite referring to the possibility of specifying issue parameters varying across parties, Alvarez and Nagler (1998: 66-7) also decide to 'estimate only one issue distance parameter for each issue' (see also Adams et al. 2005: 17). We propose to turn this assumption of homogeneity into a hypothesis to be tested empirically. The relaxation of the assumption of equal issue reactions towards all parties allows us to examine whether the saliencies voters attach to issues are actually identical across parties. As a consequence, we will be able to address the following new question: does every party attract identical levels of attentiveness on the same issue? Put differently, is every party equally successful in attracting voters' attention on issues, or are there systematic differences in the impact of issuerelated attitudes on party choice across parties?
The homogeneity assumption in spatial models of issue voting stands in sharp contrast to the insights obtained by the saliency approaches. This perspective explicitly expects varying issue saliencies at the level of the parties (supply-side issue saliency) 1 owing to the parties' incentives to strategically manipulate the saliencies of strategically selected issues Farlie 1983a, 1983b;Petrocik 1996;Robertson 1976;Stokes 1963). According to this view, parties choose to emphasise or to avoid talking about issues as opposed to announcing concrete positions on them. In particular, parties strategically aim to shift voters' attention to those issues that are advantageous to them (i.e. issues that are likely to improve their electoral performance) and to decrease the saliency of those issues that are disadvantageous to them (Budge and Farlie 1983b;Kriesi and Sciarini 2004;Petrocik 1996;Rovny 2012;Wagner 2012b). These party-varying campaign 'investments' in different issues are due to the parties' specific ideological reputations. It is the ideological backgrounds that determine which issues are the parties' 'core issues' (see e.g. recent work done by Rovny 2012). As Petrocik (1996) outlined, the continuous selective emphasis and de-emphasis of respective issues may lead to the ownership of an issue. However, most of the parties officially declare to have many or all issues in their salei.e. they present themselves as political multi-product firms.
The variation of supply-side saliency across both issues and parties is also highlighted by other recent studies. Wagner and Meyer (2014) address the question of whether parties' strategic saliency decisions differ systematically across parties. Their findings clearly demonstrate that parties that have more resources at their disposal are more likely to focus more broadly on those issues that are currently of public concern, therefore pursuing the 'riding the wave' saliency strategy. By contrast, parties possessing fewer resources and a stronger policy-oriented motivation tend to engage in the narrower saliency strategy of issue ownership. Additionally, recent efforts to integrate positional and saliency approaches also provide strong empirical evidence in this regard (see e.g. De Vries and Hobolt 2012;Rovny 2012;Van de Wardt 2014). By highlighting the importance of the multidimensional character of party competition and the parties' varying interest in these different issues, Rovny (2012) demonstrates that parties benefit from both emphasising favoured issues and adopting rather 'blurred' positions on disadvantageous issues. Van de Wardt (2014) shows that parties tend to respond to internal divisions by strategically decreasing the saliency of those issues on which there is great amount of disagreement among their support base.
Based on these recent results and in view of our proposed relaxation of the spatial identity assumption, we argue that the varying issue saliency at the supply-side level should be reflected in asymmetrically attached issue saliencies at the level of the voters, and hence at the demand-side of politics. We expect that voters respond electorally to both position-taking and issue emphasis efforts of the parties. Therefore, voters' issue reactions should also vary across parties. The identification of those party-varying issue saliencies at the level of the voterswhat we call party-specific issue votingwill allow us to improve our knowledge of the connection between parties' issue-related strategic decisions and the individual vote decision based on issues. In particular, our approach permits us to indirectly measure the success of such strategies of position-taking and saliency.

Hypotheses on Party-Varying Issue Voting
If voters' reactions towards issue distances are distinct across parties, the obvious next question is: which parties are more likely to induce such asymmetric voter reactions on specific issues? More specifically, what explains the relative strength of such reactions? Following the recent extensive discussion of the effectiveness of different issue-related strategies in party competition, we propose three hypotheses to explain the occurrence and the differential strength of party-varying issue voting. We expect party-specific issue reactions to be higher (1) for parties taking polar positions, and (2) for niche parties. More specifically, we argue that polarisation strategies actually pay out only for niche parties.
