Government+Formation+in+Parliamentary+Democracies

Government Formation in Parliamentary Democracies* Lanny W. Martin Visiting Assistant Professor Department of Political Science University of Houston Houston, TX 77204 E-mail: lmartin1@houston.rr.com Phone: (713) 743-3893 Fax: (713) 743-3927 (Corresponding Author) Randolph T. Stevenson Assistant Professor Department of Political Science Rice University Houston, TX 77005 E-mail: stevenso@ruf.rice.edu Phone: (713) 527-8101 (x2104) Fax: (713) 285-5273 January 2000 Abstract. The literature on cabinet formation in parliamentary democracies is replete with theoretical explanations of why some cabinets form and others do not. This theoretical richness, however, has not led to the development of a healthy empirical literature designed to choose between competing theories. In this paper, we try to rectify this problem by developing an empirical model that can adequately capture the kind of choice situation that is inherent in cabinet selection and then using it to evaluate the leading theories of cabinet formation that have been advanced in the literature. For example, this analysis allows us to make conclusions about the relative importance in cabinet formation of traditional variables like size and ideology, as well as to evaluate the impact that recent new-institutionalist theories (such as Laver and Shepsle 1996) have on our ability to predict and explain cabinet formation over and above the more traditional explanations. * The authors wish to thank Bing Powell, Renée Smith, David Austen-Smith, Neal Beck, and seminar participants at the University of Rochester for their helpful comments on previous drafts, as well as Paul Warwick for sharing his data with us. All models were estimated using Stata 6.0. The data and program codes are available from the authors upon request. The question of which parties get into the government in parliamentary democracies has long been a favorite of scholars with a theoretical inclination. For those with a strong substantive interest in the workings of parliamentary government, this focus needs little justification. As Laver and Schofield (1990) put it: There can be no doubt at all that the government formation process…is one of the fundamental processes of European parliamentary democracy. Understanding how a given election result leads to a given government is, when all is said and done, simply one of the most important substantive projects in political science (89). For this reason, students of parliamentary democracy, game theorists, and contributors from many other specialties have made the theoretical literature on government formation one of the most active areas of research in the discipline, and, as a result, the literature is now replete with theoretical hypotheses about why some potential coalitions form while others do not. Unfortunately, it is not at all clear that this theoretical richness has resulted in systematic progress in the explanation and prediction of real-world governments. That is not to say that empirical evaluation has not been a part of this literature. Indeed, most (but not all) of the important theoretical studies of government formation include empirical work. The problem, however, is that this empirical work has produced no real tradition of controlled statistical analysis, but instead has relied on less powerful empirical designs—such as detailed accounts of coalition bargaining across a number of countries (Luebbert 1986; de Swaan 1973), analyses of particularly prominent cases (Strøm 1994), or uncontrolled comparisons of the characteristics of large samples of cabinets (Laver and Schofield 1990)— to evaluate ever more complicated, and often quite different, theories of government formation. Because we face such a large number of competing hypotheses in this literature, however, controlled statistical analysis is exactly the tool we need to choose among them. Uncontrolled comparisons and case studies are inadequate for this task and will simply not lead to the kind of empirical development that is necessary in this increasingly theoretically sophisticated field. This is, of course, an elementary point and we do not suggest that other scholars have been unaware of it.1 Instead, the hesitation to try multivariate statistical work has come from limitations (only recently overcome) in 1 The general difficulty in testing theoretical models of coalition formation has been explicitly recognized by Laver and Schofield (1990, 90) who say that “we should not expect too much from the confrontation between theories that deal with coalition formation and the formation of real-world government coalitions….” We would disagree with this assessment insofar as we think it derives more from an inadequacy in the methods used to test coalition theories than from any inherent non-testability of the theories themselves. the set of methodological tools available to most political scientists. Specifically, although theories of government formation have formulated the basic problem as the selection of a single coalition from the set of all possible coalitions, none of the commonly used statistical models is able to accommodate a dependent variable of this kind. A close reading of those studies that have attempted to force the problem into the standard frameworks shows that their results are largely dependent on a number of