Making the first move

Public ChoiceDOI 10.1007/s11127-007-9267-5

Making the first moveA two-stage analysis of the role of formateurs in parliamentarygovernment formation

Hanna Bäck · Patrick Dumont

Received: 11 December 2006 / Accepted: 29 November 2007© Springer Science+Business Media, LLC. 2007

Abstract A standard conclusion of theorists who model bargaining as a non-cooperativegame is that the party designated to make the first move—the formateur party—will deter-mine the bargaining outcome. Most empirical studies of parliamentary coalition formationhave paid surprisingly little attention to the formation process. In this paper we model gov-ernment formation as a two-stage unordered discrete choice problem that better reflects thisprocess. The first step involves the selection of a formateur party, and the second involvesthe choice of partners by the predicted formateur. We evaluate several hypotheses for the twostages, using a data set of all cabinets formed in the Western European countries from 1970to 2006. In our analyses of formateur selection, we find that party size is clearly the domi-nant feature. In the second stage, we show that when predicting government composition itis fruitful to add information drawn from a first stage analysis.

Keywords Formateur parties · Coalition bargaining · Government formation ·Parliamentary democracies · Two-stage model

1 Introduction

In any parliamentary democracy, generally in the aftermath of an election, a process ofgovernment formation takes place. This process usually starts by the Head of State selectinga person or party that is given the first opportunity to form government. Because no partytypically gains a majority of the seats, this formateur has to negotiate with other parties inorder to make a government proposal, which may be accepted or declined in parliament.The course of cabinet formation in most parliamentary democracies more or less followsthis two stage-process where a formateur is selected in a first stage, and in a second stage agovernment is formed.

H. Bäck (�)MZES, University of Mannheim, 68131 Mannheim, Germanye-mail: [email protected]

P. DumontUniversité du Luxembourg, Avenue de la Faïencerie, 162a, 1511 Luxembourg, Luxembourge-mail: [email protected]

Public Choice

Most coalition studies have paid little attention to the process of government formation.The traditional coalition theories are all based on cooperative game theory, and make pre-dictions about the government outcome, in terms of its composition of parties and size,without making any statements about the bargaining process. As a result, most (large-n)

empirical studies of coalition formation have not made any reference to the process of gov-ernment formation, but have instead focused solely on predicting the composition of partiesin government.

On the other hand, non-cooperative game theorists concentrate on the bargaining process,and a standard conclusion is that the party designated to make the first move—the formateurparty—should always be able to form a government, and can bias government compositionin line with its preferences. Only a few empirical studies of formateur selection exist, suchas Diermeier and Merlo’s (2004) evaluation of two modelling assumptions regarding recog-nition rules, or Warwick’s (1996) study of individual parties’ odds of becoming the PrimeMinister party. And more recently a genuine debate over formateur advantage in govern-ment formation has sprung from scholars’ empirical investigations of portfolio allocation(see Warwick and Druckman 2001, 2006; Ansolabehere et al. 2005; Fréchette et al. 2005;Laver 2006).

In this paper, we combine these two approaches to studying government formation, i.e.,studies focused on predicting who becomes the formateur and those predicting full coalitioncomposition. We model government formation as a two-stage unordered discrete choiceproblem that better reflects the coalition game. The first step involves the selection of a for-mateur party, and the second involves the choice of government composition by the forma-teur. Furthermore, we generate and evaluate original hypotheses differentiating the selectionof the first party given the charge of forming a government, from the selection of the suc-cessful, last formateur (and thus Prime Minister) party. We also run separate analyses for dif-ferent samples of formation opportunities, with a special aim to tap the consequences of theinstitutionalization of the bargaining process on the identity of the party taking the respon-sibility of forming a government, and on cabinet composition. We thereby explore aspectsof coalition formation that have been barely touched upon (De Winter and Dumont 2006).

2 The government formation process in Western European democracies

In many parliamentary democracies, forming a government can be a complex process in-volving a number of consecutive stages that owe both to formal and informal rules of thegame. Constitutional and legal rules often do not reflect the way governments fall or form inpractice (De Winter 1995). For instance, the Head of State is generally less determinant forthe formation outcome than what most constitutions would suggest, even if sometimes thisconstitutional authority has seized the formal powers at his/her disposal to influence gov-ernment formation. The process of government formation generally starts with the formalconsultations of the Head of State with a number of actors. The leaders of most parties rep-resented in parliament are invited by the Head of State in order to inform him/her on theirexpectations and preferences as main actors of the formation game. Most of the time theoutgoing Prime Minister, the speaker(s) of the legislative chamber(s), and leaders of socio-economic groups are also invited, and on the basis of these consultations the Head of Statewill appoint a formateur.

In some constitutional monarchies, the Head of State facing a complex political situationmay prefer to choose an informateur first. This person will carry on consulting on his/herbehalf and will be in charge of finding viable government alternatives, thereby narrowing

Public Choice

the range of subsequent choices. Once appointed, the formateur will then seek to form agovernment, usually first by trying to follow the indications given by the Head of Stateand/or the informateur, but facing difficulties in fulfilling this mission, he/she may either bedischarged or present new alternatives to the Head of State.

When a coalition formula is found, each party selects a number of negotiators, and dele-gations convene, following an agenda set by the formateur. As the negotiations between del-egations proceed, the formateur reports to the Head of State up until an agreement is signedby all negotiating parties. Note that the role of the Head of State in the process may varyfrom an effective involvement in the full process as depicted here, to a purely ceremonialfunction as in the cases of Germany and Ireland, where parliament presents a PM-designateto the nomination of the President, after negotiations between parties have come to an agree-ment. The former type of systems will here be called “formateur” systems and the latter arecalled “free-style” systems.

Government negotiations end up with the signing of a document between parties, whichmay be a comprehensive or a minimal governmental agreement, and a proposal of portfolioallocation. Government parties then usually seek “party investiture”, i.e., endorsement byinternal bodies. At this point, failure of a party to endorse the governmental agreement maylead to renegotiations or the end of the formation attempt. In most of the countries consideredin the present study, a vote in parliament must be taken before a new government is formallyappointed.1

3 Hypotheses for the two stages of government formation

The coalition literature has so far almost exclusively focused on the role of parties in ex-plaining coalition outcomes. As we have seen, however, in certain systems, the Head ofState may have a significant power in the selection of formateurs. More generally, whetherselected exogenously by the Head of State, or emerging out of a norm or context by whichinvited partners recognize voluntarily its legitimacy, a formateur party is given or takes theinitiative of composing a government. Thus, regardless of how it became the formateur, thisparty is in the driver’s seat and conducts the cabinet formation process by proposing a dealto a set of partners of its choice.

Modeling government formation as a two-stage process even when the formateur is notimposed exogenously may appear inappropriate to some readers who would, given the cur-rent state of coalition theory, prefer to draw a neat line between systems best depicted asnon-cooperative games and those where cooperative game theory should apply. We arguethat in analytical terms the full composition of a government formation is not a choice madecollectively by the system, as one actor takes the initiative of choosing a formula and pre-senting it to a set of parties to the exclusion of others. Therefore decomposing the sequenceis warranted. As argued above, we here combine two approaches to investigating govern-ment formation by modelling the process as a two-stage game. In this section we review themain thrust of these two literatures.2

1In Sweden, an investiture vote is required but the government must only be tolerated by a majority inparliament. In some countries, a “positive” investiture vote formally exists, which may concern the PM only(Germany, Ireland) or the whole cabinet (Belgium, Italy, and Finland since 2003), whereas in others it is notconstitutionally or legally required, but customarily held in practice (Netherlands, Luxembourg).2Due to lack of space we here leave out a number of theories predicting coalition composition, such asinstitutional theories focused on explaining the frequent occurrence of minority government in some countries(see, e.g., Strøm et al. 1994; Bergman 1995).