Firstly, we expect that those parties engaging in a strategy of polarisationi.e. parties occupying extreme positions away from the centre of a given issueare more likely to become the object of issue voting and to induce greater demand-side issue saliencies. As shown by Rovny (2012) and Wagner (2012b), parties can raise the saliency of their beneficial issues by offering polar, extreme positions on these issues (see also De Vries 2010;Kriesi and Sciarini 2004;Lachat 2011). By adopting polar stances, parties can use shrill, strident and intransigent simplifications to promote their favoured issues. Often, they set aside details on implementation of policy measures and disregard office-seeking considerations. As a consequence, polarising strategies are easily communicated and get attention from media and voters. In particular, by frequently emphasising their polar stances on selected narrow issues, these parties attract the voters' attention and establish an easy link between the issues and the respective parties. Rabinowitz and Macdonald (1989) in developing their 'directional model' already suggested that parties taking a polar anchor position attract the attention of voters. Whereas these authors propose to 'misuse' the original spatial measurement by adopting a multiplicative calculus on the issue scale, we maintain the spatial logic but nevertheless try to detect the effectiveness of polarisation strategies. This leads to the following hypothesis: H 1 Polar Party Hypothesis: Polarising parties attract relatively higher issue effects as compared to those parties signalling moderate positions.
Secondly, drawing on the extensive debate that focuses on the question of whether different types of parties (mainstream vs. niche) behave differently in campaigns and whether these strategies contribute in a distinguishable way to their electoral performance, we will examine whether this also holds according to our new design. We expect the type of party to moderate the occurrence and relative strength of party-varying issue voting. 2 Previous research on niche and small party behaviour stresses the idea that this type of party follows quite distinct issue-related campaign strategies as compared to their larger counterparts (see e.g. Adams et al. 2006Adams et al. , 2012Ezrow 2008;Ezrow et al. 2011;Meyer and Wagner 2013). In particular, they point out that niche parties primarily aim to pursue policy-seeking goals in order to meet their supporters' policy demands. By contrast, mainstream parties follow vote-maximising and office-seeking strategies. Moreover, niche parties should by definition be more homogeneous internally (activists, members) as well as externally (targeted voter segments). Due to this higher cohesion resulting from their organisational structure, they are able to send less ambivalent issue signals as compared to mainstream parties. They have to make fewer rhetorical compromises when taking up and emphasising issue positions. As a result, issue reactions should be more likely to occur with regard to such parties. In contrast, mainstream parties seek to attract large and more heterogeneous voter segments in order to pursue their vote-or office-seeking incentives. Due to these motivations, mainstream parties may tend to refrain from adopting clear, unambiguous stances on issues on which there may be a high potential for disagreement among their support base. Because of their policy-oriented motivation and their organisational structure, we expect that niche parties induce a greater effect of issues on voting: H 2 Niche Party Hypothesis: Compared to mainstream parties, niche parties induce larger issue voting effects.
In fact, we observe both niche and mainstream parties taking polar positions on issues. Do these polarisation strategies really pay out the same way for both types of parties? Or should we expect, based on the arguments above, that only niche parties really profit from choosing a polarisation strategy? Considering this, our third hypothesis addresses the question of whether the strategic decision of polarisation on an issue is equally promising for mainstream and niche parties. We argue that the success of the polarisation strategy to induce party-specific issue voting is actually moderated by the type of party. Specifically, we expect that particularly niche parties benefit from polarising strategies on issues.
Actually, recent studies of niche party behaviour demonstrate that these parties are especially motivated to take and to accentuate selected extreme positions in order to 'differentiate' themselves from their mainstream counterparts and to occupy a niche to survive electorally (Spoon 2009(Spoon , 2011Wagner 2012b). The 'policy-differentiation' incentive of niche parties leads to the selection of some extreme issue stances. As soon as mainstream parties move to the poleswhich happens particularly in order to satisfy extreme segments and to deter entry of new partiesthis necessarily provokes disagreement from moderate segments of these parties. As a result, a strategy of policy moderation on their core issues should be electorally costly for niche parties, whereas a polarisation strategy should always bear a high risk for mainstream parties to be accused of radicalisation Ezrow 2008;Ezrow et al. 2011). Drawing on these arguments and empirical results, we hypothesise that niche parties benefit from taking polar stances on issues. This leads to our last hypothesis on the party-induced asymmetry of voters' issue reactions: H 3 Niche and Polar Party Hypothesis: Niche parties taking polar positions on an issue attract larger issue effects as compared to mainstream parties taking polar positions.