arbitrary data manipulations (Franklin and Mackie 1984; Browne 1970). In this paper, we attempt to close this empirical gap in the literature on government formation by analyzing data on over thirty thousand potential governments (drawn from 285 coalition formation opportunities in fourteen post-war democracies) in a multivariate statistical model that includes the operationalization of most of the major theoretical hypotheses that scholars have put forward to explain government formation in the last several decades. Unlike previous multivariate analyses of government formation that have relied on a linear regression approach, we employ a statistical specification specifically designed to deal with the multichotomous nature of the problem—that is, that a government is a single choice out of a finite set of potential governments. 2 These empirical techniques allow us to address in a systematic and rigorous manner some of the most important substantive issues arising from previous research on government formation. First, we identify a set of questions from early coalition research that center on several fundamental characteristics of potential governments, such as their size and ideological profile. For example, are majority governments more likely to form than minority ones? Are governments more likely to contain few or many political parties? Do the policy preferences of these parties have a discernable impact on their chances of entering government together? In addition, we focus on a number of questions posed more recently by coalition theorists—questions that, to our knowledge, have until now never been tested empirically, even in a standard regression framework—that emphasize the importance of political institutions. How do the formal and informal constraints surrounding coalition bargaining influence government formation? Do certain types of coalitions, or certain political actors, enjoy an institutional advantage in the formation process? Do more recent coalition theories that focus on the 2 Throughout this paper, we refer to a “potential coalition” or “potential government” to mean a combination of parties, or (in a convenient abuse of the term coalition) an individual party, that could form a government in some bargaining situation or “government formation opportunity.” 2 institutions structuring policy-making after the government has formed improve significantly on the explanatory and predictive power of traditional approaches? We begin to answer these questions in the next section, where we review most of the important studies of government formation that have appeared in the last several decades. We draw from them three sets of hypotheses reflecting different theoretical approaches to the subject. Following this review, we use the conditional logit model to test these hypotheses and to produce a set of predictions of government composition that dramatically improve upon previous attempts and definitively refute the claim that theoretical efforts in this area have failed empirically (von Beyme 1983) or that subjecting these theories to rigorous empirical testing is somehow unachievable (Laver and Schofield 1990). We discuss our conclusions from this analysis in the final section of the paper. Theories of Government Formation Any selection of hypotheses from a literature as theoretically developed as that of cabinet formation will surely overlook some worthwhile contributions. Further, our analysis must purposefully ignore some interesting theoretical approaches (e. g., Luebbert 1986) that are not amenable to quantification in a statistical analysis. That said, we have tried to include most of the important hypotheses in the theoretical literature and especially those fitting in to what we see as the two principal theoretical approaches to the subject—the size and ideology approach and the new institutionalism approach. The Role of Size and Ideology Theories constituting the size and ideology tradition assume only a minimum of institutional detail regarding the government formation process. Developed mostly in the 1960s and 1970s, these theories rely on the tools of cooperative game theory and spatial modeling to produce hypotheses about the effects of a potential coalition’s size and ideology on its chances of formation. The earliest theories in this tradition, which assumed that the primary goal of parties (or of the politicians controlling them) is to gain office, produced a series of hypotheses relating various aspects of a coalition’s size to its chances of forming. These theorists saw government formation as a zero-sum game in which cabinet portfolios are the payoffs. The most basic hypothesis that comes out of such a conception of cabinet formation is that only majority cabinets will form. Since holding office is the only consideration in coalition bargaining, a majority opposition would never tolerate the formation of a minority cabinet excluding them from office benefits. 