Lieven De Winter & Patrick Dumont. Handbook of Party Politics. Editor: Richard S Katz & William Crotty. 2006. Sage Publishing.

Introduction

The process of government formation constitutes a crucial phase and arena in democratic governance: it concerns the translation of electoral and parliamentary power into executive power, and the possibility of implementing policies that have been democratically endorsed and legitimated by the electorate. As an increasing number of Western European governments are based on a coalition of political parties, studying the process of coalition formation has never been as relevant for the understanding of parliamentary democracies.

This chapter focuses on the executive office-and policy-seeking behavior of parties and on the institutional mechanisms and contexts that make possible or impede responsible party government, especially in the case of government coalitions. The following questions will be addressed:

• Which parties get into government, and how is coalition composition decided?
• How are portfolios allocated between and within parties?
• How are governmental policies defined during government formation?

Coalition Composition

The most substantive research question studied in recent decades concerning parties and governments has undoubtedly been ‘Which parties get in?’ This is most relevant in ‘minority situations,’ that is, when election outcomes do not fully decide this question by awarding one party a majority of seats in parliament. With the increasing fragmentation of European party systems (Lane and Ersson, 1999: 142), minority situations and executive power sharing are more and more the normal outcome of elections. Hence, the process of bargaining over who will share power, and under what terms and with what policy content, is a core moment of European politics, and in many countries central to defining public policies.

Given this substantive importance, the theoretical literature on government coalition formation is one of the most active areas of research in the discipline, and, as a result, the literature is now replete with theories, hypotheses, and empirical tests regarding why some coalitions form while others do not, produced by scholars from different disciplines using ever more sophisticated statistical methods and increasingly rich data sets. However, this high level of scientific endeavor has hardly resulted in significant comprehensive progress in explaining and predicting real-world government compositions.

Office and Policy

The first school of coalition theory was strongly inspired by game theory (Von Neumann and Morgenstern, 1953; Riker, 1962) and spatial theories of party and electoral competition (Downs, 1957; Black, 1958). Political parties and their leaders are conceived as rational actors, searching to maximize their utility by gaining office.

The ‘size school’ (also referred to as office-seeking or policy-blind theories) has formulated several ‘classic’ propositions or rules:

• The ‘winning’ proposition stipulates that only majority governments will form, as the core feature of parliamentary government is that a government can only survive if it is supported by a majority in parliament. A minority cabinet reaping all the benefits of office would not be tolerated by a majority opposition consisting of pure office-seeking actors.
• The ‘minimal winning coalition’ (Von Neumann and Morgenstern, 1953; Riker, 1962) proposition stipulates that coalitions should not contain any ‘surplus’ members (i.e., parties whose omission would not make the coalition lose its parliamentary majority). As the pay-offs (the number of ministerial offices) of coalition are fixed, the inclusion of a surplus party would force coalition partners to share ‘unnecessarily’ the spoils of office with the surplus member(s) in such a constant-sum game.

These two propositions predicted an often high number of equiprobable ‘rational’ outcomes, and at the same time were rather unsuccessful in predicting coalitions formed in the real world. Theoretical refinements of the size principle aimed at increasing its predictive efficiency by reducing the number of rational outcomes:

• The ‘minimum seats’ proposition states that in the case of different minimal winning solutions, the minimal winning coalition that controls the minimum number of seats will form (Riker, 1962). If in forming a minimal winning coalition a party can choose between a larger or smaller partner, it will opt for the smaller, assuming that each partner will receive ministerial offices in proportional to its weight in terms of parliamentary seats in the coalition (Gamson, 1961).
• The ‘minimum parties’ proposition—also called the ‘bargaining proposition’ (Leiserson, 1966)—stipulates that when different minimal winning solutions exist, the minimal winning coalition that includes the smallest number of parties will form. Here, the argument is not based on the size of the rewards, but on ‘bargaining facility.’ The smaller the number of partners at the negotiation table, the more smoothly the bargaining is supposed to go, the easier it is to reach an agreement, the more rapidly office rewards can be reaped, and the more durable these rewards will be (increasing the number of partners would raise the probability of conflicts between cabinet parties and would thus endanger the coalition’s stability).