A Statistical Model of Spatial Issue Voting with Party-Varying Saliencies
Issue voting with party-varying saliencies can be easily identified using established statistical models. We illustrate this new design with an application to the German multi-party system. In accordance with the Spatial Theory of Voting we focus on controversial political issues, so-called 'position issues'. Issue distance coefficients are usually interpreted as the importance or weight voters attribute to issues. This is equivalent to our notion of issue saliency detected at the voter level (demand-side issue saliency). That these saliencies are identical across parties has previously been taken for granted. We propose to scrutinise this non-theorised assertion by turning it into an empirically testable hypothesis. The implementation will be demonstrated in the following. Our model is based on absolute issue distances 3 between the most preferred position of voter i on an issue k, x ik , and the corresponding individually perceived position of party j, p ijk . The disutility function is weighted by the issue-specific saliency coefficient α k : For our new research design we propose to use established statistical models. 4 They allow us to go beyond current practice and to estimate party-specific issue saliencies. We assume that the overall utility of party j for voter i depends on a deterministic component, V ij , and a stochastic component, e ij such that: where e ij follows an iid extreme value distribution. It is assumed that rational voters (i.e. maximising their utility) choose the party in which the difference in the deterministic term exceeds the difference in the stochastic component. Therefore, the probability P ij of voter i choosing party j compared to choosing any other party h is given by: The probabilities reflect different degrees of preferability of parties (Random Utility). According to McFadden (1974) the logit choice probabilities are given by: The deterministic utility component V ij may consist of individual-specific attributes s i (e.g. sex, attitudes) and alternative-specific attributes z ij (e.g. distance of voter position towards party position): With regard to individual characteristics s i , the corresponding coefficients b j indicate varying segment-specific evaluations of parties. On the contrary, note that with regard to alternative-specific attributes z ij , the related coefficients a have no alternative-specific subscript. In the case of so-called 'generic' attribute coefficients, only one parameter for an alternative-specific attribute is estimatedimplying that the attribute is valuated identically with regard to all parties. However, statistically the generic coefficient can be split up into as many alternative-specific coefficients as there are parties to compete. The usage of such alternative-specific coefficients is well known in transportation economics and econometrics, but has to our knowledge not been discussed or used in political science so far. 5 When transferred to the issue voting context, we strongly expect that voters' reactions to issues vary across parties due to the obvious fact that each party strategically manipulates the saliencies of selected issues. For example, one would expect that the distance in the case of the issue of immigration may induce different saliencies or reactions when voters assess far-right parties as compared to more liberal partieseven in the case of identical distances. Furthermore, one would expect that voters react more strongly to positions of Green parties with regard to environmental or energy issues because Green parties claim to 'own' these issues and to hold the highest respective issue competence.
In the following, we formally outline the specification of this model by highlighting the difference between fixed or generic and alternative-specific issue distance parameters. We follow the requirement of subjecting hypotheses of issue voting to a competitive multivariate test (see Adams et al. 2005;Alvarez and Nagler 1998;Thurner 2000) by controlling for party identification and candidate evaluations as the most important non-policy factors. Assume that V ij consists of the following components: (1) a base utility of a party j, 6 constituted by the alternative specific constant β j0 ; (2) the individual-specific evaluation of candidate j (i.e. voter's evaluation of party jth leader) represented by β j Candidate i ; and (3) the perceived issue distance between voter i and party j represented by αIssueDistance ij . Then, the following equation results: The issue distance parameter α captures the estimated issue saliency. As it is statistically fixed and constrained to be identical, the substantial assumption one buys is that the issue saliency is identical for all parties. Provided the generic issue distance coefficient is statistically significant, the previous literature presumeswithout further explicit discussionthat the estimated issue reactivity of the voters does not vary across parties. Moreover, in the case of a statistically insignificant generic issue distance coefficient, the literature so far asserts that there are no issue voting effects with regard to all parties. Both assumptions are not necessarily correct: First, it may be the case that issue reactions with regard to different parties differ markedly, and/or they may exist only for a subset of parties. Second, it is possible that the generic coefficient is not statistically different from zero, and at the same time one or several significant alternative-specific coefficient(s) exist(s) behind the fixed insignificant coefficient. Therefore, we strongly recommend testing these assumptions explicitly and empiricallyby 'splitting' the generic parameter into so-called alternative-specific parameters (i.e. by specifying for each party a specific partial utility function for each kth issue, and therefore by estimating for each party-issue-distance combination a separate parameter).