3 The first major refinement of this logic is probably the best-known result in the coalition formation literature - the minimal winning hypothesis first proposed by von Neumann and Morgenstern (1953), later revised by Riker (1962), and applied explicitly to government formation by Gamson (1961). This theory suggests that if parties are seeking office to capture a fixed quantity of office benefits, they will form coalitions that win (that is, control a majority of seats in the legislature) without including any members unnecessary to the government’s majority. This “size principle” ensures that individual members of the coalition obtain the maximum possible office-related benefits. A related version of the “size principle” is due to Lieserson (1968), who proposed that government members would seek to minimize the number of parties in the coalition, since smaller groups of parties should presumably find it easier to reach agreement. Thus, two-party coalitions should be more likely to form than three-party coalitions, and so forth. Some more recent (and less well known) work that harkens back to this early concern with coalition size emphasizes the role that the largest party in the legislature can play as a focal point in coalition negotiations (van Deemen 1989; Peleg 1981). Arguing that the largest party is the “centripetal” actor in coalition bargaining and is therefore difficult to exclude from power, these authors hypothesize that coalitions that include the largest party will be more likely to form. The principal alternatives to the office-oriented (or size) theories are those that assume politicians are motivated by policy goals as well as (or instead of) the simple desire to get into office. The first of these theories retained a focus on office but tried to add policy as a secondary consideration. In particular, Axelrod (1970) suggested that office-motivated politicians are interested not only in maximizing their office benefits but also in minimizing the transaction costs of coalition bargaining that policy divisions impose. Axelrod predicted that politicians would try to form only minimal connected winning coalitions, that is, only those winning coalitions containing ideologically adjacent parties. De Swaan (1973) presented another variant of this idea, holding that parties will endeavor to form the minimal winning coalition with the smallest ideological range. These theories, then, generate hypotheses in which a mix of size and policy considerations comes into play. Still other work in this tradition has abandoned office motivations entirely and asked instead which coalitions will form if parties care nothing about office benefits but are concerned only with obtaining their preferred policies. The most influential answer thus far comes from the analysis of majority-rule spatial voting models, which have shown that when parties compete along one policy dimension (for example, a left-right 4 dimension), the party controlling the median legislator is essentially a policy dictator. With respect to government formation, then, the prediction is that the median party get into the government (Laver and Schofield 1990, 111).3 In addition to the median party hypothesis, theorists have developed other (less well known) policyoriented explanations of government formation over the last few years. One argument is an extension of Axelrod’s (1970) and de Swaan’s (1973) arguments discussed above. Both these authors make the case for policy considerations because they think that an ideologically compatible cabinet will face fewer transaction costs. There is nothing in this logic, however, that commits them to an office-oriented position. One could imagine, for example, that the ideological differences between the members of any potential coalition (whether minimal winning or not) would influence its chance of formation (Martin and Stevenson 1995 make this argument explicitly). Under the policy-oriented hypotheses, then, we include the notion that coalitions that are more ideologically compact are more likely to form, regardless of their size. Finally, in their review of the coalition politics literature, Laver and Schofield (1990) contend that the ideological divisions in the opposition may be as relevant for the viability of minority cabinets as the ideological divisions in the minority coalition itself. Indeed, students of minority government have often argued that the reason many minority governments are able to function is that they are able to exploit differences between opposition parties on an issue-by-issue basis (Strøm 1990). The hypothesis following from this line of argument is that a more ideologically divided majority opposition should increase the chances that a minority cabinet will form. From the size and ideology approach, then, we isolate nine hypotheses that we test in the empirical 3 The more general result is that the party controlling the multidimensional median (or the core) of a multidimensional policy space will be a policy dictator and will therefore be included in the government. As Schofield (1978) and McKelvey and Schofield (1986, 1987) have convincingly demonstrated, however, in multidimensional policy spaces in a majority-rule environment, a “structurally stable” core (i.e., a core that is robust to small perturbations in the preferred policies of the legislative parties involved) is unlikely to exist. If a core party does exist, then it is likely to form a singleparty government in which policy is located at the core party’s ideal point. In the more likely event that no core party exists, Schofield (1993) predicts that policies, and the unique coalitions associated with them, will be located in the “cycle set,” the limited area of the policy space for which any policy located inside it is majority-preferred