Although these propositions greatly reduced the set of rational outcomes, they often still allowed for multiple predictions. Moreover, the reduction of the prediction set was accompanied by a decrease of the success rate in predicting the actual governments (Browne, 1973: 28). In order to improve the predictive power of rational office-seeking theories, policy proximity was introduced as an additional constraint on the ‘size and number’-based propositions:

• The ‘minimal range’ proposition (Leiserson, 1966) stipulates that parties wish to be part of a minimal winning coalition with minimal ideological diversity. Hence, of the minimal winning coalitions, the coalition with the smallest ideological range, defined as the distance (on the main dimension of competition) between the two most extreme members, will be formed.
• The ‘minimal connected winning’ proposition (Axelrod, 1970) also concerns policy proximity, stipulating that, amongst the minimal winning coalitions, a coalition whose members are ideologically ‘adjacent’ will form. Removing any of the partners would render the coalition either non-connected or non-winning. Hence, contrary to the previous proposition, no ‘gaps’ are permitted in the ordinal scaling of coalition parties on the main policy dimension of party competition, which sometimes necessitates the inclusion of a small surplus party ideologically situated between two otherwise unconnected partners.

While both scholars only added policy as an additional concern of office-seeking political parties, De Swaan (1973) shifted the emphasis towards policy, arguing that parties first try to maximize ‘policy coherence’ (expressed in terms of a minimal distance between coalition policies and the party’s own most preferred policy). Hence, he formulated:

• The ‘minimal policy distance’ proposition, predicting that of the winning coalitions a coalition for which the member parties expect that the coalition will adopt a policy that is as close as possible to their own most preferred policy will be formed.

Some extreme versions of the policy-seeking approach have abandoned office motivations altogether.

• The ‘median legislator’ proposition is drawn from majority-rule spatial voting models. The median legislator is the member of parliament who occupies the median position on the relevant policy dimension. Formal theory shows that when parties compete along a single policy dimension, the party controlling the median legislator can act as a ‘policy dictator,’ as it cannot be defeated by a majority of parties on its left or on its right and one can therefore predict that this policy dictator party will always get into the government (Laver and Schofield, 1990: 111). While this theory does not predict full coalition composition outcomes, it does theoretically help to solve another of the main real-world paradoxes of the game-theoretic approach, namely the frequent occurrence of minority governments that obviously violate the basic ‘winning’ proposition.

Laver and Schofield (1990: 88) argued that minority governments may occur because of policy divisions amongst opposition parties, so that they cannot combine and agree on a viable policy alternative. Minority governments exploit these divisions by forming majorities on an issue-by-issue basis. Strøm (1990) devised the following testable proposition:

• The ‘division of the opposition’ proposition: the more ideologically divided an opposition controlling a majority of parliamentary seats, the higher the chances that a minority cabinet will form. For an ideologically extreme party, sustaining a minority solution can represent a good solution, as participating in a coalition could cause electoral damage due to the policy compromises it would trigger, while bringing down a minority government could result in the subsequent formation of a majority coalition ideologically more distant from the extreme party than is the current minority government.

Generally, adding policy to the propositions has improved the predictive value of the early pure office-seeking theories. Still, the empirical results remain modest. While obviously richer in theoretical terms, policy-driven coalition theories require parties to be placed in an ideological space (which can be one- or multidimensional) before they can be tested empirically. Whatever the placement method chosen (manifesto content analysis, survey analysis of electorates’ or elites’ policy positions, expert surveys), measuring party positions remains a difficult exercise. It is all the more so for multidimensional models of coalition formation, to which we now turn.

Most early policy-driven theories were unidimensional, although the policy space in some of the countries on which the theories were tested were recognized to be multicleavage polities (Lipset and Rokkan, 1967; Lijphart, 1984: 130). Hence, by only focusing on the left-right dimension, researchers may miss an important part of the real-world picture of party policy competition. Especially in multidimensional fragmented party systems, there is a wide range of smaller parties that may seek to realize their policy preferences elsewhere than on the left-right divide (centre-periphery postmaterialism, etc.). Hence, including such parties may produce a coalition that appears to be unconnected, simply because one is looking at the wrong policy dimension, while the coalition may be perfectly connected on other policy dimensions more relevant to those parties. Hence, a single-minded focus on the left-right divide may well obscure rational policy-seeking behavior on other dimensions. The incorporation of multiple dimensions is increasingly warranted, given the emergence of new cleavages, the shift from social-structural voting to issue voting (Dalton, 1996) and increasing party system fragmentation.