After splitting the individual-specific candidate evaluations (into J−1 coefficients) and the generic issue distance coefficient into J coefficients, the following alternative-specific, partially observable utility functions for the main German parties result (C: CDU/CSU = Christian Democrats; S: SPD = Social Democrats; F: FDP = Liberal Party; G: Greens; L: Leftist Party): We are now able to ask the following new research question: are voters actually attributing different issue saliencies depending on which party is considered? As a baseline, we formulate the following null hypothesis: H 0 Null Hypothesis: There are no party-specific issue effects, implying that issues are identically valuated with regard to all parties.
Here we have to distinguish between the following situations. (1) There are no issue voting effects with regard to all parties on an issue. Thus, the generic issue distance coefficient should not be statistically different from zero. However, (2) it might be the case that behind a statistically non-significant generic issue distance coefficient there may be 'hidden' significant party-specific issue distance coefficients. If, and only if we observe such party-specific issue voting effects for at least a subset of the parties, this would be a necessary and sufficient indicator that the assumption of identical issue reactions with regard to all parties has to be rejected. As a consequence, this would indicate that not all parties are equally successful in attracting voters by their issue-related strategies in election campaigns.

Data and Operationalisation
Our model of issue voting with party-varying issue saliencies is applied to cross-sectional surveys of the German parliamentary elections of 1987, 1990, 1998, 2002, 2005 and 2009. 7 These studies include questions in which respondents were asked to locate parties' positions and their own positions on several issues 8 on a 7-or 11-point scale. 9 The dependent variable 'vote choice' is operationalised via the stated vote intention. The set of alternatives includes only parties receiving a vote share total of at least 5 per cent in the respective election. These parties are: the Christian Democratic Parties (CDU/CSU), the Social Democratic Party (SPD), the Liberal Party (FPD), the Green Party, andsince election year 1998the Leftist Party (PDS/Leftists). Our key explanatory variables are the absolute distances between the individually perceived positions of the parties and the self-reported positions of respondents on a given issue. 10 To provide a hard test of the role of issue considerations in the individual vote decision we follow the approach of Adams et al. (2005) and control for party identification and candidate evaluations 11 as the most important non-policy factors. Polar parties are defined as those parties that are perceived to take the most extreme positions on an issue while at the same time being significantly differently located from every other party. 12 Based on their relatively smaller issue portfolios, we define the Greens, the Leftists as well as the Liberal Party (FDP) as niche parties. 13

Empirical Results
The presentation of empirical results is divided into two parts: In the first part we test whether voters actually exhibit party-varying issue saliencies. The second part explains the strength of such party-specific saliencies along the proposed hypotheses.

Identifying Party-Varying Issue Saliencies
As a first step of our analysis, we aim to determine whether the saliencies voters attach to issues vary across parties by estimating (nested) conditional logit models 14 with party-specific coefficients for spatial issue distances. By estimating these coefficients for each of the parties separately, we are able to scrutinise the conventional assumption of a constant issue distance parameter. In order to illustrate the difference between a model of issue voting with fixed generic issue distance coefficients and a model with party-varying issue distance coefficients, we contrast the results of these models. Figure 1 depicts and compares the estimated issue distance coefficients with 95 per cent confidence intervals. The left-hand column of Figure 1 shows the generic issue effects, the right-hand column depicts the party-specific issue reactions that are statistically significant at the 5 per cent level. 15 Also note again that we control for party identification 16 and candidate evaluations in the background. As expected, party identification proves to have a strong and statistically significant positive impact for all parties on vote choice. In addition, candidate effects discriminate considerably with regard to their impact on choosing different parties (see Appendix A in the supplemental material).