Public Choice

3.1 Bargaining theory and formateur predictions

Building on the work of Austen-Smith and Banks (1988), who depicted bargaining as anon-cooperative game, and the more general model of Baron and Ferejohn (1989), severalbargaining models have emphasized that procedural rules and the sequence of bargainingmatter. A standard conclusion of these models is that the party designated to go first informing a government—the formateur party—will be able to significantly shape bargaining.For example, Baron (1991) concludes that the party selected first to form a government isalways successful in forming a government. Because all proposals contained in the statusquo’s majority win-set would be accepted, the formateur party can exploit its proposal powerto achieve the highest payoff among the alternatives in the win-set. As a result, this party willtypically receive a payoff higher than if it had been in cabinet without being the formateur(Diermeier and Merlo 2004). In terms of office payoffs, the formateur party should receivean disproportionate number of ministerial seats (see, e.g., Warwick and Druckman 2001,2006; Ansolabehere et al. 2005; Fréchette et al. 2005), and in terms of policy payoffs, theformateur should be able to bias “the ideological profile of the cabinet in its favor” (Martinand Stevenson 2001, p. 36).

Hence, if we are interested in predicting the cabinet composition, we should first askwhich party becomes the formateur. Crombez (1996) makes this point explicit in his modelby arguing that the characteristics of this party have an impact on the type of government thatwill form.3 According to Diermeier and Merlo (2004), all proposer models build on eitherone of two bargaining procedures; (1) formateurs are selected randomly proportional to thedistribution of seat shares in the legislature, or (2) formateurs are selected in a deterministicfashion based on their rank order in terms of their share of seats, i.e., the largest party isselected first and then the second largest, and so on. Both of these procedures are based onthe idea that the only party feature that matters for formateur selection is a party’s size, anda conclusion is thus that large parties, and in particular the largest party, should become theformateur. This is in line with the cooperative game theoretical assertion that in systemsdominated by one party, the latter is difficult to exclude (see, e.g., Van Deemen 1989), andwith the normative expectation that the party with a vote plurality should have an advantagein government formation since it is likely to form a cabinet reflecting the voters’ preferences.

Some have argued that other characteristics make parties more or less likely to becomeformateurs. According to Warwick (1996), prior governmental experience should advantagea party for formateur status. This advantage could materialize if there is a more or lessformal institution that says that the outgoing formateur is given the first chance to forma new government and/or the previous government is not formally obliged to resign. Thisis most likely to occur in systems where the Head of State has a purely ceremonial rolein government formation (in free-style systems). This suggests that the party holding thePMship in the previous government will take the responsibility of forming a new one. In“formateur systems” the experience gathered by a former PM party may also be valued byHeads of State who aim to appoint a successful formateur, knowing that any other choiceentails transaction costs.

Another feature that could affect a party’s likelihood of becoming the formateur is itsideological position along some relevant policy dimensions. According to the median voter

3Crombez (1996) argues that, as the largest party becomes larger and more ideologically centrally located,the probability of a minority government increases, whereas the smaller and away from the centre the largestparty, the more likely that a surplus government forms.

Public Choice

theorem, if we assume that the policy space is one-dimensional, the party controlling the me-dian legislator will have increased bargaining power, since there is no other policy positionthan the median legislator’s ideal position that is preferred by a majority of legislators. So ifparties take the charge of forming a government on the basis of their likelihood to succeedin forming a government, this should make median parties likely to become formateurs. Ac-cording to a non-cooperative model developed by Morelli (1999), the median party is mostlikely to be picked as first mover where the Head of State has discretionary power over theselection of formateurs, as the Head of State has an incentive to select parties that are mostlikely to succeed in forming a stable government that will reflect the policy preferences ofthe median voter. Even though we do not expect Heads of State to have discretionary powerover the selection of formateurs, we expect that the effect of the median party variable isgreater in systems where the Head of State selects the formateur, i.e., in formateur systems.

Institutional and bargaining environment factors may strengthen or reduce the effect ofparty-specific characteristics. For instance, in countries using a positive investiture vote pro-cedure, we can expect that larger parties should be favored, as they are close to the 50% hur-dle. The same applies for countries where a qualified majority is requested for constitutionalamendments or important legislation, as the aim is to build (large) majority governments.Likewise, the fragmentation and polarization of the party system may influence formateurselection. For instance, high fragmentation should favor larger parties, since building a ma-jority cabinet is more difficult in such systems and these parties are more likely to allow theformation of such governments.

Even though Diermeier and Merlo (2004, p. 784) conjecture that “there is no evidencethat the procedures to select a new government differ depending on whether the formationfollows an election or a parliamentary vote of no-confidence”, we argue that a systematicanalysis of the selection outcomes according to the timing of the formation opportunityin the electoral cycle is in order. For instance, inter-election formations frequently occurbecause a surplus coalition loses a member, or because a coalition is enlarged. In theseinstances, the incumbent PM party is therefore more likely to keep its formateur status.Table 1 gives an overview of the hypotheses tested in this paper.4

3.2 Traditional coalition theories predicting government composition

One of the most well-known coalition theories is the cooperative game theoretic minimalwinning solution, which predicts that only coalitions in which all members are necessaryfor the government to control a majority of seats in parliament will form (Von Neumannand Morgenstern 1953). As parties are interested in maximizing their benefits of being inoffice, the addition of a surplus member would be irrational, as coalition partners would haveto share unnecessarily the spoils of office with the surplus member(s). A variation of thistheory was suggested by Leiserson (1968), who predicts that the minimal winning coalitionwith the smallest number of actors will form, since bargaining becomes more difficult asthe number of actors increase. A basic concern of these theories is that coalitions shouldcontrol a majority of seats, as an opposition controlling a seat majority would never toleratethe formation of a minority cabinet as its members are denied any benefits of office (see alsoRiker 1962).

4We also recognize that several authors have stressed the importance of electoral responsiveness, for example,Warwick (1996), and Mattila and Raunio (2004) test, and find support for the idea that, apart from absolutesize, electoral gains and losses influence a party’s chance of becoming the formateur, which we also find.Since the inclusion of this variable restricts the analysis to post-electoral cases, we have here excluded thisvariable.

Public Choice

Table 1 Hypotheses for the two stages of government formation

1st stage hypotheses: formateur selection 2nd stage hypotheses: government composition

FH1: Parties are more likely to become formateursif they are the largest party in the legislature

CH1: Potential governments are more likely toform if they are minimal winning coalitions

FH2: Parties are more likely to become formateursif they held the PMship in the previousgovernment

CH2: Potential governments are more likely toform the fewer the number of parties they include

FH3: Parties are more likely to become formateursif they control the median legislator

CH3: Potential governments are more likely toform if they are majority coalitions

FH4: Median parties are more likely to becomeformateurs in formateur systems

CH4: Potential governments are more likely toform the smaller their policy range

FH5: Incumbent PM parties are more likely tobecome formateurs in free-style systems

CH5: Potential governments are more likely toform if they are incumbent cabinets

FH6: Largest parties are more likely to beformateurs in systems characterized by positiveinvestiture rules, qualified majority requirements,high polarization and high fragmentation

CH6: Potential governments are more likely toform if they include the party predicted to be theformateur on the basis of its size, experience andpolicy position

FH7: Incumbent PM parties are more likely to beformateurs in inter-election governmentformations

CH7: Potential governments are more likely toform the smaller the ideological range between thepredicted formateur and the policy position of thecabinet

Link between 1st and 2nd stage hypothesis

LH: Potential governments are more likely to form according to traditional government compositionpredictions if the formateur party was successfully predicted (through its size, ideological position andexperience) in the first stage

Other authors have argued that we need to consider that parties are also motivated bypolicy-seeking concerns. For example, coalitions are expected to form if they consist ofparties that are adjacent to each other along a specific policy dimension, or the smaller theirideological range (see Axelrod 1970; De Swaan 1973). A more recent idea is that incumbentcabinets are favored in negotiations since they represent “the reversion point in the event theother parties fail to agree on an alternative” (Strøm et al. 1994, p. 311). Incumbents may alsobe advantaged since parties who have cooperated before should have established routines forgoverning together, suggesting that transaction costs increase with a change of partners (see,e.g., Bäck and Dumont 2007).