A large number of multidimensional coalition formation theories have been formulated7. They all share the assumption that parties want to be members of a winning coalition that is as close as possible to their own ideal position in the multidimensional space. The theories differ strongly, however, in their definitions of which coalition produces the highest utility, that is, satisfies best the smallest distance assumption (De Vries, 1999: 16-17).

Multidimensional models also share several theoretical and practical problems. First, most of them generally do not predict a single or limited number of outcomes unless rather unlikely conditions are fulfilled. Second, most spatial theories are not designed to predict coalitions but to search for an ‘undominated policy point’ in the space. Additional inferences are thus needed to identify which parties stand a better chance of being included in the coalition. Third, they require metric data on party positions on the relevant dimensions. Because dimensions other than left-right are dealt with in party manifestoes, most authors use the Manifesto Research Group data. However, these data were collected to measure the salience of issues and not parties’ policy positions. It may thus require a heroic leap to infer party positions from such a data set. On the other hand, expert surveys are only snapshots and may be subject to other sources of errors such as the contamination of experts’ placement of parties by their knowledge of prior coalition government experiences. Most often, different data sets produce different party positions (De Vries, 1999: 240), and the need for measurement on multiple dimensions increases the risks of such discrepancies, which in turn generate different predictions for the same models. Although multidimensional models aim to be more realistic, they are quite unstable, and empirical results are unimpressive. Finally, multidimensional models assume that party leaders are aware of their exact policy positions, and those of all the other parties represented in parliament, on a number of relevant dimensions of competition, and are, moreover, sophisticated enough to carry out quite complicated calculations to determine the coalition that will be closest to their preferred policy positions.

The Role of Institutions in Coalition Formation

New institutionalism emerged in the early 1980s as a major alternative to the traditional institution-free approaches by emphasizing the role of different types of institutions in structuring the outcomes of the coalition formation process. The new institutionalist approaches do not reject office—and policy—seeking rationales, but add constraints imposed by institutional rules. Institutions are defined as any restriction on the set of feasible cabinet coalitions that is beyond the short-term control of the players (Strøm et al., 1994). Hence, differences in coalition outcomes (especially relevant for cross-country comparisons) are predicted on the basis of institutional differences with regard to coalition bargaining rules and norms that allocate power differently between (party) actors in the bargaining process.

One can distinguish neo-institutional theories based on the rules and norms governing the process of government formation itself, from the (more recent) theories that focus on the rules that structure cabinet decision-making and inter-party bargaining in the post-formation phase.

Institutions Structuring the Formation Bargaining Process

The first group of institutions (for inventories, see Strøm et al., 1994; Laver and Schofield, 1990; Mershon, 1994) includes the following:

