Keeping the party together

An online appendix with mathematical proofs will be made available as part of this article at https://​files.​nyu.​edu/​hm484/​public.

In fact, single-party majority governments are not different from explicit coalitions as they both need to maintain a parliamentary majority (Laver 2006).

In the rest of the paper, I use “he” to refer to the minority faction leader and “she” to refer to the party leader.

Patty (2008) also treats party discipline as an endogenous outcome depending on the size of the majority. More specifically, Patty (2008) argues that the strength of the majority party’s leadership in enforcing party discipline is a decreasing function of the majority party’s seat share. Nevertheless, he focuses on the incentives of the majority party leadership to determine its members’ vote choices; whereas here, the focus is on the incentives of the party leader to keep her party united by preventing splits.

In the rest of the paper, R1 and R2 are used to denote both faction leaders and faction/party names.

One might ask why R2 would want to challenge R1 in the first place. It is assumed here is that all governments are in equilibrium at the time of their formation following policy bargaining among actors with diverse preferences (Browne et al. 1986). However, over their lifetimes parties face policy shocks, which can change party positions on key policy dimensions, creating conflict among party factions (Laver and Shepsle 1998).

The party leader gets a zero payoff because she can implement her preferred policy and therefore does not experience disutility stemming from the policy distance between her policy position and the resulting policy.

Empirically, this would imply an equitable allocation of ministries, which have jurisdiction over the policy areas for which they are held responsible.

Note that the non-leadership faction is assumed to be pivotal.

Even if the splinter faction is not centrally located, a winning coalition is most likely to be formed between one of the larger parties and the small party (Baron 1991). Another point that needs clarification is that the focus here is on legislative politics and not electoral politics. Electoral politics would affect the expectations about the governments that are likely to form in the event of a split. If, for instance, one actor expects to increase its seat share, this would change the coalition alternatives by making some coalitions more likely than others and therefore affect the probability of a party split. However, the main intuition of the current model would still apply to a model that also takes electoral considerations into account.

The k parameter includes the cost related to the electoral system, but it is meant to capture more than that. In particular, in Eastern European countries, in the early years of 1990, k was small not only because of the permissiveness of the electoral system, but also because none of the existing political parties had time to build name recognition and reputational capital, hence lowering the cost of forming new parties.

It is assumed here that coalition parties will have a proportional say in the government’s policy. This is in fact in line with the prediction of Gamson’s Law, which refers to the strong empirical regularity that the proportion of cabinet ministries received by each government party tends strongly to be equal to the proportion of legislative seats contributed by that party to the government’s seat total (Gamson 1961).

This is analogous to having an indicator function that takes the value 1 when the party leader holds the premiership position, and 0 otherwise.

The formal derivations of the equilibria are presented in the Online Appendix.

Remember that the maximum value that \(\lambda\) can take is 1.

The theoretical model presented in this paper is on single-party majority governments analyzed in the context of two-party systems. In a multi-party system, such as Italy, the coalition formation dynamics following a split are expected to be different, making some coalitions more likely than others and affecting the actors’ actions ex-ante. However, the main intuition of the model’s results extends to multi-party cases since in multi-party systems, as in two-party systems, political competition often takes place between the government’s main party leader (who is usually also the leader of the largest party) and the main opposition party leader (who is also usually the leader of the second largest party) who compete for the prime minister’s position. Comparative data on governments shows that out of the 304 governments formed in 12 advanced industrialized democracies between 1960 and 2005, 242 (\(80\,\%\)) had the largest party’s leader and 38 (\(12\,\%\)) had the second largest party’s leader as the PM (Mutlu-Eren 2010).

Following Ezrow (2007), the polarization variable is measured as the weighted average of the absolute deviations of party positions from the weighted mean of all parties. Alternative measures include Dalton’s polarization measure (Dalton 2008) and the standard deviations of party policies. Party positions are coded using Comparative Manifesto Data (Klingemann et al. 2006).