Linking what bargaining theorists predict with regard to formateur advantages in shap-ing government composition, and what has been found in empirical studies of formateurselection, we hypothesize for our two-stage model of government formation that potentialgovernments including the predicted formateur are more likely to form, and that potentialgovernments with a weighted policy position closer to the formateur’s ideal position aremore likely to form. These hypotheses thus link our first stage analysis predicting who be-comes the formateur to our second stage analysis.

Furthermore, as we assume that a formateur party will try to be successful in forminga government, and will aim to maximize the chances of keeping the benefits of being theleader of a cabinet as long as it can, we expect that the propositions regarding government

Public Choice

composition laid out above will hold in this party’s choices.5 For instance, as a large numberof partners and a wide policy range between the potential coalition partners make bargainingand maintenance more difficult, formateur parties will avoid forming coalitions with suchcharacteristics. For this reason, and as we expect the formateurs correctly predicted in thefirst stage of our analysis to be the best predisposed to successfully compose a new gov-ernment, we hypothesize that these predicted formateurs will form cabinets that conform tothe traditional theoretical predictions regarding full composition. This hypothesis therebyexplicitly concerns the implications of the first stage of our analysis on the second.

4 Methods and data

4.1 Previous methodological approaches in coalition studies

Empirical evaluations of theories that aim at predicting what type of governments will formimply calculating the theory’s success rate, or the proportion of times that a theory’s predic-tion set includes the cabinet that actually formed. Warwick (1996) and followers performan analysis that fixates on the questions of what affects a party’s likelihood of becoming theformateur, and of becoming the formateur’s coalition partner, hence making parties the unitsof analysis. A problem with this approach is that units are not independent since groups ofobservations are parties in the same formation situation. The more straightforward approachwhen we aim to study which party obtains the formateur role, is to model selection as adiscrete choice problem, where the units are the formation opportunities and the partiesare the alternatives, and applying a conditional logit model (McFadden 1973, 1974). Suchan approach has for example been used by Diermeier and Merlo (2004) in their study ofrecognition rules.

To determine which independent variables are most important to include when explain-ing coalitions, we should ideally incorporate variables drawn from different types of theoriesin one multivariate model. An early multivariate test, which allows for this, was performedby Franklin and Mackie (1984). The authors adopt a regression approach where all of thepossible combinations of the parties are seen as units, which implies 2n − 1 potential gov-ernments in each formation situation (where n is the number of parties). A more appropriateapproach was presented by Martin and Stevenson (2001), who model government formationas a discrete choice between the set of all potential combinations of parties that might forma government. Martin and Stevenson assume that the central actors in a system collectivelychoose one of the governments that may form, which makes a conditional logit model anappropriate model.6

This methodological approach however appears problematic as it models governmentformation as a choice made by the system as a whole, and not by the individual parties.Since parties are the actors that make coalitional choices, we would ideally use an approachwhere parties make discrete choices between potential governments. Trying to solve thisproblem, Becker and Shikano (2004) argue for an actor-oriented approach and model gov-ernment formation as a choice for different types of parties (such as “center” or “dominant”parties) between the potential governments in which the party is itself included in. Using

5This is in line with the approach advocated by Diermeier et al. (2006), who argue that expectations aboutcabinet duration are likely to be anticipated at the time of formation, and thereby influence the choices madeat this stage by the formateur party.6See Martin and Stevenson (2001, p. 38) for arguments why the conditional logit model is used.

Public Choice

this approach, the number of potential governments that a specific party can choose from is2(n−1) (where n is again the number of parties).

4.2 A two-stage approach to analyzing government formation

In this paper, we model government formation as a two-stage unordered discrete choiceproblem. Greene (2000, pp. 432–433) describes a two-step estimation procedure where anonlinear regression model depends on a second set of parameters estimated separately. Themodel is described as such:

y = h(x,β,w,γ) + ε, (1)

where y is the outcome, x is a set of variables determining the outcome and β are the re-gressors. In this model, γ is estimated separately in a first step in a model that depends onan additional set of variables w. The parameters γ then enter h(·) through some function ofγ and w. The second step of the analysis then consists of a regression of y on h(x,β,w, c),where c is the first-round estimate of γ.

The goal in the analysis on government formation is to predict y, the outcome of bargain-ing, more precisely, the government composition. In the model predicting this outcome, weinclude a number of features, x, of the governments that could potentially form, such as avariable indicating whether the potential government is a minimal winning coalition. Sincethe formateur should be able to significantly shape the coalition outcome, we would, in thismodel, also like to include information on which party is the formateur, which can here bedescribed by γ. But simply including information on which party is the formateur by lookingat the post-formation result would not suffice, since this implies that we are using informa-tion on the final result to predict the outcome. Thus, we have to use some other information,w, to predict the formateur, and then use the predictions made in this first-stage analysis,giving us an estimate c, which is included in our model of government composition (y).

Thus, a first analysis should be performed aimed at predicting which party becomes theformateur. This analysis is straightforward. We create a dataset where the units are the for-mation opportunities, and the choice alternatives are the different parties, and apply a condi-tional logit model.7 In this first-stage analysis we can compute the predicted probabilities foreach party becoming the formateur, which can be used in the second-stage analysis. In thisstudy, the latter is structured both using Stevenson’s (2001) traditional approach, and usingan actor-oriented approach, which enables us to study the choices of the predicted formateur(or the actual PM).8

7In the conditional logit model, the probability that individual i chooses alternative j is:

Pr(Yi = j) = eβzij

J∑

j=1eβzij

,

where j = 1,2, . . . , J for a total of J alternatives (Greene 2000, p. 862). In our applications, i is the systemat a point in time, and j are the different parties, or the potential governments.8Optimally we would like to statistically model the two stages as interdependent, e.g., using a similar ap-proach of analyzing strategic interaction as presented by Signorino (1999). Problems of application in thissetting however abound: first, we are here dealing with a two-stage game where one actor (e.g., the Head ofState) is making a choice in the 1st stage, and another actor (the formateur) is then interacting with otheractors (the other parties); second, we are dealing with a large number of choice alternatives, and the numberof choices vary across formation opportunities; finally, we are here dealing with some theories that make

Public Choice

4.3 Data and operationalizations

We use a dataset consisting of cabinets formed in 12 Western European democracies be-tween 1970 and 2006. The countries and years in our sample are: Austria 1970–2002, Bel-gium 1972–2003, Denmark 1971–2005, Finland 1972–2003, Germany 1972–2005, Iceland1971–2006, Ireland 1973–2002, Italy 1972–1993, Luxembourg 1974–2004, Netherlands1971–2003, Norway 1973–2005, and Sweden 1970–2002.9 The sample includes 176 for-mation opportunities, and over 160,000 potential governments.

The main problem in most analyses aimed at predicting which party becomes the ac-tual formateur is to determine which parties are the formateurs in the different stagesof the bargaining process. Most studies of formateur selection (see, e.g., Warwick 1996;Ansolabehere et al. 2005; Martin and Stevenson 2001) have used PM status as a proxy offormateur status, since this type of information is typically easily attained. If the first for-mateur is successful in forming a government, the formateur party will (most likely) indeedbe the PM. But if the first party selected to form a government is not successful, a secondformateur, who may be the same individual as in the first step or not, and belong to the sameparty or not, is selected to become the formateur and may thus end up as the Prime Minister.The process may take several rounds, with several formateurs who do not become the PrimeMinister, before a cabinet is eventually formed.10

We have here chosen to perform two types of analyses, one where the PM of the gov-ernment that actually formed is coded as the formateur party (since the PM is typically thefinal formateur), and one analysis where we use the first formateur as dependent variable.This variable is coded by the authors on the basis of the dataset developed for the Coali-tion Governance Project, thus abiding by the coding instructions used in Müller and Strøm(2003)—where one party is publicly recognized (either appointed by the Head of State ornot), as trying to form a cabinet, and negotiating parties are clearly identified—updated andsupplemented by information taken from Keesing’s Record of World Events, various issuesof the European Journal of Political Research Political Data Yearbook and by country ex-perts of the Coalition Governance Project.