• ‘Recognition rules’ that stipulate which party or parties will be asked to form a government, and in what order. This recognition may be enshrined in formal and even constitutional rules, or invested in other actors such as the head of state (Bogdanor, 1984). Recognition invests a formateur (and his party) with the power to propose the coalition alternatives over which bargaining will take place. Potential partners must accept or reject the proposals brought forward by this formateur before bargaining over other proposals can proceed. Hence, a formateur party should be able to guarantee its own entry into the government as well as to propose and have accepted coalitions composed of parties that it finds to be most compatible with its own policy preferences. As formal recognition rules are quite rare, in the real world the largest party generally tends to become the formateur party (Diermeier and Merlo, 2004), while according to Morelli (1999) it is the median party in systems where the head of state has discretionary power over the selection of formateurs.
• The power to control the timing of the bargaining over a new cabinet, especially by incumbent prime ministers. Strøm et al. (1994) argue that an incumbent cabinet that manages to stay in office during the (usually post-electoral) formation bargaining over its successor, enjoys an advantage in the coalition negotiations, as it constitutes the fall-back or ‘reversion outcome’ if bargaining goes on endlessly or breaks down. Thus, parties whose most preferred outcome coincides with this reversion outcome enjoy a particular advantage, as they have an interest in boycotting or sabotaging any other alternative coalition formation attempt to which they are invited.
• Investiture rules stipulating whether a new government must pass a formal vote in the legislature, and with what kind of majority. Theoretically and empirically, minority governments are more likely to form in the absence of an investiture rule, since their general policy program need not be subjected to a formal parliamentary go-ahead at the cabinet’s initiation. In countries with such ‘negative investiture’ rules, minority governments can avoid instant ‘political death sentences’ expressed by a motion of defiance from majority opposition by skillfully manipulating the legislative agenda and building ad hoc majorities on each separate issue the government submits to a vote in the legislature (Bergman, 1995).
• Rules constraining the party composition of a government (such as the Belgian constitutional rule that requires an equal number of French- and Dutch-speaking ministers) or the size of a ‘winning’ majority, which has for too long been narrowed down to controlling a simple majority. In certain systems, controlling a majority of seats in the parliament is just not enough to implement an ‘ordinary’ policy agenda. First, there are policies that require constitutional reform, for which in many countries a special threshold (of two-thirds or more) has to be attained, sometimes spanning two legislatures (Lijphart, 1999: 217-23). Also, in many countries particular policy sectors require special majorities (Müller, 2000: 91). Recently, scholars set out to explore the consequences of bicameralism for government formation. Parties seem to anticipate the potential instability induced by the presence of a second legislative chamber and decide to form larger coalitions in the lower house (Diermeier et al., 2002, forthcoming) or, minority coalitions if they rely on a majority in the upper house (Druckman et al., forthcoming).

Some behavioral rules, not enshrined in formal institutions, also affect the bargaining process and outcome. The most prominent is the respect of pre-electoral commitments between parties to form the next government, election results permitting. This can either take the form of a positive statement that commits a party to form a government with another party (thereby implicitly excluding other parties that did not sign the pact), or negatively, as an ‘anti-pact’ in which parties declare that they will not coalesce, for instance by ruling out either any coalition that includes a particular ‘pariah party,’ or the party’s participation in a specific coalition or any coalition. Given the general moral principle of pacta sunt servanda and credibility as a central ingredient for successful coalition building, public pre-formation commitments to rule or not to rule with some other parties constitute very powerful real-world constraints on coalition bargaining (Martin and Stevenson, 2001; Golder, 2004).

Most existing theories are ‘history blind’ in another way, by assuming that the formation of a particular government will not be influenced by the formation of preceding governments. Some authors introduced parties’ past histories in the form of mutual satisfaction based on past gains. In particular, ‘familiarity’ between parties that governed together in the past enhances the probability of governing together in the future, thereby showing that governments are not formed ab nihilo (Franklin and Mackie, 1983).

Institutions Structuring Post-Formation Government Decision-Making

Other recent theories explore the constraints on coalition composition of rules that structure government decision-making once a new government is in place, thus effectively ruling out certain combinations and making others more likely by anticipation. These include cabinet operational rules (e.g., the balance between collective cabinet and individual ministerial policy jurisdiction (Laver & Shepsle, 1996; for a critique, see Dunleavy and Bastow, 2001), the prime minister’s power to reshuffle or deselect ministers (Strøm, 1998), political responsibility and resignation rules); parliamentary rules (decision rules such as qualified majority votes, the right to turn any vote on a specific issue into a vote of confidence (Huber, 1996)); rules for dissolving the legislature and for calling early elections (Strøm and Swindle, 2002); electoral system rules; rules granting power to external veto players (head of state, domestic pressure groups), etc. Strøm et al. (1994) demonstrated on a small sample that the institutional constraints model manages to predict the cabinets actually formed better than simple size- and policy-driven models, but verification of the full institutional model on a large number of cases is still lacking.

Bargaining Theories and Actor-Oriented Theories

In some theories, parties’ chances of being included in government depend on their strategic position in legislative bargaining games. Hence, they focus on the existence and properties of such special ‘powerful’ or ‘dominant’ players, and the probability of these parties getting into government, rather than the probability of particular coalition formulae.

