![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
To what extent does the federal political arena contaminate the regional one in Germany? Does a party’s position as government or opposition on the federal level have a systematic impact on its performance in Land elections? Land elections are often characterised as second order elections, but existing empirical studies that use real election data suffer from important methodological problems. Unlike previous approaches using survey data or comparing vote shares in regional and federal elections, we analyse contamination in two ways. First, we test whether a party’s role at the federal level has a systematic impact on gaining or losing office at the Land level. Second, we examine the vote difference of parties relative to their result in the previous election in the Land. Drawing on a complete dataset of all Land elections from 1949 to 2017, we find confirmation for two phenomena well known in comparative electoral studies. First, the anti-incumbency effect: government parties tend to lose votes. In the German context, as in many other multilevel systems, this is exacerbated by the second effect: contamination. Gaining power or votes on the Land level is very difficult when a party is in government on the federal level.
ABOUT THE AUTHORS
Christopher Baethge, MD, is Professor of Psychiatry at the Department of Psychiatry and Psychotherapy of Cologne University Medical School where he specialises in affective disorders. His research topics include uni-, bipolar and schizoaffective disorders as well as meta-research on medical publications. He is involved in the production of clinical guidelines (e.g. for bipolar disorders) and serves as a member of the German Drug Commission. Christopher Baethge is Chief Scientific Editor of Deutsches Ärzteblatt, the journal of the German Medical Association. In that capacity, he is also a member of the International Committee of Medical Journal Editors, ICMJE.
Mirko Dallendörfer, MA in Economics, University of Cologne, is a research assistant at the Cologne Center for Comparative Politics.
André Kaiser is Professor of Comparative Politics at the Cologne Center for Comparative Politics, University of Cologne and Faculty in the International Max Planck Research School ‘The Social and Political Constitution of the Economy’ and the Cologne Graduate School in Management, Economics and Social Sciences. His recent publications include ‘Policymaking in Multilevel Systems. Federalism, Decentralization, and Performance in the OECD Countries’ (ECPR Press 2013, with Jan Biela and Annika Hennl).and articles, among others, in American Journal of Political Science, Comparative Political Studies, Electoral Studies, European Journal of Political Research, European Union Politics, Journal of Legislative Studies, Journal of Theoretical Politics, Party Politics, Political Studies, and West European Politics. His research focuses on the relevance of institutions for political action. For further information see http://www.cccp.uni-koeln.de/en/public/team/professuren/kaiser1/.
ORCID
Mirko Dallendörfer http://orcid.org/0000-0001-8653-0165
Notes
1 The early election in Lower Saxony on 15 October 2017 is not included. There are a number of reasons why we think an update of the original article is a valuable addition to research on multi-level voting in Germany. First, the original article was published in German. Second, it is now much more feasible to compare the timeframes before and after 1990 as more observations have been added. Third, an analysis of the effects of federal-level grand coalitions is also more feasible now as another period of grand coalition government has been added to the dataset. Fourth, we had the opportunity to screen the database by comparing with different sources and thus correct errors in the original database previously used. The dataset is available online at https://www.cccp.uni-koeln.de/de/research/data/.
2 Take for example a simple model with two parties, where the vote shares for party X (e.g. the CDU) are independently normally distributed around mean 50 with a standard deviation of 5 per cent points in each Land and federal election. The party Y (e.g. SPD) receives the rest of the vote share 100-x. The party receiving more than 50 per cent forms the federal government. A simple Monte Carlo simulation of 100 pairs of simulated elections, in which the difference between Land and preceding federal election results of party X is regressed on X’s federal incumbency status yields a significantly negative coefficient for federal government status of −7.98 (when averaging the coefficients across 100,000 simulation runs). Compare this to Lohmann, Brady, and Rivers’s (Citation1997) results who find a coefficient of −7.76 for incumbency status of the CDU and of −6.74 for the SPD between 1960 and 1989 to see that a random process could explain such results.
3 Spearman’s rank correlation coefficient between change in vote shares relative to the previous election and change in government (coded −1 = loss of power, 0 = no change, 1 = gain of power) in our dataset is r = 0.2386 for the CDU, p = 0.2832 for the SPD, r = 0.3747 for the FDP and r = 0.1879 for the Green Party.