For the independent variables, we use the dataset referred to above to create a variablethat describes which party is the largest party in parliament and which party held the PMpost in the previous cabinet (previous PM). In order to gauge whether median parties aremore likely to become formateurs, we need information on parties’ policy positions alongsome key dimension. This key policy dimension is in most coalition studies assumed to bethe left–right economic policy dimension. To measure parties’ ideal policy positions alongthis dimension, we here rely on data derived from two of the expert surveys that have beenadministered during the last decades; the Laver and Hunt (1992) expert survey administeredin 1989 (used only for Luxembourg), and the Huber and Inglehart (1995) survey adminis-tered in 1994.11

multiple equilibrium predictions, which makes it difficult to assign probabilities over outcomes (Signorino1999, p. 294). An alternative statistical approach would be to use some sort of nested model, which wouldallow for sequential choices (see, e.g., Greene 2000).9We count a change of cabinet whenever: (1) general elections are held, (2) there is a change in the partycomposition of the cabinet, (3) there is a change of Prime Minister, and we only consider minority situations(where no single party holds a majority of seats).10Coalition scholars such as Michael Laver (2006) have stressed the problem of endogenously coded forma-teurs in datasets built to estimate formateur effects in pay-off allocations.11A drawback with using these data instead of manifesto data (the main available alternative), is that theexpert survey data do not vary over time, that is, we have to assume that the parties’ policy positions have

Public Choice

In the analysis of formateur selection we also include a number of variables that de-scribe the context within which government formation takes place. These variables only varyacross formation opportunities and thus have to be interacted with a choice-specific variable(here, the largest party variable) in order to be included in the conditional logit analysis.Positive investiture denotes whether there is a strong norm or a formal rule that says thata new government should receive support from a majority in the parliament. We also in-clude a variable indicating if there are qualified majority requirements. To measure partyfragmentation, we use a variable controlling for the effective number of parties (Laakso andTaagepera 1979), and we measure polarization as the absolute distance between the partieson the left–right extremes.

As argued above, the role of the Head of State in setting the agenda of cabinet formation,and his/her overall influence on the process, varies amongst countries as well as across time.In some systems, the Head of State acts at least in part as an exogenous constraint on theprocess of government formation, and appoints a formateur that he/she considers most apt toform a cabinet, which we refer to as formateur systems. In other systems, so called free-stylebargaining systems, the parties are not constrained in their initiatives by the prerogatives ofthe Head of State. A number of comparative descriptions and in-depth studies of governmentformation processes allow us to provide a classification of our 12 countries according to thisdistinction.

Laver and Schofield (1990, p. 64) and later Strøm et al. (1994, p. 312) point to the casesof Finland, Italy, and the Netherlands, where the Head of State plays a significant role ingovernment formation, compared to the Heads of State in Ireland, Germany and Swedenwho play no part, or only a formal part, in the process. According to De Winter’s (1995,p. 126) survey of country specialists, among (semi-)presidential systems the impact of theHead of State on the nomination of the PM designate is greatest in Finland and lowest inAustria. Following Müller (2003), no effective intervention of the Head of State occurredin the period under consideration in the present study, allowing us to consider Austria ashaving a free-style bargaining system. On the other hand, Iceland, which was ranked inan intermediate position by De Winter (1995), belongs to the formateur systems. Finally,among constitutional monarchies, De Winter’s (1995) survey indicates that Norwegian andDanish Heads of State are formally involved through constitutional provisions, but have noimpact on cabinet formation. In contrast, the Benelux countries share quite similar custom-ary prerogatives regarding the input of their monarchs in the formation process, and practicereveals that they belong to the formateur type of systems. We therefore identify five coun-tries with a formateur system and six with a free-style system. Finland is coded differentlythrough time, as it switches in 2000 from a formateur to a free-style type.

In the second-stage analysis, we include several variables describing the potential govern-ments. We include three variables aimed at evaluating the traditional office-seeking theories(minimal winning, number of parties,12 and minority governments13), using information on

stayed stable between 1970 and 2000. There are however several reasons for using expert surveys instead ofmanifesto data. One is that party manifestos are likely to reflect the image that parties want to give to the votersrather than their “true” policy positions, and that they may also anticipate future coalitions when writing theirpolicy program, which makes manifesto data unsuitable to use when studying coalitions. Moreover, eventhough the main advantage of using party manifesto data is to provide measures for each election year, Budgeand Klingemann (2001) find that in the post-war era, parties’ locations remain fairly stable and they rarelyleapfrog each other on a left–right scale.12Note that although Leiserson suggested that amongst MWC those with the smallest number of partiesshould form, we use the same variable as Martin and Stevenson (2001).13We give minority governments the value 1 (see Martin and Stevenson 2001).

Public Choice

government composition and distribution of seats provided by Müller and Strøm (2003).From the same source we also draw an incumbent cabinet variable indicating whether thepotential government was the previous cabinet. Using the data on parties’ policy positions,we can identify the policy range of a potential coalition measured as the distance betweenthe parties on the extremes.

In the second stage we also include some variables aimed at testing whether the formateuris advantaged in government formation. First, we include a variable created on the basis ofthe predictions made in the first-stage analysis, that describes if the party that we predictfor formateur status is included in the potential government (predicted formateur included).We also create a variable describing the predicted probability to be formateur for the partywith the highest predicted probability, thereby assuming that the party which is most likelyto be formateur in a specific potential government would indeed be the one leading to itsformation. Lastly, we include a measure describing the ideological distance between theformateur and the parties in the potential government (distance formateur), by taking theabsolute distance between the formateur’s position and the government’s weighted policyposition.

5 A two-stage analysis of government formation

5.1 Stage 1: An analysis of formateur selection

In Table 2 we present a number of multivariate analyses aimed at evaluating what featuresdetermine who becomes the formateur. The dependent variable either describes whether ornot a party was selected for the Prime Minister post, or if a party was selected to be thefirst formateur. These two variables are highly correlated, but in cases where there havebeen several formation attempts (in about a third of all cases), the first formateur and thePM may come from different parties (in roughly half of these cases).14 In the tables wepresent unstandardized conditional logit coefficients, which indicate if a variable increasesor decreases the likelihood that a party will become formateur.15

Model PM1 is our base model, including only the most important actor-specific vari-ables, the largest party, the previous PM, and the median party variables. The effect of allthree variables are positive and significant, suggesting that these features are important toconsider when we aim at explaining formateur selection. To fully evaluate our models wealso generate predicted probabilities from each of them and predict that the party with thehighest predicted probability in each formation opportunity will become the formateur. Thispoint prediction is then compared to the party that actually became the formateur, and weare thus given a prediction rate for each model. Model PM1, including our base variables,correctly predicts 70% of the PM parties.

We also evaluate the base model applied to five different samples, first comparing theeffects of our main variables across formateur and free-style systems. One important dif-ference is the results for the previous PM variable, which exerts a positive and significanteffect in free-style systems (model PM3), whereas the effect in formateur systems is not

14Amongst post-electoral formations, failed attempts amount to 36.5% and the proportion of multiple forma-tion attempts with formateurs belonging to different parties is roughly one half. The occurrence of multipleformation attempts involving different parties at the helm is underestimated and too rapidly discarded by mostempirical studies of formateur selection.15The conditional logit model is nonlinear. Thus, the unstandardized coefficients cannot be directly inter-preted as changes in probabilities when we alter a choice-specific variable.