Focusing on the issue distance coefficients, we highlight the following points resulting from Figure 1. Firstly, 12 out of 17 generic issue distance coefficients are statistically significant indicating that issues are relevant choice criteria, even when controlling for party identification and candidate images. 17 This suggests that in most cases there is a separate issue effect with regard to at least one party on these issues.

SPATIAL ISSUE EFFECTS IN GERMAN PARLIAMENTARY ELECTIONS 1987-2009
Note: Left-hand column displays generic issue distance coefficients; right-hand column shows party-specific issue voting effects significant at the 5 per cent level.
Secondly, by splitting up the fixed generic coefficients into J party-specific issue reactions, we can show that a non-significant generic issue distance parameter does not necessarily imply that there is no impact from such an issue. With regard to the five non-significant generic coefficients (unification and abortion in 1990, immigration and EU in 1998, EU in 2002, we estimate for two of these issues significant separate party-specific effects. Even though the corresponding non-significant generic issue coefficients would suggest that there are no issue effects with regard to all parties, the specification of partyvarying issue saliencies enhances our insight that a non-significant generic coefficient does not necessarily imply that there are no issue reactions at all. With regard to the EU issue in 1998 and 2002, it is indeed the case that behind a statistically non-significant generic issue distance coefficient there are 'hidden' significant party-specific issue reactions. For example, consider this issue in 2002: By de-fixing the generic coefficient and specifying instead party-varying issue saliencies, it can be shown that only for one party we identify a statistically significant issue effect, namely for the Leftist Party. In other words, on this issue only the Leftist Party proves to be a significant object of issue voting. A closer inspection of the perceived party positions on this issue provides a possible reason: this party constitutes the only polar party on this issue (see Table 1). In addition, when examining all electoral manifestos on this issue, it turns out that the Leftist Party unconditionally advocated the acceptance of all European Union accession candidates (see its electoral manifesto 2002: 24)in contrast to the moderate positions of all the other parties. Consequently, the statistically achieved result can be validated externally. In sum, by estimating only one issue distance parameter, we would falsely conclude that the EU issue played no role in this election.
Thirdly, a significant generic coefficient does not incontrovertibly indicate that there are identical issue reactions with regard to all parties. In particular, the examination of Figure 1 shows that for almost all significant generic issue distance coefficients, we identify party-specific effects only for a subset of parties. Exclusively in the case of the issue of unemployment in 1987, we observe for all parties (the PDS was not included in this analysis) separate significant issue effects. 18 Thus, our null hypothesis, stating that issues are identically valuated with regard to all parties, can be rejectedexcept for the issue of unemployment in 1987. In most circumstances, only subsets of issue distance coefficients prove to be statistically significant. This is a non-ambiguous sign that voters' issue reactions indeed vary across parties. Take, for example, the issue of taxes in 2009. All expert observers unanimously agreed after the fact that the Liberal Party (FDP) was most successful in promoting the reduction of taxes and in effectively politicising this issue. Our results support this interpretation. However, our findings additionally suggest that the Leftist Party and the Christian Democratic Parties (CDU/CSU) also effectively played on this ground. The example again demonstrates that not all parties are equally successful in attracting voters by their issue-related strategies in campaigns, and that our research design provides a reliable evidence-based procedure to detect such asymmetries. There is additional evidence that the Greens predominate in politicising issues. For example, in the case of the issue of immigration, the Greens manage in almost all elections to become the object of issue voting. Less counterintuitively, this can also be observed for the issue of nuclear energy. In four out of six possible issue effects, we see a party-specific effect with regard to the Greens.