4 We have updated Baethge’s (Citation2011) original dataset and cross-checked the data with Schakel’s (Citation2013) data on regional election results and Röth and Kaiser’s (Citation2018) dataset of regional government composition and change.
5 By design, this study can only reveal statistical associations and no causal relationships. Following common usage in political science, however, and since we are convinced that our findings point to a causal link, in this research note we write ‘effect’ rather than ‘association’.
6 All calculations were carried out with STATA 12.
7 Strictly speaking, under the operationalisation of gaining government power applied here, parties may also gain power as part of government, but only in the rare case of attaining the Land prime ministership after having been a junior coalition partner in the previous government.
8 Note that an absolute average loss of 0.65 per cent points for the Greens does not contradict an average relative gain of 2.89 per cent of its size. This is because the arithmetic means of percentage gains and percentage losses over time are weighted differently. For example, an increase of one percentage point from a vote share of 5 per cent and a subsequent loss of one percentage point by a party in the same Land would lead to an average absolute change of 0, but an average relative increase of 1.67 per cent of its size. We are, however, calculating averages not only across time but also across Länder.
9 When looking at the overall time period from 1949 to 2017 the CDU/CSU and the Greens show a significant negative correlation with the time since the last federal election when in opposition, while the FDP shows a significant positive correlation when in opposition and a significant negative correlation when in government.
10 The differences between Stadtstaaten and Flächenstaaten before 1990 are not statistically significant at a 5 per cent level according to a two sided t-test without equal variance assumption for either government status (p = 0.2088) or opposition status (0.3464).
11 The differences between Stadtstaaten and Flächenstaaten after 1990 are not statistically significant at a 5 per cent level according to a two sided t-test for equal means without equal variance assumption for either government status (p = 0.2309) or opposition status (p = 0.4028).
12 The difference between grand coalition governments and non-grand coalition governments for all government parties is statistically significant at a 5 per cent level according to a two-sided t-test without equal variance assumption (p = 0.0083).
13 The difference between opposition to a grand coalition and non-grand coalition government for all opposition parties is not statistically significant at a 5 per cent level according to a two-sided t-test without equal variance assumption (p = 0.3175).
14 The recent success of the right-wing populist Alternative für Deutschland (AfD), for example, contributes to this phenomenon. Dissatisfaction with grand coalition governments in which the need to compromise is stronger than in a CDU, CDU-FDP or SPD-Greens government and differences between the two major parties become blurred thus likely leads to increased protest votes outside of the established party spectrum. We additionally tested the effect of grand coalition formations on the electoral strength of new parties in a multivariate regression including different government formats on the federal and Land level, time since the last federal election and Land fixed effects. We find that a grand coalition format clearly has a strong effect.
15 The difference between being in a grand coalition government and a non-grand-coalition government is not statistically significant at a 5 per cent level for the CDU according to a two-sided t-test without equal variance assumption (p = 0.2585).
16 The difference between being in a grand coalition government and a non-grand-coalition government is not statistically significant at a 5 per cent level for the SPD according to a two-sided t-test without equal variance assumption (p = 0.1713). A one-sided t-test (when excluding a priori the possibility that the SPD could benefit from being in a grand coalition) would yield statistical significance only at a 10 per cent level (p = 0.0857).
17 The difference between being in opposition to a grand coalition and non-grand-coalition government is not statistically significant at a 5 per cent level for the FDP according to a two-sided t-test without equal variance assumption (p = 0.0780), a one-sided t-test (when excluding a priori the possibility that being in the opposition to a grand coalition may hurt the FDP, (p = 0.0390) shows statistical significance, however).
18 The difference between being in opposition to a grand coalition and non-grand-coalition government is not statistically significant at a 5 per cent level for the Greens according to a two-sided t-test without equal variance assumption (p = 0.1043).
19 The difference between being in opposition to a grand coalition and non-grand-coalition government is not statistically significant at a 5 per cent level for the Linke according to a two-sided t-test without equal variance assumption (p = 0.6875).
20 All significance testing has to be viewed as exploratory, not confirmatory.
21 We find confirmation for this in our multivariate regression although one has to keep in mind that it could also simply be a regression to the mean phenomenon.
22 This is the case even though we do not find strong statistical evidence for it when checking for individual parties because the number of observations is too small.