Public Choice

Tabl

e2

Con

ditio

nall

ogit

with

PMpa

rty

orfir

stfo

rmat

eur

asde

pend

entv

aria

ble

(1st

stag

e)

PMpa

rty

asde

pend

entv

aria

ble

Firs

tfor

mat

eur

asde

pend

entv

aria

ble

Mod

elP

M1

all

case

s

Mod

elP

M2

form

a-te

ursy

stem

s

Mod

elP

M3

free

-st

yle

syst

ems

Mod

elP

M4

elec

-to

ral

case

s

Mod

elP

M5

inte

r-el

ectio

n

Mod

elP

M6

faile

dat

tem

pts

Mod

elP

M7

all

case

s

Mod

elP

M8

form

a-te

ursy

stem

s

Mod

elP

M9

free

-st

yle

syst

ems

Mod

elF

1al

lca

ses

Mod

elF

2fo

rma-

teur

syst

ems

Mod

elF

3fr

ee-

styl

esy

stem

s

Mod

elF

4el

ec-

tora

lca

ses

Mod

elF

5in

ter-

elec

tion

Lar

gest

part

y2.

19**

*

(0.2

6)2.

79**

*

(0.3

4)1.

90**

*

(0.2

9)2.

21**

*

(0.2

6)2.

23**

*

(0.3

5)2.

93**

*

(0.4

2)−0

.69

(0.9

4)−1

.99

(2.1

1)−0

.02

(1.9

6)1.

81**

*

(0.2

0)2.

05**

*

(0.2

9)1.

76**

*

(0.2

8)1.

84**

*

(0.2

4)2.

00**

*

(0.3

7)

Prev

ious

PM1.

04**

*

(0.2

3)0.

47(0

.34)

1.20

***

(0.2

9)0.

74**

(0.2

9)1.

64**

*

(0.3

9)−0

.16

(0.4

0)0.

95**

*

(0.2

3)0.

37(0

.37)

1.13

***

(0.3

0)1.

24**

*

(0.2

1)0.

82**

(0.3

2)1.

33**

*

(0.2

8)0.

90**

*

(0.2

7)1.

80**

*

(0.3

7)

Med

ian

part

y1.

15**

*

(0.2

3)2.

08**

*

(0.3

7)0.

44(0

.34)

1.20

***

(0.2

8)0.

91**

(0.4

4)2.

39**

*

(0.4

3)1.

36**

*

(0.2

6)2.

32**

*

(0.3

9)0.

61(0

.39)

0.88

***

(0.2

2)1.

52**

*

(0.3

1)0.

25(0

.34)

0.65

**

(0.2

7)1.

29**

*

(0.4

2)

Pos.

inve

stitu

re×

larg

est

––

––

––

−0.7

6*

(0.4

4)−2

.31**

(0.9

7)−0

.13

(0.9

2)–

––

––

Qua

l.M

ajor

ity×

larg

est

––

––

––

0.75

*

(0.4

1)0.

51(0

.66)

0.26

(0.6

0)–

––

––

Eff

.no

part

ies

×la

rges

t

––

––

––

0.19

(0.1

7)0.

40(0

.29)

−0.4

7(0

.43)

––

––

–

Pola

riza

tion

×la

rges

t–

––

––

–3.

03**

(1.4

4)5.

78*

(3.0

6)5.

68*

(3.2

9)–

––

––

Public Choice

Tabl

e2

(Con

tinu

ed)

PMpa

rty

asde

pend

entv

aria

ble

Firs

tfor

mat

eur

asde

pend

entv

aria

ble

Mod

elP

M1

all

case

s

Mod

elP

M2

form

a-te

ursy

stem

s

Mod

elP

M3

free

-st

yle

syst

ems

Mod

elP

M4

elec

-to

ral

case

s

Mod

elP

M5

inte

r-el

ectio

n

Mod

elP

M6

faile

dat

tem

pts

Mod

elP

M7

all

case

s

Mod

elP

M8

form

a-te

ursy

stem

s

Mod

elP

M9

free

-st

yle

syst

ems

Mod

elF

1al

lca

ses

Mod

elF

2fo

rma-

teur

syst

ems

Mod

elF

3fr

ee-

styl

esy

stem

s

Mod

elF

4el

ec-

tora

lca

ses

Mod

elF

5in

ter-

elec

tion

N (par

ties)

1,37

382

454

986

351

045

61,

373

824

549

1,37

382

454

986

351

0

Pseu

doR

20.

530.

620.

450.

490.

610.

520.

550.

650.

460.

460.

510.

430.

390.

63

Log

likel

ihoo

d−1

65.5

3−7

3.70

−84.

89−1

13.6

0−4

9.96

−54.

30−1

58.9

7−6

9.52

−82.

51−1

88.5

1−9

6.40

−87.

96−1

36.5

8−4

7.47

Pred

ictio

nra

te0.

700.

750.

710.

720.

770.

650.

710.

760.

670.

680.

690.

680.

720.

77

Not

e:Si

gnifi

cant

at*

the

0.10

leve

l,**

the

0.05

leve

l,**

*th

e0.

01le

vel.

Para

met

eres

timat

esar

eun

stan

dard

ized

cond

ition

allo

gitc

oeffi

cien

ts.E

ntri

esin

pare

nthe

ses

are

stan

dard

erro

rs.P

redi

ctio

nra

tes

are

obta

ined

byge

nera

ting

pred

icte

dpr

obab

ilitie

s(u

sing

STA

TA’s

pred

ictc

omm

and)

from

each

mod

el.A

ssum

ing

that

the

part

yw

ithth

ehi

ghes

tpre

dict

edpr

obab

ility

will

beco

me

the

form

ateu

r,th

ispo

intp

redi

ctio

nis

com

pare

dto

the

actu

alou

tcom

e

Public Choice

significant (PM2). Hence, despite the fact that Heads of State may have built relations basedon trust with the former PM, it appears that there is no previous PM advantage in forma-teur systems. Either these relations turned out to be conflictual, and the non-selection of theprevious PM is intended by the Head of State, or other forces constrain the latter’s choices.Another important difference is the fact that the median party variable exerts a significanteffect in formateur systems, but not in free-style systems. This is in line with Morelli’s pre-diction that the Head of State will select parties that are likely to form a cabinet that reflectsthe median voter’s preferences.

In Table 2, we also present a model applied to only inter-election cases, i.e., cases whichare governments formed between elections (PM5), and a model applied only to cases wherethere have been several attempts to form a government (PM6). By comparing these modelsto PM4, the base model applied to post-electoral cases, we are given some indication as towhat factors influence formateur selection in distinct bargaining circumstances. The onlyvariable that has a consistent effect across all models is the largest party variable. The pre-vious PM variable exerts a significant effect in both the post-electoral model and the inter-election model, but being the previous PM makes a party more likely to become the partyleading the new government only when the latter forms without there being a failed forma-tion attempt. This suggests that the previous PM often is selected to be the first formateur,but if this party fails to form a cabinet, it is not more likely to become the PM party than anyother. We should also note that the effect of the previous PM variable is, as hypothesized,stronger for inter-election cases than for post-electoral ones. Conversely, median party statusseems to be more important when a government forms directly after an election and after afailed formation attempt, than when a cabinet forms between elections. Hence, the factorsdetermining formateur selection differ according to the timing of cabinet formation.