to any policy located outside it. Unfortunately, we are not aware at present of any data available on the existence of core parties or cycle sets for any of the countries and time periods in our study. Clearly, this is a problem that future research on government formation should address, especially since the core party concept represents the main “multidimensional” alternative to the Laver and Shepsle (1996) Portfolio Allocation approach, which we discuss in more detail below. An important question that will remain unanswered even after our analysis, then, is whether the more complex institutional apparatus 5 analysis that follows: Hypothesis 1: Potential governments are more likely to form if they control a majority of seats in the legislature; Hypothesis 2: Potential governments are more likely to form if they are minimal winning coalitions; Hypothesis 3: Potential governments are more likely to form the fewer the number of parties they contain; Hypothesis 4: Potential governments are more likely to form if they contain the largest party in the legislature; Hypothesis 5: Potential governments are more likely to form if they are minimal connected winning coalitions; Hypothesis 6: Potential governments are more likely to form if they are ideologically compact minimal winning coalitions; Hypothesis 7: Potential governments are more likely to form if they contain the median party; Hypothesis 8: Potential governments are more likely to form the smaller their ideological divisions;4 Hypothesis 9: Potential minority governments are more likely to form the larger the ideological divisions within the potential opposition they will face. The Role of Institutions Due to the lack of obvious empirical success of the size and ideology theories, and because game theorists have gradually come to be more interested in institutions and less interested in cooperative game theory, an important alternative to the traditional approach to government formation began to emerge in the early 1980s. We refer to this approach as “new-institutionalist” because of its emphasis on the role of institutions in structuring the outcomes of the coalition formation process; however, different new-institutionalist scholars have tended to emphasize the effects of different sets of institutions. Specifically, while much of the new-institutionalist literature focuses on the impact of the rules and norms governing the process of government formation itself (Baron and Ferejohn 1989; Austen-Smith and Banks 1988), more recent work finds an important role for the rules that structure cabinet decision-making after the government is in place (Laver and Shepsle 1996). imposed on the government formation process by Laver and Shepsle fares any better in predictive terms than the more established concept of the core party in multidimensional settings. 4 This hypothesis is distinct from hypothesis 3 above, which refers to the most ideologically-compact minimal winning coalition, as opposed to the most compact coalition overall. 6 Institutions Structuring Pre-Formation Bargaining. Unlike the earlier size and ideology models, the work we discuss in this section attempts to specify explicitly the rules governing the process by which bargaining over governments occurs. In general, these theories tend to generate two types of hypotheses. The first type looks very much like the hypotheses that have emerged from the size and ideology literature and depend, as the earlier ones did, on the specific assumption a researcher makes about the motivations of politicians. Consequently, those new-institutionalist theories that assume that politicians pursue office benefits tend to produce minimal winning hypotheses just like those of Riker and Gamson. Likewise, those theories that focus on policy motivations invariably find a special role for parties that are median or otherwise “central” in the policy space. These predictions about size and ideology, then, do little to distinguish new-institutionalist theories from their predecessors. What is different, however, is that these theories, with their rich institutional detail, generate a number of additional hypotheses that predict cabinet composition from variables other than the distribution of party ideology and legislative seat shares. The first several hypotheses we discuss reflect two of the clearest messages that have emerged from the formal literature on legislative bargaining—that procedural powers are valuable in bargaining and that the reversion point that obtains in the case of bargaining failure can affect bargaining outcomes. With respect to government formation, the meaning of procedural powers and reversion points is embodied in the rules and norms of the coalition bargaining process. Further, since many countries use similar procedures (and abide by similar norms) when forming cabinets, institutionalists believe that it is possible to make empirical generalizations across systems about the effects that particular kinds of institutions have on formation outcomes. Indeed, much of the power in institutionalist theory comes from the ability to predict differences in outcomes across systems based on institutional differences. There are at least two important procedural powers identified in the coalition politics literature that help to structure the process by which cabinets get formed in most parliamentary