Explaining Government Formation in Multi-level Settings: Coalition Theory Revisited—Evidence from the Spanish Case

Abstract

This article argues that regional government formation in decentralized countries follows different rules than national government formation in unitary states. It revises some basic assumptions that classical coalition formation theory makes, positing that in multi-level settings parties do not behave as unitary actors, that the goals they pursue might vary across levels at any given time, that regional coalition formation is part of a two-level game and that the policy space in which coalitions are mapped is often two-dimensional. Employing a combination of quantitative and qualitative techniques, several classical propositions are tested in light of these revised assumptions on data about Spanish regional governments. We find that classical predictors do their fair share, but multi-level factors are crucial in explaining the making and breaking of regional governments.

Keywords:

• Coalition theory
• regional government formation
• political parties
• mixed-methods design
• Spain

Acknowledgement

This research was performed as part of the Interuniversity Attraction Poles financed by the Belgian government. The author is grateful to Richard Katz, Mariano Torcal, Kerstin Hamann, Bonnie Field, Josep Colomer, Patrick Dumont, Marc Debus, Uwe Jun, Kris Deschouwer and Tània Verge for comments on earlier versions. All remaining errors are the author's.

Notes

The core party is the party whose policy position in this multi-dimensional space lies within the area determined by the intersection of all dimension-by-dimension medians. Unlike the simple median, the core policy space is rarely occupied by an existing party (see Schofield and Sened, 2005).

The selection is based on de Winter and Dumont's (2006) review of the state of the art in coalition theory and on Martin and Stevenson (2001). Due to space limitations only those propositions related from the office and the policy schools are addressed here.

Take, for example, the situation of a five-party legislature in which parties are placed as follows. On Dimension 1, A–B–C–D–E, with E quite far removed from the rest of the parties towards the right-end of the policy axis. On Dimension 2, the order of the parties is A–E–C–D–B, this time with B quite far removed from the rest of the parties towards the higher-end of the policy axes. In this situation, if both axes are relevant for coalition formation, parties B and E will be excluded because of their radical position on the second and first dimension, respectively. If we further suppose that C and D are far from commanding a majority of seats together, it follows that the only possible alternative left is a coalition between A, C and D, which is disconnected (i.e. it includes non-adjacent parties) on both dimensions.

The only combinations that are excluded from the start are those that contain the anti-system Unity of the People (Herri Batasuna) and We, the Basque Citizens (Euskal Herritarrok) parties in the Basque Country and in Navarra. The Communist Party of the Basque Territories (Partido Comunista de las Tierras Vascas) is also excluded, being a highly improbable coalition ally.

Between 1982 and 2005, if we exclude the situations of one party enjoying a parliamentary majority status on its own, a total of 84 minority or coalition cabinets were formed at the regional level in Spain. Some of these governments are unfortunately excluded from the analysis as the corresponding data about party policy positions at the regional level are missing. I argue that valid conclusions can be drawn even if not all Spanish regional governments are included—generally data is missing only for those governments which have formed during the 1980s, that is, at the beginning of the analyzed period, when the party system was still in flux.

This type of data is collected only in Galicia, the Basque Country and Catalonia. Respondents are asked to place parties on a scale ranging from minimum minority nationalism (1) to maximum minority nationalism (10). As Navarra is also a case in which this second dimension plays an important structuring role in party competition and government formation and as, with the exception of the CDN, all other parties that compete in Navarra also compete in the Basque Country, scores for Navarra were generated on this second dimension based on data from the Basque Country. As previous research shows that it is only a matter of degree that separates the positions of CDN and the Union of the People of Navarra (Union del Pueblo Navarro—UPN), these two parties were allocated the average scores of the Popular Party (Partido Popular—PP) from the Basque surveys.

Available from http://www.sfu.ca/∼warwick/program/.

In this model, the predicted probability of observing alternative m on the dependent variable y is:

where Z_im contains the values of the independent variable for alternative m for observation i (Long and Freese, 2006: 297).

I use the procedure developed by Martin and Stevenson (2001: 39), which drops a random 10% of the alternatives (with the exception of the government that eventually formed) and produces the average Hausman test statistic over 20 iterations. Obtaining an average test statistic <0.05 indicates that the IIA assumption is problematic. Average test statistics for the IIA assumption are reported in each table. All models also correct for heteroskedasticity by using robust standard errors (see Bäck, 2008).

However, one should be cautious with any absolute interpretations of pseudo-R²s, as they tend to be unreliable estimates of model fit (Long and Freese, 2006). The McFadden pseudo-R² reported here is intended to estimate the relative improvements in the various models’ predicting power.

This time the variable used for measuring the policy distance between the two most remote parties in the coalition was weighted by the salience that parties attach to each policy dimension. The salience scores were estimated by the author and the salience-weighted measure developed in Horizons 3D. Details are available from the author upon request.

Model 4B retains only those variables that were statistically significant in Model 4A and adds variables pertaining to the territorial policy dimension. The two models also differ in the specification of the minimal range variable—the latter includes a saliency weight in the calculation of policy distances, while the former does not.

For an extended discussion of coalition congruence in Spain, see Ştefuriuc (2009).

All interviews were carried out between April and June 2006. The anonymity of interviewees is preserved, as many of them hold public functions of high visibility.

The Spanish constitution states that “the state should guarantee the effective application of the solidarity principle … so that differences between their statutes of autonomy may in no case imply economic or social privileges” (Moreno, 2001: 99).

Following national elections held in March 2008, the incumbent governing party maintained itself in office, albeit again in a minority situation. The seat distribution made it impossible to continue the preferential alliance with ERC, as the latter became an insignificant player in the Spanish parliament. Talks were held instead with the CiU, which holds ten seats and the Basque Nationalist Party (Partido Nacionalista Vasco—PNV), which holds six seats. Ultimately, however, no definitive agreement was struck with any individual party. The government is negotiating parliamentary support on an issue-by-issue basis with the opposition parties in the parliament.