The most basic and best-known version of this approach is the median legislator theory mentioned above. Another actor-oriented bargaining approach developed voting power indices, of which the Shapley-Shubik (1954) index and the Banzhaf (1965) index are the most renowned. The voting power of a player, which is calculated by listing all the coalitions in which it makes the coalition win or lose (the number of pivots or swings), is compared to the voting power of other parliamentary parties. The voting power of individual parties may differ starkly from their weight in seats: a rather small party may be as powerful (and, depending on the index used, may be even more powerful) than a much bigger one within a given distribution of seats. Although hitherto rarely employed, the inclusion of the power indices instead of the party weights, and their combination with the assumptions on policy distance and institutional constraints used in formal theories, could well be one of the most promising new avenues for formal as well as descriptive research on coalition formation in terms of composition, portfolio allocation, policy formulation and process. Notice also that following Warwick’s (1996) analysis of factors affecting individual parties’ odds of becoming government formateur or coalition partner, some scholars looked more specifically at parties’ results in the elections preceding the formation of a government (Mattila and Raunio, 2004; Isaksson, 2005).

Discussion

After four decades of comparative research on the party composition of cabinets, this research field has become highly mature, in terms of the diversity of theoretical approaches and paradigms that are competing, their degree of formalization, the variety and sophistication of methods applied, and the scope and richness of the data sets used for testing hypotheses.

However, there are still major shortcomings. First, existing theories do not predict and therefore sufficiently explain a significant proportion of cabinet compositions formed in the real world. Whereas the latest comprehensive model of Martin and Stevenson does predict correctly an impressive number (about half) of real-world coalitions, this is done by lumping together two dozen variables drawn from three main schools, and the model therefore lacks parsimony and internal consistency. Still, a systematic comparative testing of theories and families of theories against each other, in combination with each other, as well as against randomly generated solutions seems a promising path in evaluating the predictive capacity of coalition theories (De Vries, 1999; Martin and Stevenson, 2001). Studies that concentrate on parties getting into government as the unit of analysis rather than the full cabinet composition have better prediction rates, but they clearly explain less. More generally focusing on predictions is only one part of a causal explanation, which needs both an account of causal effects of independent variables and a verification of the real-world presence of the causal mechanisms posited by a theory. This goal may be achieved by combining quantitative tests of existing theories with a qualitative treatment of cases that confirm and cases that disconfirm the theory, by tracing the process and the variables that caused the observed outcome (Bäck and Dumont, 2004).

Second, in order to test existing spatial theories, especially multidimensional models, one needs more reliable data on party policy positions. Also many formal models theorize variables that in practice are hard to operationalize, especially for cabinets in the more distant past (such as actors’ electoral expectations or satisfaction with former experiences with partners in government). It comes as no surprise that the authors who formulate such abstract models at best give one or two examples that seem to fit their model. The insertion of institutional variables into formal models, albeit generally easy to collect and boosting the prediction rate, tends to lead to unacceptable simplifications of reality and/or to unmanageable mathematical complexities (De Vries, 1999).

Finally, although most scholars acknowledge that during government formation a lot of bargaining goes on within parties (Luebbert, 1986; Laver and Schofield, 1990; Müller and Strøm, 1999), for the sake of model simplicity as well as data collection problems on internal party divisions, almost all theories treat parties as unitary actors (for exceptions, see Robertson, 1976; Budge and Farlie, 1983; Maor, 1995). One operational indicator of internal divisions could be voting behavior during party investitures (De Winter et al., 2000: 345). The support expressed for the upcoming coalition within intra-party arenas may serve as a more valid proxy for party cohesion than parliamentary group cohesion in investiture votes.

Portfolio Allocation

The question ‘Who gets what in coalition government?’ is even older than the question of ‘Who gets in?.’ The basic finding of the first analyses inspired by rational choice, that is, the existence of an ‘iron law’ of proportionality (Gamson, 1961), has generally been confirmed by subsequent research (Browne and Franklin, 1973; Browne and Feste, 1975; Budge and Keman, 1990). While this proportionality norm may seem trivial, it actually is not, especially seen from a bargaining perspective. Hence, interest in the question of portfolio allocation has revived in recent years in order to provide a more solid theoretical grounding for this empirical law, to explain marginal but significant deviations from the rule, and to produce weights and data (through expert surveys) for different types of ministerial portfolios (prime minister, senior ministers, ordinary and junior ministers).