In Table 2 we also present results regarding the selection of a party as first formateur(models F1–F5), regardless of whether the latter will be successful in forming a governmentor not. Because the dependent variable is not dramatically different than the one studiedabove, most results are similar. Interesting nuances emerge however: the previous PM vari-able has a greater effect than for PM selection in some models, with the most importantdifference being found for formateur systems—the effect of the previous PM variable issignificant for the choice of the first formateur in these systems. Median party status issomewhat less important. Reconstructing a plausible sequence from these results, the previ-ous PM seems to be given the first chance of forming a government. If it fails in its mission,other factors such as median party status are likely to drive formateur selection. Overall, notonly individual variables reveal nuances: the prediction rate and the Pseudo-R2s are slightlylower in models F1 to F3 (about 68% predicted), suggesting that, using information on par-ties’ size, experience and policy positions, it is more difficult to predict which party startsthe bargaining process.

In Table 2, we also present three models including contextual features interacted withthe most important party characteristic (size). Model PM7 is based on the full sample offormation opportunities and should thus be compared to model PM1, whereas models PM8and PM9 should be compared to models PM2 and PM3 respectively. In these “contextual”models we include four interaction terms: the largest party variable interacted with positiveinvestiture, qualified majority, effective number of parties and polarization. Two of the inter-action terms are positive and significant when studying the entire sample (PM7), qualifiedmajority× largest, and polarization× largest, but positive investiture× largest is on the otherhand negative and significant. Thus, large parties are likely to become the PM in systemswith qualified majority requirements and with a high polarization, but less so in countrieswhich use a positive investiture vote to install new governments, a finding that runs counterto common expectations.

Public Choice

Focusing on different samples, we see that largest parties seem to be advantaged in polar-ized systems whatever the process of government formation in use, but the size of the partysystem does not exert any significant effect. In formateur systems (PM8), the largest party isless likely to be the PM when the country uses a positive investiture vote procedure, whereasthe effect of the positive investiture× largest is not significant in the free-style sample. Over-all, contextual factors seem to be more important when aiming to predict who becomes theformateur in systems where the Head of State is involved in selection, as is also indicated bythe fact that the prediction rate of our model is much higher for the formateur sample thanthe free-style sample (76% compared to 67%).

Hence, we find that what matters for formateur selection are parties’ size, previous PMexperience, and ideological positions. Size clearly dominates formateur selection: it is highlyprobable that the largest party becomes the formateur. Further, studying different samplesand dependent variables allows us to reveal previously undocumented variations in the ex-planatory strength of these three party features.

5.2 Stage 2: An analysis of government composition

We now turn to the second stage of our analysis, focusing on predicting the governmentcomposition. In Table 3 we present several conditional logit models where the potentialgovernments are the choices.16 By generating predicted probabilities from our models andpredicting that the government with the highest predicted probability will form, we canassess our ability to predict Western European governments.17 In model 1 we include fivevariables aimed at evaluating traditional theories. We find that governments are more likelyto form if they are minimal winning, if they control a majority of seats, the smaller thepolicy range between the parties, and if they are incumbent cabinets. The number of partiesincluded exerts an effect in the expected direction, but this effect is not significant. Model 1enables us to predict 31% of the governments.

In model 2 we include a variable indicating whether the party we predicted to be theformateur in the first stage is included in the potential government. This variable exerts asignificant, positive effect, and the prediction rate is increased by 4 percentage points (to35%) when the variable is added, suggesting that we are better able to explain the coalitionoutcome if we consider that some features make parties more likely to become the formateur.In model 3 we include the three party features used to predict formateur status. The medianparty and the largest party variables exert positive and significant effects, suggesting thatwhen these types of parties are included in a government, it is more likely to form. The ad-dition of these variables increases the prediction rate by 2 percent only, which suggests that,on its own, our predicted formateur variable performs better than the party features mea-

16A possible drawback to using the conditional logit model is that the model imposes the property of inde-pendence of irrelevant alternatives (IIA), which may be problematic here since the choices between potentialgovernments are not likely to be independent of each other. Following Martin and Stevenson (2001), we eval-uate whether the IIA assumption is problematic by dropping a random set of alternatives from each formationopportunity and applying a Hausman test. We here drop 50% of the choices and then apply the Hausmantest. Looking at the average p-value over 50 replications we find that it for none of the models is lower than0.20, which suggests that we cannot reject the null hypothesis that IIA holds. Thus, assuming independenceof irrelevant alternatives does not seem to be problematic in this application. We thank Martin and Stevensonfor providing a program to perform this test.17As some models make multiple equilibrium predictions we present predictive efficiency rates (the totalnumber of predictions is the denominator rather than the number cases).

Public Choice

Tabl

e3

Con

ditio

nall

ogit

offu

llgo

vern

men

tcom

posi

tion

(2nd

stag

e)

Syst

emch

oose

sbe

twee

ngo

vern

men

tsFo

rmat

eur

choo

ses

betw

een

gove

rnm

ents

Mod

el1

Bas

em

odel

Mod

el2

Pre-

dict

edfo

rma-

teur

Mod

el3

Part

yfe

atur

es

Mod

el4

Prob

abil-

ity form

a-te

ur

Mod

el5

Pre-

dict

edsu

b-sa

mpl

e

Mod

el6

Part

yfe

atur

es

Mod

el7

Dis

tanc

efo

rma-

teur

Mod

el8

Dis

tanc

ePM

part

y

Mod

el9

Form

a-te

ursy

stem

s

Mod

el10

Free

-st

yle

syst

ems

Min

imal

win

ning

0.73

***

(0.2

7)0.

84**

*

(0.2

7)0.

86**

*

(0.2

7)0.

84**

*

(0.2

7)0.

28(0

.34)

0.19

(0.3

5)0.

07(0

.33)

0.38

(0.2

7)0.

30(0

.31)

2.66

**

(1.0

7)

Num

ber

ofpa

rtie

s−0

.17

(0.1

2)−0

.14

(0.1

2)−0

.20

(0.1

3)−0

.13

(0.1

2)−0

.37**

(0.1

5)−0

.45**

*

(0.1

7)−1

.45**

*

(0.1

4)−1

.25**

*

(0.1

1)−1

.30**

*

(0.1

4)−1

.20**

*

(0.2

3)

Min

ority

gove

rnm

ent

−2.5

5***

(0.3

4)−1

.89**

*

(0.3

6)−1

.51**

*

(0.3

6)−1

.88**

*

(0.3

6)−3

.36**

*

(0.4

4)−2

.07**

*

(0.4

6)−1

.84**

*

(0.4

3)−1

.55**

*

(0.3

4)−2

.87**

*

(0.4

9)1.

85(1

.13)

Polic

yra

nge

−8.9

6***

(0.8

5)−9

.12**

*

(0.8

7)−8

.85**

*

(0.9

1)−9

.12**

*

(0.8

7)−9

.90**

*

(1.1

3)−1

0.51

***

(1.2

5)–

––

–

Incu

mbe

ntca

bine

t3.

00**

*

(0.2

1)2.

68**

*

(0.2

1)2.

78**

*

(0.2

3)2.

65**

*

(0.2

2)3.

52**

*

(0.2

4)2.

93**

*

(0.2

7)3.

38**

*

(0.2

8)3.

29**

*

(0.2

5)2.

89**

*

(0.3

6)3.

15**

*

(0.3

9)

Pred

icte

dfo

rm.

incl

uded

–1.

08**

*

(0.2

4)–

––

––

––

–

Lar

gest

part

yin

clud

ed

––

1.12

***

(0.2

5)–

–−0

.34

(0.8

9)−0

.51

(0.8

8)−0

.95**

(0.4

7)−1

.10**

(0.5

6)−1

4.71

(795

.43)

Prev

ious

PM incl

uded

––

−0.1

1(0

.25)

––

−0.8

3(0

.59)

−0.6

0(0

.57)

−1.0

2***

(0.3

9)−0

.72

(0.5

1)−1

.40

(0.9

5)

Public Choice

Tabl

e3

(Con

tinu

ed)

Syst

emch

oose

sbe

twee

ngo

vern

men

tsFo

rmat

eur

choo

ses

betw

een

gove

rnm

ents

Mod

el1

Bas

em

odel

Mod

el2

Pre-

dict

edfo

rma-

teur

Mod

el3

Part

yfe

atur

es

Mod

el4

Prob

abil-

ity form

a-te

ur

Mod

el5

Pre-

dict

edsu

b-sa

mpl

e

Mod

el6

Part

yfe

atur

es

Mod

el7

Dis

tanc

efo

rma-

teur

Mod

el8

Dis

tanc

ePM

part

y

Mod

el9

Form

a-te

ursy

stem

s

Mod

el10

Free

-st

yle

syst

ems

Med

ian

part

yin

clud

ed

––

1.05

***

(0.2

3)–

–1.

