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

Abstract

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.

Acknowledgements

We would like to thank Jim Adams, William Greene, Samuel Merrill III, Bernard Grofman, and the anonymous reviewers for their valuable comments.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Supplemental Data

Supplemental data for this article can be accessed at http://dx.doi.org/10.1080/01402382.2015.1026562.

Notes

1. Since the location of saliency here is at the supply-side of politics, empirical strategies to identify such saliencies have mainly used party manifestos as exemplified in the Comparative Manifesto Project (CMP). For a recent assessment see Dolezal et al. (2014).

2. According to Meguid (2005, 2008), niche parties are defined as those parties seeking to emphasise new issues not captured by the traditional Downsian left–right cleavage, and these parties contest only on a selected subset of issues. However, note that Adams et al. (2006) and Ezrow (2008) conceive of niche parties as taking extreme ‘non-centrist’ positions on the traditional left–right cleavage. These definitions are critically discussed by Meyer and Miller (2015) or Wagner (2012a).

3. By using linear instead of Euclidean squared loss functions, we follow recent insights by Singh (2014), who offers both theoretical arguments and empirical evidence in favour of the simple linear loss function.

4. See McFadden (1974) and Ben-Akiva and Lerman (1985) or Glasgow and Alvarez (2008); for applications in electoral research see Alvarez and Nagler (1998) or Thurner (2000).

5. See Ben-Akiva and Lerman (1985: 168): ‘An important aspect of the specification of discrete choice models is the distinction between alternative-specific and generic attributes. A generic specification imposes restrictions of equality on a more general model with alternative-specific attributes’.

6. This base utility contains the unmeasured utility components – e.g. the party label.

7. There are no position issues included in the 1994 National Election Survey. For the 1990 election we use only the survey conducted in West Germany in order to stick to a coherent design for the whole period. A description of the national election studies and variables is provided on request.

8. The following issues are included: European Union (1998, 2002, 2005), immigration (1990, 1998, 2002, 2009), unification (1990), nuclear energy (1987–2009), abortion (1990), unemployment (1987), taxes (2009).

9. To ensure that the issue effects are comparable over time, we standardised the issue scales before estimating the models.

10. Since the prevalent specification of utility losses in empirical voting models is based on quadratic loss functions, we re-estimated all the models below by specifying quadratic utility losses instead of linear ones. These analyses yield almost identical results and are available in Appendix B.1 in the supplemental material. Additionally, the linear specification improves the model fit and outperforms the quadratic specification (see also Singh 2014).

11. The National Election Surveys include questions in which respondents were asked to use 11-point scales to evaluate the party leaders. To keep the model parsimonious and to ensure comparability across time, we only include the candidate evaluations of the two large parties CDU/CSU and SPD.

12. In most cases, the means of the parties’ positions are significantly different at the 5 per cent significance level. For the issue of unification in 1990 the perceived party positions do not statistically differ from each other and no polar parties could be defined. For the issue of unemployment in 1987 and the issue of the EU in 2002, we are only able to identify one polar party on one side of the issue scales. For each of the remaining 14 issues, we have been able to exactly identify one polar party on both sides of the scales (see Table 1).

13. Note that this classification does not perfectly match the definitions suggested by, for example, Meguid (2005, 2008) or Adams et al. (2006). Since it is difficult to apply these definitions to the German multi-party system when including only parties having a total vote share of at least 5 per cent (the remaining parties cannot be included in the analysis due to extremely small number of observations), we rely only on these parties.

14. Logit models are based on the assumption of Independence from Irrelevant Alternatives (IIA). Since Hausman–McFadden specification tests indicate that the IIA assumption is violated in several cases, we captured potential unobserved similarities between subsets of alternatives by estimating two-level nested conditional logit models (see McFadden 1984; Thurner 2000). Where necessary, we partitioned the choice sets into two clusters including the ‘large parties’ (CDU/CSU and SPD) and the ‘small parties’ (Greens, FPD, and the Leftists), respectively.

15. To facilitate readability of important party-specific issue reactions, we report only those party-specific coefficients that are significant at the 5 per cent level, whereas we display all generic coefficients. Appendix A in the supplemental material provides the tabled estimation results presenting all variables and parameter estimates.

16. Since party identification is also a relational construct (i.e. voter i has a loyalty towards a party j, or not), it presents an alternative-specific variable and can also be specified with J alternative-specific effects. It is, therefore, additionally possible to test whether party identifications (PI) vary across parties. The literature on party identification has, to our knowledge, not yet theorised the possibility of a party-varying PI explicitly – despite this being not counterintuitive.

17. Since party identification and candidate evaluations might be controversial in European politics, we assessed the robustness of our findings by excluding these variables and considering instead the meta issue Left–Right and socio-economic variables as controls. Even though omitting these variables naturally leads to ‘overestimation’ of party-specific effects, our main arguments and conclusions hold (see Appendix B.2 in the supplemental material).

18. Note that we performed a Likelihood-Ratio test to examine whether these party-specific issue coefficients are equal. The result indicates that in this case splitting up the generic coefficient is not necessary and that one fixed issue coefficient is sufficient.

19. Nash Equilibria are calculated with the R Package ‘nopp: Nash Optimal Party Positions’ by Curini and Iacus (2012), which implements the iterative algorithm by Merrill and Adams (2001) and Adams et al. (2005). In order to assess the uncertainty, we used bootstrapping with 1,000 replications.

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.