The Iron Law of Proportionality

The starting point for the proportionality rule is the assumption that players have a specific weight, usually assumed to be proportional to the number of seats in parliament. Gamson (1962: 158) postulates that ‘any participant will expect others to demand from a coalition a share of the payoff proportional to the amount of resources which they contribute to a coalition.’ Still, a minor player should be capable of increasing its office share if it were necessary to keep a third player out of the coalition (Gamson, 1961). As all players consider proportionality as their bottom-line demand and concede that the other players take proportionality as their minimum expectation, proportionality becomes the only allocation principle on which all can agree.

Empirical analyses confirm the proportionality thesis (over 90% of the variance explained), sometimes with smaller parties getting bonus portfolios, probably in order to avoid their potential defection to another coalition that would offer a better payoff or simply because their small size would not have allowed them to receive any portfolio under pure proportionality.

Still, this very strong relationship between seat weight and portfolio allocation is puzzling from a bargaining perspective: if bargaining is the predominant logic behind coalition formation, and if a party’s bargaining power is dependent on its strategic position rather than on its size alone, what in practice prevents parties with great bargaining power from claiming a disproportionate share of the portfolios?

Qualitative Portfolio Allocation

Apart from the quantitative questions of portfolio allocation, there is the question of which party, intra-party faction, and individual gets which type of portfolio (in terms of policy domain, prestige, spending power, etc.). Until the 1990s, few empirical theories on qualitative portfolio allocations were tested (Budge and Keman 1990). Laver and Shepsle’s (1996) assumption of ministerial portfolio dictatorship, coupled with the hypothesis that the median party on the relevant policy dimension will get the specific portfolio, has drawn new attention to the qualitative aspect of portfolio allocation.

A first qualitative distinction between portfolios is policy domain. Some link was found between parties’ ideological profile and their control of a particular domain of ministries. Browne and Feste (1975) explain this party-portfolio link by the ‘possible reinforcement of the loyalty of each party clientele’s group.’ A basic problem of the empirical testing of this nexus is the a priori assumption of the static nature of party families’ portfolio preferences, which are assumed constant over time and between countries (Budge and Keman, 1990). The empirical findings are therefore rather weak. Budge and Keman (1990: 98-102) also focused on the prime ministerial portfolio, predicting that the premiership goes to the major party (which actually occurs in 80% of the post-war European coalition governments). Again, under the proportionality rule, one can expect that the largest party can claim the largest spoils, and therefore can choose the biggest prize. The few deviations can be explained by presidential nomination power, by a rough equality of size between the main coalition parties, and by intra-party dissensus impeding the largest party from nominating a candidate for this office.

This points to a wider question of party unity in seeking portfolios. Case studies on factionalized parties—such as Christian Democrats in Latin countries (Blondel and Cotta, 1996; Mershon, 2001)—highlight the role of factions in distributing ministerial portfolios within parties. In addition, even in unitary parties, party leaders have to take into account a series of equilibria, in terms of the territorial background of ministers, constituency party support, gender, distribution between first and second chamber, etc. Hence, intra-party decision-making rules and constraints, and the old question of party cohesion, have entered the most recent research (Laver and Shepsle, 2000; Pennings, 2000).

Discussion

By linking the coalition formation process to the portfolio allocation process, some recent formal theories have tried to overcome the limits of the first generations of quantitative and qualitative portfolio allocation theories. This linkage is made by assuming that portfolios are all about policy, and that obtaining ‘policy portfolios’ rather than ‘offices and their perks’ is the core of the coalition bargaining process (Austen-Smith and Banks, 1990; Schofield, 1993; Laver and Shepsle, 1990, 1996). Although these theories have been empirically tested on only a few selected cases, these formal theories also introduce a variety of institutional conditions under which the policy portfolio bargaining is supposed to occur, and thus, unlike their predecessors, are not institution-blind. They include rules concerning the nomination discretion of the prime minister and the head of state, the power of the finance minister, the role of party leaders, etc. Yet, rather than equating portfolio bargaining to policy bargaining, one should keep these goals analytically separate, as parties may differ in the emphasis they put on maximizing policy, office or vote (Strøm and Müller, 1999). These disparities allow for tradeoffs between the different payoffs, which in principle should facilitate coalition formation. For instance, office-seeking parties may want to trade policy concessions to the policy-seeking parties in return for a disproportionate share of portfolios.