democracies. The first of these is the power to propose the coalition alternatives over which bargaining will take place. The second is the power to control when bargaining over a new cabinet will occur. In the real world of coalition formation, politicians that have the power to propose coalition alternatives 7 are known as formateur parties.5 Potential partners must accept or reject the proposals brought forward by these institutionally designated actors before bargaining over other proposals can proceed. All of the existing noncooperative game theoretic models of coalition formation have utilized the notion of the formateur party to begin the process of government formation in their models (e.g., Austen-Smith and Banks, 1988; Baron 1991, 1993) and so provide predictions about the way that this formateur (usually identified only by party) can shape formation outcomes. Almost universally, these theories, along with less formal theories of coalition bargaining that rely on a formateur party to get coalition negotiations rolling, conclude that the formateur party should be able to shape the coalition in important ways, including guaranteeing its own place in the government that forms as well as biasing the ideological profile of the cabinet in its favor. Baron’s (1991) spatial model of coalition formation is typical in this respect. In his model, a cabinet is composed of both a set of parties and a negotiated policy position. Coalition bargaining proceeds sequentially with a proposal by a formateur followed, eventually, by a majority vote on the proposal. A clear prediction of his model is that the party that is able to make the first proposal for the formation of a government (the formateur party) will be able to offer (and have accepted) a policy proposal that is closer to its ideal point than to that of its partner. This result mirrors other theoretical efforts that suggest that formateur parties should be able to form coalitions that are biased toward their preferred outcomes.6 In the empirical work that follows, we examine one aspect of this bias by asking if formateur parties are systematically able to guarantee their place in the cabinet and whether they are able to propose (and have accepted) coalition alternatives that consist of parties that they find to be most ideologically compatible. A second procedural power that may help determine the outcome of coalition negotiations is the power 5 Descriptions of such parties can be found in Laver and Schofield (1990) and in the many qualitative country-specific studies of coalition formation. Sometimes the process is complicated by the use of an informateur (especially in the low countries), who is appointed by the head of state to investigate coalition formation possibilities but who will not take a position in the cabinet that forms. In cases in which this informateur simply recommends a particular formateur to the head of state, these systems are effectively the same as those that appoint a formateur directly. On the other hand, when the informateur essentially selects both the prime minister and the coalition partners at the same time, the informateur, rather than the formateur, is the individual with proposal power. In our data, the latter cases are extremely rare, occurring occasionally in The Netherlands and Belgium. Strengthening this theoretical result is the empirical observation that unsuccessful proposals for governments are normally not followed by a change in the proposing party (Stevenson 1997). This finding suggests that even in the face of uncertainty about whether certain proposals are acceptable, a formateur may try different options, be defeated, and yet go on to get a different proposal accepted. 6 8 that incumbent prime ministers and (to a lesser extent) incumbent coalition partners have to choose the timing of the negotiations. In most parliamentary democracies incumbent governments have considerable control over the timing of government termination, and although formal work on this topic is just beginning (Lupia and Strøm 1995; Baron 1998; Diermeier and Stevenson 1999), many scholars have suggested informally that incumbent governments may be able to time subsequent government negotiations to their advantage (Laver and Schofield 1990, 213). The party that is particularly advantaged by these rules is the party of the prime minister, who can often control the government agenda (bringing up or blocking “coalition breaking” issues), reshuffle the government to appease unhappy members (preventing a cabinet termination at an undesirable time), or even dissolve the legislature and call early elections (Huber 1996). The second well-accepted generalization to come out of the bargaining literature is that the reversion outcome influences the behavior of the actors involved in the bargaining, generally benefiting those actors whose most preferred outcome is the reversion outcome. This result is a part of almost all bargaining models, and coalition theorists have used it to suggest that whichever coalition one expects to come to power (or remain in place) in the event of a breakdown (or protraction) of coalition bargaining is the very coalition that should be able to bias the outcome of the negotiation in its favor. For example, Strøm et al. (1994) argue that an incumbent cabinet (usually acting in a caretaker capacity) that stays in office while