The identification of these issue effects strongly validates the usefulness of our new approach to estimating party-specific issue reactions. This brief exploration of electoral programmes and campaign statements substantiates that the detection of party-varying issue saliencies at the level of the voters allows us to improve our knowledge of the connection between parties' issue-related strategic behaviour and the individual vote decision based on issues. Our statistical model of issue voting with party-varying saliencies allows us to systematically guide our assessment of the effectiveness of parties' issue-related strategies in election campaigns and offers a new way to combine the insights of saliency approaches with the Spatial Theory of Voting. In sum, 'de-fixing' the generic issue coefficient leads to interesting insights into issue votingwhich vary across elections, issues and parties. In most cases, only some party-specific issue distance parameters turn out to be statistically significant. Thus, the assumption of identical issue reactions with regard to all parties has to be rejected. This implies that not all parties are equally successful in attracting voters based on their issue-related campaign efforts. Additionally, it is now possible to compare the differences between generic versus party-specific coefficients with regard to choice probabilities and market shares. Figure 2 shows how the conditional predicted probabilities of party choice based on distances on the issue of nuclear energy in the 1998 election vary between these two different specification strategies. It demonstrates that, in the worst case, a fixed generic coefficient averages out highly different party-specific saliency parameters. Note also that the 95 per cent confidence intervals of the party-specific reaction functions with regard to the CDU/CSU and the Greens, for which we identify separate party-specific issue effects, do not overlap. Thus, it is also necessary from a statistical point of view to use alternative-specific coefficients. Moreover, based on the party-specific specification of the issue distance coefficients we are able to show that this also leads to quite different Nash Equilibria on an issue dimension. Figure 3 displays the parties' optimal positions, their corresponding vote shares as well as the uncertainty around these positions for the issue of immigration in 1990. 19 The visualisation of the competition on the issue of immigration in 1990 demonstrates that the Green Party, for which we identify a party-specific issue effect, actually attracts voters who would otherwise choose the SPD. Accordingly, the Green Party's vote share increases from 11.8 per cent to 18.4 per cent by moving further to the right and getting closer to the optimal position of the SPD. The Liberal Party (FPD), for which we also estimate a party-specific issue effect, gains additional votes by differentiating itself more from the mainstream party CDU/CSU. Thus, the varying effectiveness of parties' issue-related campaign strategies may also result in very different optimal locations.
For all these reasons, we strongly recommend using this special feature of conditional logit and probit models in future studies of spatial voting in order to avoid averaging out highly different saliency parameters when specifying fixed generic coefficients.

Test of Hypotheses on Party-Varying Issue Voting
The extensive estimation results show that effects of issue voting vary substantially across parties. They corroborate our expectation that voters do not react identically to every party. Whether these findings are completely contingent on a campaign, or whether they follow a general pattern will be tested in this section. We hypothesise that issue reactions are primarily induced (1) by those parties taking polar positions (H 1 ), (2) by niche parties (H 2 ), and more specifically (3) by niche parties taking polar positions (H 3 ) on an issue. Table 1 offers a synoptical summary comparing the observed party-specific issue effects with our expectations. In order to test these hypotheses analytically we propose the following multivariate design. We specify linear regression models including all estimated party-specific issue effects (i.e. all party-issue combinations) for all election years as the dependent variable, generating 79 party-specific issue distance coefficients. 20 We regress these issue effects against the two party types considered (niche party, polar party on an issue). 21 The results are reported in Table 2. In order to test Hypotheses 1-2, Model 1 includes dummy variables for each category of the parties under consideration. Model 2 additionally

Election
Year Issue Unification includes a multiplicative interaction term between polar party and niche party in accordance with Hypothesis 3. As can be seen from Model 1 in Table 2, there are remarkable differences in the impact of issue considerations on party choice when accounting for different types of parties. 22 Firstly, polarising parties attract significantly larger party-specific issue voting effects as compared to those parties signalling moderate positions. Polarising is therefore an effective strategy for politicising voters. Secondly, niche parties exhibit significantly larger party-specific issue voting effects as compared to mainstream parties. To disentangle whether the strategic decision to engage in an issue polarisation strategy is equally promising for niche and mainstream parties, Model 2 in Table 2 additionally includes a multiplicative interaction term between polar party and niche party. The estimated coefficient for the interaction term is large and statistically significant, indicating that niche party status indeed moderates the effect of polar positiontaking strategies on party-specific issue voting. According to our expectations, niche parties taking polar positions on an issue induce higher party-specific issue voting effects as compared to other parties. Niche parties clearly benefit from polarising position-taking strategies, whereas for mainstream parties such voter reactions are not visible. 23 The following interpretation of the interactive relationship between niche party and polar party type on party-specific issue voting in Table 3 is based on conditional incremental effects (see e.g. Berry et al. 2012;Brambor et al. 2006). As can be immediately seen, the effect of polarising on issues for mainstream parties is very small and insignificant. By contrast, the effect of polarising strategies for niche parties is large and highly statistically significant; therefore, as expected, niche parties taking polar positions induce party-specific issue voting to a much higher degree than mainstream parties. As suggested by Berry et al. (2012), we also offer in Table 3 predictions about how the effect of niche party status differs with moderate/polar position-taking strategies, resulting in additional support for our expectations.  In summary, the results of our analysis provide clear evidence in favour of both the Polar Party Hypothesis and the Niche Party Hypothesis. By recombining the controversial discussion on the operationalisation of niche parties and specifying an interaction term between niche party status and the polarisation efforts on an issue it becomes possible to provide a clear answer: Polarising niche parties become successful issue owners; moderating office-seeking niche parties are not rewarded.