Moreover, a number of questions remain unanswered regarding the dependent and independent variables, the process that links them and the potential effects on coalition governance and outputs in this type of research. As far as the dependent variables are concerned, the study of the full range of relevant offices has only very recently started. Junior ministerships are often used as spare change to round off or fine-tune portfolio deals but sometimes also are considered to be policy-relevant (Mershon, 2001; Thies, 2001; Manow and Zorn, 2004). In some countries, the ‘offices cake’ should be extended to include the Speaker in parliament or the European Commissioner.

Although important efforts have recently been made to weight a wide range of portfolios (Druckman and Warwick, 2005), we still know little about how parties in practice weigh portfolios and which criteria they use to do so: policy relevance (Dumont, 1998), interests of traditional or new clienteles, patronage opportunities, prestige, visibility, distributive versus redistributive policy departments, goodness of fit with the qualifications of the ministrables available in the party, the appetite of incontournable ministrables, etc.). In several countries, parties or the head of state allot certain portfolios to non-partisan technocrats (De Winter, 1991; Strøm, 2000; Amorim Neto and Strøm, 2004), a practice that certainly contradicts the office-seeking drive on which most portfolio theories are based.

With regard to the policy impact of portfolio bargaining, it is also important to know when portfolio bargaining actually starts: at the end of the policy negotiations (as most research seems to suggest), or before. This question is linked to decision-making within parties. Are ministrables members of the negotiating teams that draft the policy agreement and bargain over portfolios? Do they help themselves to a portfolio of their liking? Does such interlocking of principal and agent, when party leaders bargain over portfolios that they themselves would like to occupy (Andeweg, 2000), allow ministers to become unaccountable policy dictators (Laver and Shepsle, 1996), in spite of the formal existence of a collective coalition agreement? To date, most empirical analyses suggest that only the allocation between parties matters with regard to policy outputs (Laver and Shepsle, 1994; Klingemann et al., 1994).

Policy Formulation

There are two contrasting views concerning the policy relevance of coalition negotiations (Timmermans, 2003): the ‘positive’ version considers the formulation of coalition agreements as a genuine opportunity for parties to influence the future government’s policy agenda (Peterson et al., 1983; Peterson and De Ridder, 1986); the ‘sceptics’ consider coalition negotiations as ‘policy irrelevant,’ being either just a ‘ritual’ carried out to ease the transition from election campaign competition to inter-party governmental co-operation (Luebbert, 1986), or because the link is only conditionally relevant, that is, it is relevant only if policy proposals supported in the coalition agreement are supported by the party that receives the relevant ministerial portfolio (Laver and Shepsle, 1996: 42).

Empirical studies tend to support the positive view. In most countries, policy bargaining is the main subject of the formation negotiation process. This can be inferred from numerous indicators. First, there is a growing tendency in West-European governments to draft a government policy agreement (Müller and Strøm, forthcoming). Policy bargaining consumes most time in the bargaining process (often weeks or months), while the allocation of portfolios is settled in a few hours or days (De Winter and Dumont, forthcoming). Apart from Italy, portfolio allocation follows the conclusion of a general agreement on policy between the parties that will constitute the next government (Budge and Laver, 1992: 415), and most formation attempts break down on policy, not on the allocation of portfolios. Coalition agreements tend to grow in length and detail, and they mostly cover substantive policy areas (Müller and Strøm, 2000). Second, coalition agreements contain issues that are salient to the member parties and for which they have formulated policy pledges, even when they disagree on the solution of such issues (Thomson, 1999). They do not focus only on non-divisive issues (Klingemann et al., 1994), and when divisive ones are not mentioned in the agreement, mechanisms to deal with these during the life of the cabinet are often specified (Timmermans, 2003). But overall, during the negotiations the pledges most salient to each party tend to receive explicit attention.