bargaining continues is the reversion outcome in the event of protracted bargaining or in the event that bargaining breaks down. Consequently, they argue that the results of formal bargaining theory indicate that these incumbents will enjoy an advantage in the coalition negotiations. Based on this hypothesis, then, our empirical work examines whether incumbent cabinets are likely to form again immediately after they collapse. We should note, however, that the logic connecting this particular lesson from the bargaining literature (i.e., that reversion outcomes matter) to the real world of coalition negotiations seems a bit strained—specifically in the claim that incumbent cabinets provide the relevant reversion outcome. First, this claim ignores the fact that many cabinets end exactly because one of the governing parties is unhappy with the current coalition; thus, the reversion outcome for such a party is not a desirable outcome and so provides no advantage in coalition bargaining. Second, it may not be empirically accurate to think of the caretaker cabinet (the incumbent during negotiations) as the relevant reversion outcome in coalition bargaining. While this 9 coalition provides the status quo during negotiations, its caretaker status often limits its ability to act as a true governing body (that is, new proposals may not be brought forward by caretaker cabinets in some countries). Further, in the case that bargaining breaks down completely, most countries will require a new election—not a continuation of the incumbent government. In this case, then, party expectations about the outcome of a new election are the relevant reversion points in bargaining.7 Some of the other hypotheses in the new-institutionalist literature are not products of the formal bargaining literature as much as they are the result of years of work by country experts and other students of coalition formation who have recognized the role that the particular institutions and norms that each nation uses to structure its own process of coalition bargaining have in determining the outcomes produced in those systems. One such hypothesis that has been of particular interest to students of Scandinavian politics concerns the effect of investiture requirements on the possibility that minority governments will form (Strøm 1990). An investiture rule requires that a prospective government, once its members have agreed to form, must pass a formal majority vote in the legislature. The argument is that a minority government is more likely to form in the absence of an investiture rule. Presumably, since its entire legislative program need not face a single vote, a minority government not facing an investiture requirement is able to manipulate its legislative agenda to build ad hoc majorities on each separate issue it presents to the legislature. Three other hypotheses in this vein are qualitatively different than the ones reviewed above in that they are concerned primarily with behavioral norms governing cabinet formation rather than with more formal rules. As a result, some comparative researchers question their status as “institutional” explanations for coalition bargaining outcomes.8 Given the inclusive nature of the present empirical project, however, we have decided not 7 In contrast to the argument made by Strøm et al., recent theoretical work by Baron and Diermeier (1999) suggests that the incumbent cabinet is actually disadvantaged in coalition negotiations. Specifically, their full equilibrium model of proportional rule elections, government formation, and policy choice in parliamentary democracies implies that incumbent cabinets will never form again immediately after they collapse. That is, if one of the parties in the current government becomes the formateur in the next government, then it will always switch cabinet partners. A finding that status quo cabinets do tend to reproduce themselves would thus constitute evidence against this proposition. Laver and Schofield (1990) disagree with Strøm et al. (1994) about the inclusion of these hypotheses in the “newinstitutionalist” agenda. Their argument is that the exclusion of anti-system parties from government coalitions or the formation of pre-electoral agreements are behaviors to be explained by a theory of government formation, not an institution or norm to be labeled a determinant of government formation. Strøm et al. are less concerned with this distinction, and in their own analysis they identify these kinds of behaviors as either “hard” or “soft” constraints on coalition bargaining and exclude potential coalitions from the analysis if they are so constrained. We advocate neither of 8 10 to exclude them on these grounds. The first two of these hypotheses concern pre-electoral commitments by parties to form certain governments given a particular electoral result. In general, these commitments take two forms. The first is essentially a positive statement that commits a party to try to form a government with another party. Usually, these pacts do not imply that other parties not included the pact must be excluded as partners. Also, they are often contingent on a particular electoral outcome, which suggests that the degree to which they actually dictate coalition outcomes in practice is more appropriately a matter for empirical analysis rather than assumption. Another type of pact is what