Government termination in multilevel settings. How party congruence affects the survival of sub-national governments in Germany and Spain

ABSTRACT

Past studies on government survival in parliamentary democracies have been limited to national governments. However, most societies live in a multilevel polity where different policies are decided at distinct governmental layers. So far, the conditions triggering sub-national governments’ termination have remained unexplored. Our paper makes a twofold contribution to the literature. First, we explicitly focus our analysis on the sub-national government level. Second, we expand the analytical scope by assuming a multilevel setting, in which the survival of sub-national governments is dependent on both the party composition of the national government (vertical congruence) and their sub-national peers (horizontal congruence). We test the impact of both congruence measures on the early termination risk of regional governments. Our analysis is complemented by including “traditional” factors from national government termination literature, such as structural attributes of governments and their bargaining environment, into empirical modelling. Analysing a novel dataset on 494 regional governments in Germany and Spain we find that the risk of sub-national government termination is related to varying levels of vertical congruence. Furthermore, we find interesting explanatory variation between the two countries with regard to the effect size of economic performance, regional authority and congruence.

This article is part of a series including:

Party competition and dual accountability in multi-level systems

Party competition and dual accountability in multi-level systems

Disclosure Statement

No potential conflict of interest was reported by the authors.

Notes on contributors

Javier Martínez-Cantó is a PhD candidate at the Bamberg Graduate School of Social Sciences (University of Bamberg). His research interests are party organization, candidate selection, legislative behaviour and federalism. His work has been published in Regional and Federal Studies.

Henning Bergmann is a doctoral candidate at the Bamberg Graduate School of Social Sciences (University of Bamberg). His research interests include coalitions with a focus on cabinet stability and the study of roll call votes in the German Bundestag. His research has been published in the British Journal of Political Science.

Notes

1 Each federal or regionalized country tends to call its subnational units in a different manner, like States in the US, Provinces in Canada and Argentina, Autonomous Communities in Spain, regions in France and Italy, or Länder in Germany and Austria. However, for the sake of simplicity, we will refer to them as “states”.

2 We assume that the formation of both state and federal governments will be influenced by the current state of alliances at both levels, as the state of the art literature does. However, we do not theorize about possible strategical behaviour lagged in time, such as the formation of a state coalition anticipating a change on the federal government’s composition.

3 Here we use a simplified version of Lupia and Strøm’s (1995) equation, as parties compare the current utility of being in government with the prospective utilities of a non-electoral replacement or a new election. Each of these include transaction costs, as well as uncertainty measures, that affect parties’ utilities, that we have left out of our explanation for the sake of parsimony, and because our research does not distinguish between different types of early terminations.

4 In order to measure how many powers do sub-national entities have, we use the Regional Authority Index (Hooghe et al. 2016), which depicts the constitutional powers a region has regarding self-rule and shared-rule characteristics. Self-rule depicts how many policies a region can decide and implement without the acquiescence of the national government, while shared-rule depicts the degree of regional government’s involvement in national policy-making.

5 The sources of our database are listed in Appendix A. Additionally, Appendix B offers summary statistics and correlation matrices of all variables used in our models.

6 We excluded post-WWII unity governments in Germany as well as care-taker governments in both countries from our analysis.

7 We do not apply a “competing risks” design that is widely used in the national government termination literature and differentiates between early elections and cabinet replacements for two reasons. First, we do not theoretically assume that our main independent variables should have different effects on the risk of early elections compared to cabinet replacements. Second, early elections are relatively rare in our sub-national sample compared to national level datasets. This fact might cause problems in model estimations as the number of failures to be explained is similar to the number of independent variables introduced into the models.

8 Due to the nonexistence of nuanced data and in order to account for the whole period, we hand-coded cohesive governments as those formed by parties ideologically adjacent on a unidimensional left-right dimension. Appendix C shows a comparison of our hand-coded measure with available data for shorter periods of time, and a replication of our results.

9 This model is suitable as all time-varying covariates fulfill the proportional hazards assumption.

10 We also performed analyses with a shared frailty design to address the potential problem that our observations (governments) within a group (region) might be correlated. Theta, the indicator of the degree of within group-correlation, is insignificant in all models, which indicates that shared frailties do not provide added value compared to our specification.

11 All figures are generated with the Stata scheme plotplain (Bischof 2017).

12 It is possible to argue that in Germany more populated regions may have a higher impact on the rest than less populated ones due to the voting composition of the Bundesrat. Therefore, we calculated the region’s horizontal congruence weighted by the percentage of votes in the Bundesrat. However, the results do not substantially change from one specification to the other.

Additional information

Funding

This work was supported by Deutsche Forschungsgemeinschaft [grant number GSC1024].