01**

*

(0.3

8)1.

61**

*

(0.3

7)1.

46**

*

(0.2

8)1.

53**

*

(0.4

5)1.

63**

*

(0.4

2)

Pred

icte

dpr

ob.

form

ateu

r

––

–1.

74**

*

(0.3

8)–

––

––

–

Dis

tanc

efo

rmat

eur

––

––

–−7

.43**

*

(2.7

1)−5

.28**

*

(1.9

9)−2

.24

(2.4

9)−5

.93

(3.8

8)

N(p

ot.g

ov-

ernm

ents

)15

9,46

315

9,46

315

9,46

315

9,46

312

9,58

164

,828

64,8

2879

,784

71,8

487,

936

Pseu

doR

20.

360.

370.

390.

380.

450.

490.

420.

360.

290.

58

Log

likel

ihoo

d−5

73.4

2−5

62.7

8−5

48.0

0−5

62.2

7−3

58.5

6−2

88.6

3−3

32.2

3−5

02.7

4−3

43.2

6−1

24.4

7

Pred

ictiv

eef

ficie

ncy

rate

0.31

0.35

0.33

0.33

0.35

0.48

0.44

0.39

0.28

0.54

Not

e:Si

gnifi

cant

at*

the

0.10

leve

l,**

the

0.05

leve

l,**

*th

e0.

01le

vel.

Para

met

eres

timat

esar

eun

stan

dard

ized

cond

ition

allo

gitc

oeffi

cien

tsw

ithst

anda

rder

rors

inpa

rent

hese

s

Public Choice

sured separately.18 In model 4 we include a slightly different predicted formateur variable,describing the predicted probability of the party with the highest probability of becomingthe formateur (in a potential government). This variable exerts a significant positive effect,suggesting that potential governments are more likely to form if they include the partiesmost likely to become the formateur. 19

Finally, we proposed that the explanatory strength of traditional theories would be greaterwhen we managed to predict the formateur successfully in the first stage. Restricting thesample to these cases in model 5, the predictive efficiency rate rises to 35%, suggesting in-deed that predicted formateurs tend to form governments that are more accurately explainedby classical coalition theories—one unexpected result being however that, contrary to mod-els 1 to 4, the effect of the minimal winning variable is here far from significant. Amongstformation opportunities for which the base model of the first stage did not manage to findthe actual formateur, only 9% of the cabinets are correctly predicted. The characteristics ofthe formateur selected in the first stage thus appear to have important implications for thepredictability of cabinet composition.

In models 6 and 7 we use a restricted dataset, concentrating on the governments includingthe correctly predicted formateur, and thus focusing on the formateur’s choices. In model 6we include the traditional coalition variables and the party characteristics to gauge whetherthe latter have an effect in the second stage. The results differ from those of model 3 withregard to three variables: whilst the effect of number of parties included is now highly sig-nificant, the effects of the minimal winning variable and the largest party dummy are nowinsignificant. Hence, mostly because the predicted formateur is very often the largest party,this variable does not exert a significant effect anymore. Had we not found that the minimalwinning variable had no significant effect in model 5, we could have conjectured that asthe predicted formateur is typically a very large party, minimal winning coalitions are over-represented in this restricted sample where the formateur is included. Instead, this resultconfirms the unexpected finding of model 5. With this model, we predict almost half (48%)of the governments.

These results hold when we add the variable measuring the distance between the forma-teur and the weighted policy position of the parties in the government (model 7). The effectof the distance to formateur variable is negative and significant, giving support to the ideadrawn from non-cooperative models that formateur parties manage to build governmentsthat are ideologically close to their ideal position.

In order to fully compare our results with previous research, we performed an analysiswhere we study the choices of the actual PM party (see Martin and Stevenson 2001). Theresults of this analysis are presented in model 8, and they are largely similar to the resultsfound in our analysis of the predicted formateur, but being the largest party or the previous

18In a model where we control for the presence of the median party or the previous PM in cabinet, the effectof the predicted formateur variable is still significant, but when adding a variable describing if the largestparty is included, the effect of the formateur variable fails to be significant, probably due to the fact that thesevariables are highly correlated as party size dominates the first stage predictions. If other variables were moreinfluential at the 1st stage, we could perhaps disentangle the effects of the predicted formateur and largestparty variables.19We also ran these models on different samples, according to the institutionalization of the formationprocess. The performance of the predicted formateur variables are similar in both samples—these variablesexert significant effects. Note however that the predictive efficiency rate is as low as 19–24% in formateursystems but reaches 42–44% in free-style systems (which is due to differences in the size of the party systems;the average number of potential governments per formation opportunity is 1570 for the former and 191 forthe latter).

Public Choice

PM party is a liability (both coefficients are negative and significant) unless these parties areselected to be the PM. Hence, when formateur status is not predicted by size, median partystatus and PM experience, the formateur party avoids inviting the largest or previous PMparty as coalition partners.

Thus, as in previous research, we find that several features of the potential cabinets, suchas their size, policy completion and incumbency status, influence their likelihood of form-ing. More importantly, we find that it is useful to add information drawn from a first-stageanalysis focusing on formateur selection when we aim to predict the government outcome.Potential governments that include the party that we in the first stage predicted to becomethe formateur are more likely to form. Also, coalitions seem to be more likely to form ifthey contain parties close to the formateur, suggesting that the formateur is able to shape theideological profile of the cabinet to his/her liking.

However, the latter finding has to be qualified since it proved to be non-robust when putto further tests, presented in models 9 and 10. When analyzing the model in the two samplesdistinguishing formateur from free-style systems, we see that the distance to formateur vari-able only comes close to significance (but still fails to be significant) in free-style systems(model 10). As these systems are characterized by a small number of parties and frequentsingle-party minority governments, this result is hardly surprising. In systems where theHead of State selects the formateur (model 9), the effect of the distance variable, on theother hand, is smaller, and completely fails to be significant.

6 Conclusions

In this paper, we have modeled government formation as a two-stage unordered discretechoice problem, where the first step involves the selection of a formateur, and the secondstep involves the choice of government composition by the formateur.

We find that what matters for formateur selection is predominantly parties’ size. We alsofind that parties’ governmental experience and ideological positions are important featuresto take into account when explaining formateur selection, and that some contextual featuresinfluence selection. Most notably, party size and median party status seem to matter morein formateur systems than in free-style systems, whereas experience of being the previousPM matters more in free-style systems. We also confronted the determinants of first and lastformateur selection, and investigated differences in formateur selection owing to the post-electoral character of the case. Our results suggest that experience as previous PM mattersmore in the selection of the first formateur, but if these parties fail, large and centrally locatedparties, who are most likely to be successful in building a government, are taking over thischarge.

Regardless of how we model the 2nd formation stage, we find support for our two-stageapproach in that potential governments are more likely to form if they include the predictedformateur, and that we are better able to predict coalition outcomes if we focus on the poten-tial governments where this party is included. Also, even though like Warwick (2001) andMartin and Stevenson (2001) we find that potential governments are more likely to formif they contain parties that are located ideologically close to the formateur, these resultshave to be qualified, since we show that such a formateur advantage does not materializeat all in systems where the Head of State is effectively involved in the selection of the firstmover.20 Overall our results indeed stress the importance of taking the institutionalization of

20Although not focused on this specific point, our study thus adds to our understanding of the power of theformateur in government formation and to the debate over evaluations of the formateur advantage using other

Public Choice

the bargaining process into account, since results for both stages differ substantially acrossfree-style and formateur systems. This is a finding not documented in previous research.