we refer to as an “anti-pact.” These kinds of pacts are declarations by parties that they will not coalesce in certain ways. Most often, this takes the form of ruling out any coalition that includes a particular party—for example, when the CDU/CSU in Germany after the 1949 and 1953 elections refused to ally with the SPD under any circumstances (Klingemann and Volkens 1992). In other cases, such as with Fianna Fail in Ireland, which for most of the post-war period declared that it would only rule alone (Laver 1992), a party might declare an anti-pact with every other party in the system. More unusual anti-pacts rule out a specific coalition, such as when the Democrats ‘66 in The Netherlands declared that it would not ally itself with both the CDA and the VVD, but that it would ally itself with one or the other of them (Tops and Dittrich 1992). In all of these cases, public commitments to rule or not to rule with some other parties probably constitute a powerful constraint on coalition bargaining, and we therefore include them in our empirical analysis. Another hypotheses is a similar to the “anti-pact” notion in that it sees some coalitions as unlikely to form because they contain parties likely to be excluded by others from a role in government. The difference, however, is that this hypothesis suggests explicitly the reason for the exclusion. Specifically, it suggests that parties that adopt an “anti-system” ideology may face systematic exclusion from the cabinet by other parties (so that coalitions that contain these “anti-system” parties will be less likely to form). Numerous coalition theorists examining specific cases have reaffirmed this logic—Budge and Keman (1990), for example, argue that because strong social norms exist against admitting parties to government who are not committed to the maintenance of the democratic system, the electoral costs of forming coalitions with anti-system parties are prohibitive. Further, these positions, but we do include these variables in our empirical models and estimate their impact on coalition formation. 11 external pressure from other countries—for example, pressure from the United States to keep Communist parties out of Italian governments during the Cold War (Mastropaolo and Slater 1992)—may make inclusion of antidemocratic parties in coalition governments especially costly. In the empirical analysis, then, we use a continuous measure of the level of anti-system rhetoric in a party’s manifesto to build a score for the level of “anti-system” views expressed by the members of each potential coalition. Those coalitions with higher scores should be less likely to form. It is important to note that these “anti-pact” and “anti-system” variables are distinct. While anti-pact restrictions for several coalitions are the result of commitments by pro-system parties to exclude anti-system parties (indeed, most of the parties traditionally thought of as anti-system parties are coded as having anti-pacts with the other parties), there are also many coalitions composed entirely of pro-system parties that are subject to anti-pact restrictions. In these cases, one party has simply made it clear that it will not form a government with another party following the election (which may be for any number of reasons not having anything to do with antisystem views). In contrast, the anti-system variable pinpoints only those parties that have expressed anti-system views in their manifestos. Clearly, this captures all the usual anti-system parties (so both this variable and the anti-pact variable will tell us whether these parties are indeed excluded), but it also gives positive scores to parties that, while normally pro-system, adopt anti-system rhetoric for a particular election. In summary, we consider seven hypotheses from this literature: Hypothesis 10: Potential governments including the formateur are more likely to form 12 Hypothesis 17: Potential governments including anti-system parties are less likely to form. Institutions Structuring Post-Formation Government Decision-Making. Laver and Shepsle (1996) provide one of the most recent and comprehensive of the new-institutionalist approaches. Their Portfolio Allocation approach, like the other new-institutionalist models reviewed above, is concerned with modeling the process of government formation that produces a government. It differs from other new institutional theories, though, in a number of important ways. Specifically, its reliance on the spatial voting model commits it to a predominantly policy-oriented view of political motivations, despite the fact that other theorists seem to be moving toward a more balanced policy/office orientation (Strøm 1990; Baron 1998). Second, the principal institutions affecting government formation in this approach are those pertaining to decision-making within the government rather than those structuring the process of government formation itself (which are almost the exclusive focus of other new-institutionalist work).9 Laver and Shepsle’s theory begins from a multi-dimensional spatial voting model somewhat akin to the policy-oriented models from the work on size and ideology discussed above. Unlike these models, however, Laver and Shepsle assume a particular set of institutional rules relating to government decision-making—namely, that ministers have dictatorial control over the dimension