Conclusion
Not all parties are equally effective and successful in attracting electoral responsiveness on the same issues. Issue voting substantially varies across parties. Thus, there are systematic differences in the impact of issue attitudes on party choice across parties. This is one major finding of this study. We outlined a new research design for the statistical identification of such party-varying issue reactions within the established paradigm of the Spatial Theory of Voting. In order to demonstrate this, we proposed and applied a previously neglected feature of conditional logit and probit modelsi.e. the estimation of partyspecific coefficients instead of fixed 'generic' issue distance effects. Our proposed research design and empirical findings in the context of the German multi-party system will have considerable implications for future theoretical and empirical analyses of spatial party competition because they challenge the conventional assumption of identical issue reactions. Hitherto, the literature oriented towards the classical spatial model, towards the directional model (Rabinowitz and Macdonald 1989), or towards the unified approach Thurner 2000) has never empirically tested the underlying assumption of models with generic issue distance effectsnamely that voters react identically towards the issue positions of all parties. Our statistical model of issue voting with party-varying saliencies allows us to improve our knowledge of the connection between parties' issue-related strategic behaviour and the

Incremental Effect of Niche Party on
Moderate Party −0.09 (0.06) Polar Party −0.29*** (0.08) Standard errors in parentheses; *p < 0.05; **p < 0.01; ***p < 0.001. Notes: Conditional incremental effects of polar party and niche party type on the expected value of the party-specific issue effects are based on Model 2 in Table 2. They are obtained by taking the arithmetic differences and show the changes in party-specific issue effects for a unit change in these discrete predictors.
individual vote decision based on issues. The proposed research design provides a new way to transfer insights of saliency approaches to the Spatial Theory of Voting. Parties not only take and shift positions; they also manipulate saliencies in order to politicise issues. Finally, by corroborating the expectation that especially niche parties engaging in polarising position-taking strategies induce such asymmetric issue voting, our analyses also have relevance for studies on niche party behaviour. The next steps will be to extend this design to a cross-national perspective in order to find out whether this relationship holds more generally in different contexts.
20. Even though the identification of party-specific issue voting in the previous section was based on the significance of the estimated party-specific effects, we test the hypotheses by using the strength of these effects (i.e. the effect size) as the dependent variable because this design allows us to detect differences in the impact of issue considerations across parties. 21. Note that these models are based on estimates of the individual-level (nested) conditional logit models, estimated separately for each election year. Therefore, the dependent variable represents an estimated dependent variable (EDV), for which, in general, the regression residuals tend to be heteroscedastic when the sampling variance varies across observations. Since sampling size does not greatly vary across election years, it is not necessary here to account for differences in the standard deviations of the issue distance coefficients by applying weighted least squares or alternative Feasible Generalised Least Squares (FGLS) approaches (see e.g. Lewis and Linzer 2005). 22. Since the estimated party-specific issue distance parameters indicate utility losses (see Equation (1)), the coefficients for polar and niche party type are also negative. They show how the increase in utility losses differs across different types of parties. 23. Since the Greens seem to predominate in politicising issues, one might argue that the results about niche and polar parties particularly depend on the Greens. To address this concern, we further inspected the robustness of our finding by excluding the party-specific issue effects with regard to the Greens in the analyses presented in Table 2. These analyses yield comparable results, indicating that even when excluding the Greens from the models testing the stated hypotheses our main arguments hold. The details are available in Table B16 in Appendix B.3 in the supplemental data. We gratefully thank one anonymous reviewer for pointing this out.