We therefore shed new light on government formation, accounting for its process andthe fact that some formation attempts fail. More research is however needed to improvemeasurement of crucial variables and understand why some bargaining attempts fail whilstothers do not. Also, even though we have here made progress in terms of empirically mod-elling government formation as a two-stage process, further advancements can be madeby developing the statistical approach used to evaluate theories of government formation.In this article we have used the standard framework developed by Martin and Stevenson(2001). An alternative would be to use some sort of nested approach, which allows for se-quential choices, modelling the first stage as a choice between “branches” (see e.g., Greene2000). This would fully endogenize formateur selection, and would make it possible to takeinto account that characteristics of the potential governments may influence the choice offormateur.

Finally, in line with a recent theoretical approach arguing that government type and ex-pected duration are selected jointly in equilibrium, we find that government compositionoutcomes are more predictable when the identity of the formateur is itself easy to predict.When the formateurs that are most likely to succeed in their task are entitled to make thefirst move, they tend to form governments that appear to be the easier to put together andthe most likely to last. Although more work is still to be done along this line of reasoning,we argue that by modelling the formation process more accurately, we also contribute to thedebates over the next steps of theory improvement.

References

Ansolabehere, S., Snyder, J. M. Jr, Strauss, A. B., & Ting, M. M. (2005). Voting weights and formateuradvantages in the formation of coalition governments. American Journal of Political Science, 49, 550–563.

Austen-Smith, D., & Banks, J. S. (1988). Elections, coalitions and legislative outcomes. American PoliticalScience Review, 82, 405–422.

Axelrod, R. (1970). Conflict of interest. Chicago: Markham Publishing Company.Bäck, H., & Dumont, P. (2007). Combining large-n and small-n strategies. The way forward in coalition

research. West European Politics, 30, 467–501.Baron, D. P. (1991). A spatial bargaining theory of government formation in a parliamentary system. Ameri-

can Political Science Review, 83, 1181–1206.Baron, D. P., & Ferejohn, J. A. (1989). Bargaining in legislatures. American Political Science Review, 83,

1181–1206.Becker, A., & Shikano, S. (2004). Empirische Überprüfung von akteurszentrierten Koalitionsmodellen mit

KOALA. Paper presented at AK Handlungs- und Entscheidungstheorie, Bamberg, 9–10 July 2004.Bergman, T. (1995). Constitutional rules and party goals in coalition formation. Umeå: Umeå University

Press.Budge, I., & Klingemann, H.-D. (2001). Finally! Comparative over-time mapping of party policy movement.

In I. Budge, H.-D. Klingemann, A. Volkens, J. Bara, & E. Tannebaum (Eds.), Mapping policy prefer-ences (pp. 51–74). Oxford: Oxford University Press.

Crombez, C. (1996). Minority governments, minimal winning coalitions and surplus majorities in parliamen-tary systems. European Journal of Political Research, 29, 1–29.

De Swaan, A. (1973). Coalition theories and cabinet formations. Amsterdam: Elsevier.De Winter, L. (1995). The role of parliament in cabinet formation and resignation. In H. Döring (Ed.), Par-

liaments and majority rule in Western Europe (pp. 115–151). New York: St. Martin’s Press.

measures of bargaining outcomes than the mere allocation of portfolios (see Warwick and Druckman 2001,2006; Ansolabehere et al. 2005).

Public Choice

De Winter, L., & Dumont, P. (2006). Parties into government: Still many puzzles. In R. S. Katz & W. Crotty(Eds.), Handbook of party politics (pp. 175–188). London/Thousand Oaks/New Dehli: Sage Publica-tions.

Diermeier, D., & Merlo, A. (2004). An empirical investigation of coalitional bargaining procedures. Journalof Public Economics, 88, 783–797.

Diermeier, D., Eraslan, H., & Merlo, A. (2006). The effects of constitutions on coalition governments inparliamentary democracies. In R. D. Congleton & B. Swedenborg (Eds.), Democratic constitutionaldesign and public policy: analysis and evidence (pp. 133–162). Cambridge: MIT Press.

Franklin, M. N., & Mackie, T. T. (1984). Reassessing the importance of size and ideology for the formationof governing coalitions in parliamentary democracies. American Journal of Political Science, 28, 671–692.

Fréchette, G., Kagel, J. H., & Morelli, M. (2005). Gamson’s law versus non-cooperative bargaining theory.Games and Economic Behavior, 51, 365–390.

Greene, W. H. (2000). Econometric analysis (4th ed.). Upper Saddle River: Prentice-Hall.Huber, J., & Inglehart, R. (1995). Expert interpretations of party space and party locations in 42 societies.

Party Politics, 1, 73–111.Laakso, M., & Taagepera, R. (1979). Effective’ number of parties: A measure with application to West Eu-

rope. Comparative Political Studies, 12, 3–27.Laver, M. (2006). They fought the law; the law won. Formal bargaining models and portfolio payoffs in

Western Europe. Paper presented at Iowa University, 3 February.Laver, M., & Hunt, B. (1992). Policy and party competition. New York: Routledge.Laver, M., & Schofield, N. (1990). Multiparty government. The politics of coalition in Europe. Oxford: Ox-

ford University Press.Leiserson, M. (1968). Factions and coalitions in one-party Japan: An interpretation based on the theory of

games. American Political Science Review, 62, 70–87.Martin, L. W., & Stevenson, R. T. (2001). Government formation in parliamentary democracies. American

Journal of Political Science, 45, 33–50.Mattila, M., & Raunio, T. (2004). Does winning pay? Electoral success and government formation in 15 West

European countries. European Journal of Political Research, 43, 263–285.McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior. In P. Zarembka (Ed.), Fron-

tiers of economics. New York: Academic Press.McFadden, D. (1974). The measurement of urban travel demand. Journal of Public Economics, 3, 303–328.Morelli, M. (1999). Demand competition and policy compromise in legislative bargaining. American Political

Science Review, 93, 809–820.Müller, W. C. (2003). Tight coalitions and stable government. In W. C. Müller & K. Strøm (Eds.), Coalition

governments in Western Europe (pp. 86–125). Oxford: Oxford University Press.Müller, W. C., & Strøm, K. (Eds.). (2003). Coalition governments in Western Europe. Oxford: Oxford Uni-

versity Press.Riker, W. H. (1962). The theory of political coalitions. New Haven: Yale University Press.Signorino, C. (1999). Strategic interaction and the statistical analysis of international conflict. American Po-

litical Science Review, 93, 279–297.Strøm, K., Budge, I., & Laver, M. (1994). Constraints on cabinet formation in parliamentary democracies.

American Journal of Political Science, 38, 303–335.Van Deemen, A. M. A. (1989). Dominant players and minimum size coalitions. European Journal of Political

Research, 17, 313–332.Von Neumann, J., & Morgenstern, O. (1953). Theory of games and economic behaviour. Princeton: Princeton

University Press.Warwick, P. V. (1996). Coalition government membership in West European parliamentary democracies.

British Journal of Political Science, 26, 471–499.Warwick, P. V. (2001). Coalition policy in parliamentary democracies. Who gets how much and why. Com-

parative Political Studies, 34, 1212–1236.Warwick, P. V., & Druckman, J. N. (2001). Portfolio salience and the proportionality of payoffs in coalition

governments. British Journal of Political Science, 38, 627–649.Warwick, P. V., & Druckman, J. N. (2006). The portfolio allocation paradox: An investigation into the nature

of a very strong but puzzling relationship. European Journal of Political Research, 45, 635–665.

Making the first move

Documents