Regarding their potential policy impact, although coalition agreements are never legally binding, in practice they often do bind partners strongly. In fact, these agreements usually are endorsed by the parties’ main decisional body (e.g., the party congress, parliamentary party or party executive), and, as such, this endorsement legitimately binds all the other sections of the party, from the rank-and-file members to constituency party organizations, individual members of parliament and ministers, the party executive, and the party leader. The party investiture is therefore a crucial moment for making agreements stick, not only between parties but also, more importantly, within parties (De Winter et al., 2000).

In most countries these coalition agreements are widely available to the general public, which expands their utility for scrutinizing government performance by party bodies, as well as by other actors, such as interest groups, the media, and retrospective voters. Coalition cabinets and parties have, in addition, set up a variety of mechanisms and rules to facilitate the smooth implementation of these agreements, to solve conflicts over the way they should be interpreted, to formulate an answer to issues not anticipated by or included in the agreement, and to amend these agreements without jeopardizing the coalition’s survival.

Recent empirical research (Müller and Strøm, forthcoming) suggests that in most countries coalition agreements are central instruments for coalition policy-making. They also have an important theoretical role in the process of governance once a cabinet has been formed: they are vital devices that make coalition government possible and help tackle some of the severe and complex bargaining problems and inherent weaknesses of coalitions, such as limited information, non-simultaneity of exchange, cyclical voting, formal ministerial discretion over departmental policy agendas, their implementation and coordination, problems of interdepartmental interdependency shared competencies, changes in external context, etc. (Lupia and Strøm, forthcoming).

Hence, from an empirical and theoretical point of view, coalition agreements emerge as one of the main institutions that make collective coalition cabinets viable. Still the practical working of ‘governing by contract’ remains to a large extent a black box.

Conclusion

The formation of a government is a crucial aspect of democratic politics, because it deals with the conquest of power and the possibility of implementing policies that have been democratically endorsed and legitimated by the voters. This field of party studies is, together with voting, probably the most mature field in party research, in terms of the variety of competing theoretical frameworks produced by scholars from different disciplines, the sophistication of statistical techniques, and the richness of available data sets. However, this high level of scientific activity has in the last decade scarcely resulted in significant comprehensive progress in explaining and predicting real-world government compositions. Models aggregating current knowledge still do not manage to predict correctly more than half of the coalitions actually formed in the real world.

The field still has several shortcomings. First, the inclusion or exclusion of explanatory variables in formal models is often determined by difficulties of operationalization rather than by theoretical coherence. Also, some crucial variables are often poorly operationalized, leading to unreliable and unstable conclusions, especially in the field of party ideological positions. There is also the problem of selection bias as the prediction success rate of composition and portfolio allocation theories differs between countries (and to a lesser extent between time periods). The goodness of fit of theories is thus conditional on the countries and time frames selected.

Furthermore, some essential components of coalition formation are traditionally neglected, like the frequently occurring formation failures (Müller and Strøm, 2000: 570). Also we know very little about the operation of the formation process, in terms of negotiators, their autonomy vis-ă-vis their party principals, their tactics and games, their criteria for evaluating alternative policy and portfolio proposals, the benefits of a retreat into opposition, etc. Also, political parties figure as the main actors in explaining coalition outcomes in almost all theories, while in certain systems the head of state, pressure groups, or foreign powers sometimes have a significant formal or informal veto power.

Many of the real-world formations remain theoretical puzzles, of which vital explanatory pieces are lacking (De Winter et al., forthcoming). Government formation, like all politics, is conducted by human actors, but unlike studies of most other political activity (e.g., voting, political participation, legislative behavior), coalition formation theory does not pay much attention to the accounts and explanations of the human actors involved in government formations themselves. This imbalance can be redressed by ‘thick’ descriptions of government formations, preferably using information from participants obtained through elite interviewing, analyses of memoirs, etc. Only in this way can we try to reconstruct actors’ preferences, motivations, strategies, evaluation of past experiences, anticipation of future developments, perception of the credibility of other negotiators, capacity to commit their party, and the perceived impact of formal and informal institutional constraints and veto players. However, the ultimate aim of such an inductive approach should not be the writing of a thriller reconstructing dramatic deviant formations, but to feed new explanations—discovered by thick descriptions—back into theory formulation. As such, inductive research should serve as a complement, not as an alternative, to existing formal theories that have considerably contributed to our current state of understanding of government formation.