Do Voters’ Coalition Preferences Affect Government Formation?

Abstract

Studies on coalition formation assume that political parties have two major goals: they aim to maximise office and policy payoffs. This paper shows that decision-making in the government formation game is also determined by the voters’ coalition preferences. Since the coalition formation process is not a one-shot game, parties have to take the coalition preferences of the electorate into account when they evaluate the utility of potential coalitions. If parties fail to comply with the coalition preferences of voters, they are likely to be penalised in future elections. The argument is tested by an analysis of government formation in the 16 German states between 1990 and 2009. The results support the argument: the formation of coalitions – at least in the German states – is not only determined by office- and policy-seeking behaviour of political parties, but also by the preferences of voters regarding their preferred outcome of the coalition game.

Notes

1. In countries that are mostly governed by coalitions, such as Austria, Germany, Israel and the Netherlands, it is quite common for information regarding voters’ coalition preferences to be published and discussed by the media (see e.g. Bäck and Rosema 2008; Blais et al. 2006: 694; Decker and Best 2010: 179–80; Faas 2010; Meffert and Gschwend 2010; Norpoth 1980; Pappi 2007). However, as the 2010 election campaign in the United Kingdom showed (Fisher and Wlezien, 2011; Quinn 2011), voters’ coalition preferences are also relevant in the election campaign and are sometimes published in countries that do not have a ‘tradition’ of governing in coalitions.

2. All of the surveys used were conducted before the respective state election, so that the answers regarding the preferred coalition are not affected by the actual election results and the first steps in the government formation process.

3. The respective datasets were made available to us by GESIS.

4. For an alternative way of identifying potential coalitions see Armstrong and Duch (2010).

5. Some of the surveys include further questions on the respondents’ opinion on specific coalitions. However, these questions are not only asked irregularly, but also in different ways, which prevents us from using these questions.

6. We do not have information on the preference of a voter for a single-party government. To account for this problem, we include a variable in the analysis that identifies potential governments consisting of only one party that has a parliamentary majority. If a party is in such a situation, it should form the government regardless of the voters’ coalition preferences.

7. Table 3 shows only the three coalitions that are by far the most common pre-electoral alliances in Germany.

8. The conditional logit model assumes the independence of irrelevant alternatives (IIA). That is, ‘the odds of choosing one alternative over another do not depend on any other alternatives in the choice set or on the values of the covariates associated with those alternatives’ (Martin and Stevenson 2001: 39). We check whether the IIA assumption is violated by applying the test procedure developed by Martin and Stevenson (2001). This is not the case in any regression model presented here.

9. We only include the residuals of a logistic regression on the variable ‘pre-electoral alliance’ in our regression model to control for potential endogeneity problems. The results of this regression are (robust standard errors in parentheses; single-party ‘coalitions’ excluded): Cross-cutting coalition –1.90 (0.58) Cross-cutting coalition X formation process pivotal  –0.54 (0.86) Incumbent government 3.47 (0.53) Minimal winning coalition –12.24 (0.41) Bargaining proposition 13.20 (0.64) Strongest party 0.12  (0.55) Economic policy distance –0.22 (0.07) Social policy distance –0.07 (0.05) Constant –1.67 (0.44)with N = 1255, Log pseudolikelihood = –101.32, Pseudo R² = 0.417.

10. Because pre-electoral pacts and other characteristics of potential coalitions like the policy distances between parties might also affect the coalition preferences of voters, we include – similar to the proceeding with the variable that covers information on pre-electoral alliances – only the residuals of the voters’ most preferred coalition government in the conditional logit model. The results of the logistic regression on the voters’ most preferred coalition government are (robust standard errors in parentheses): Pre-electoral alliance 3.94 (0.64) Incumbent government 1.41  (0.45) Cross-cutting coalition –0.67 (0.47) Cross-cutting coalition X formation process pivotal  0.28 (0.46) Minimal winning coalition –11.80  (0.30) Bargaining proposition 13.13 (0.39) Strongest party 1.78  (0.47) Economic policy distance –0.17 (0.06) Social policy distance 0.11 (0.04) Constant –4.67 (0.35)with N = 1575, Log pseudolikelihood = –182.86, Pseudo-R² = 0.405.