![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
Previous research has shown that in addition to party identification chancellor candidate evaluations play an important role in determining vote choice in German federal elections. In this article, we evaluate the impact that such evaluations had on vote choice in 2009, 2013, and 2017 using the German Longitudinal Election Study. In contrast to a popular narrative that Chancellor Merkel was less of a factor in 2017 than she had been in the two previous elections where the Union parties ran especially personalised campaigns around Merkel, we find that the Kanzlerbonus (incumbency benefit) and Merkel’s attractiveness to voters was strong in all three elections, especially among non-party identifiers. More importantly, we find that increasingly the Social Democrats had chancellor candidates unattractive to voters and perceived as bland across all of these elections, and demonstrate that this was a factor which significantly weakened their vote. Thus we conclude that beyond the structural weaknesses of the Social Democrats over the last three election cycles the lack of appeal of the party’s chancellor candidates has played a significant role in the SPD’s lack of electoral success.
About the Authors
Michael A. Hansen (corresponding author) is an Assistant Professor in the Department of Politics, Philosophy, and Law at the University of Wisconsin – Parkside. His research focuses on parties of the radical right, gender politics, and political behaviour in the U.S. and Europe.
Jonathan Olsen is professor and chair of the Department of History and Government, Texas Woman's University. His research focuses on parties of the radical right and radical left in Germany, Europe, and the European Union.
SUPPLEMENTAL DATA AND RESEARCH MATERIALS
Supplemental data for this article can be accessed on the Taylor & Francis website, https://doi.org/10.1080/09644008.2019.1669020.
ORCID
Michael A. Hansen http://orcid.org/0000-0001-5116-5751
Notes
1 The data used for the empirical analysis are the German Longitudinal Election Studies (GLES) conducted in 2009, 2013, and 2017. The post election studies were utilised for the study since the dependent variables are vote choice.
2 The GLES is the only comprehensive survey that measures German citizens’ attitudes towards federal election. The years 2009, 2013, and 2017 were chosen because these are the only years that the survey was conducted. Of course, we would prefer panel data and a longer period of time. However, the GLES study is the most useful, complete data for studying the topic.
3 Relevant models were estimated using post-stratification survey weights.
4 In the German electoral system parties may earn more seats than they are entitled to by the second vote (PR) if their number of first vote seats won in a state exceeds the number of seats determined through the second vote. This is the so-called ‘overhang mandate.’ Following a German Constitutional Court ruling and a new electoral law, overhang mandates are now compensated through ‘Ausgleichsmandaten’, which proportionally compensate other parties without overhang mandates so that the seats earned through the election remain proportional.
5 Why are both votes explored? Even though the first vote is for a particular candidate, and that candidate might not be the chancellor candidate, studies have solidly demonstrated that party leadership is utilised as a vote heuristic when casting a vote. Therefore, we have to account for the fact that the chancellor candidate evaluations might impact individual vote for other members of the same party.
6 The inclusion of each individual candidate evaluation item in a vote choice model is problematic for model estimation due to multi-colinearity since the individual items are highly correlated. The lowest bivariate correlation between individual items was .34. The highest correlation was .756. On average, estimating a correlation matrix between the items, the items correlated at .59 or higher. Also, see model selection critique in Appendix B.
7 We must note that the number of observations in 2009 and 2013 are lower than one would hope when estimating binary choice models. In order to explore the robustness of the results, Bayesian binary models were estimated as a potential check using Markov Chain Monte Carlo Simulation (MCMC). The prior means and variance were specified in a number of ways. Overall, the results from the Bayesian binary models indicated that the results were fairly robust.
8 Variable coding and any statistical tests performed for creating the independent variables (i.e. Cronbach’s Alpha scores, factor analysis, and binary correlation tests) are provided in the appendices.
9 In order for models to be comparable across elections, model specification must be identical (i.e. independent and control variables). Therefore, the inclusion of election specific issue questions is avoided, since the inclusion in one election and not another would not allow for cross election comparison.
10 The Party Manifesto ideology scores (rile) indicate that the SPD has remained ideologically consistent, while the CDU/CSU has moved around (Volkens et al. Citation2018). In 2009, SDP was at −18.297 and CDU/CSU was at 8.724. In 2013, the SPD was at −23.568 and CDU/CSU was at 2.564. Finally, in 2017 the SPD was at −21.437 and CDU/CSU was at 2.757. The statistics would indicate that in 2009 the CDU/CSU was its most conservative out of the three elections and was fairly centre right in 2013 and 2017. On the other hand, the SPD has been fairly solid left in all three elections.
11 Appendix B provides a discussion on the independent variables that were excluded from the final analysis.
12 Full candidate evaluation descriptive statistics are provided in Appendix C.
13 It is worth noting that in Supplemental Appendix D (Tables D1–D4) output is provided where linear regression models are estimated in order to predict respondents’ evaluations on each of the four candidate characteristics in isolation. Importantly, similar trends are found when predicting each of the four candidate characteristics that are found when predicting the singular, latent candidate characteristic measure.
14 In Appendix F, we include output from models that estimate the impact that ideological distance between the respondent and the candidate has on candidate evaluations. Taking the respondent’s ideological positioning and subtracting the position from where they placed each candidate created each measure. Then, we standardised the distance. The findings indicate that respondents evaluate the candidate lower as the ideological distance between the respondent and the candidate increases.
15 It is worth noting that the AfD was not included as a category for the party ID variable or vote choice dependent variables in 2009 and 2013 because there were not enough respondents indicating that they either identified or voted for the AfD in order to include in the multivariate models.