of policy associated with their ministry—and from this are able to generate a number of new hypotheses about the ability of certain kinds of parties to obtain a place in the government. These hypotheses center on the existence of parties that enjoy a strategic advantage in coalition formation because of their relative ideological positions and relative size.10 These lucky parties, which come in two varieties, “very strong parties” (VSPs) and “merely strong parties” (MSPs), should be able to guarantee their place in the government or even rule alone in a minority government.11 Laver and Shepsle define a VSP as a party that, in a weighted majority-rule voting game in which preferences over governments are defined by the expected 9 One exception is Baron (1998), who focuses on how different confidence vote procedures can impact formation. In many ways, Laver and Shepsle’s model is reminiscent of traditional models, especially in their emphasis on size and ideology as key independent variables determining outcomes. The difference in the Laver and Shepsle approach, however, is that their policy-oriented parties are not free to adopt any policy in the multidimensional policy space (which would lead to cycling in an unconstrained majority-rule environment) but can only select policies implied by the lattice defined by the policy ideal points of the parties on each policy dimension. The relevance of this lattice, which induces a non-cyclic equilibrium outcome, stems directly from the institutional assumption giving parties dictatorial power on the dimensions of policy over which they gain control in government negotiations. 10 13 policy produced once the government forms, some majority coalition prefers to give all the government portfolios rather than to support any other alternative government. This party should thus be able to form a single-party minority government that has majority support in the legislature or, relaxing the formal argument, at least guarantee itself a position in the government that forms. A MSP is not quite as well placed as a VSP, but it still should be able to get into the government. In particular, while some legislative majority prefers at least one coalition government to the government in which a MSP gets all the portfolios, the MSP is a member of each of these majorities. Theoretically, then, a MSP is in a position to veto the formation of all governments with any chance of forming and may therefore be able to hold out for its most preferred outcome (a government in which it rules alone). In practice, however, a standoff might occur that can prevent a MSP from successfully forming a single-party minority government. Still, no government should form without the participation of the MSP. This argument, then, implies four additional testable hypotheses: Hypothesis 18: Governments are more likely to form if they contain a VSP; Hypothesis 19: Single-party governments are more likely to form if they contain a VSP; Hypothesis 20: Governments are more likely to form is they contain a MSP; Hypothesis 21: Single-party governments are more likely to form if they contain a MSP. Laver and Shepsle’s formal analysis of the technical conditions for the existence of strong parties shows that “the strong party concept is not simply restating something obvious” (92). That is, mathematically these are not just median parties, multidimensional median parties, core parties, or some other previously distinguished kind of party. Rather, the conditions that define a strong party are much more complex and do not always correspond to a party’s centrality or size. That said, however, the fact that small non-central parties can sometimes be strong parties does not tell us that they actually are strong parties in the kinds of party systems that we observe in practice. To examine this, Laver and Shepsle provide evidence, using simulations, that in fact a pronounced tendency exists for strong parties to be more central than other parties—indeed, it is virtually a requirement that they hold the median position on at least one dimension of policy and that they hold the most legislative seats. These simulation results suggest, then, that despite the mathematical fact that the strong parties are conceptually 11 Throughout the discussion, we assume that no party controls a majority of seats in the legislature. 14 distinct from earlier notions of median parties or dominant parties, in practice the concept may not do any better than these more simple formulations at predicting the composition of real-world governments. A controlled statistical analysis that directly confronts Laver and Shepsle’s variables against the hypotheses reviewed above (that median parties and dominant parties will get into the government) is therefore essential to the empirical assessment of their theory. Later in this paper, we make such an assessment and find that several of Laver and Shepsle’s hypotheses fare quite well. Data and Methods The twenty-one hypotheses reviewed above represent a good sampling of both the received wisdom on the systematic determinants of government formation as well as the most recent innovations in this field. Studies of government formation, however, have yet to evaluate most of them in the context of a multivariate model.12 Indeed, the empirical methodology in this literature has not advanced far beyond simple uncontrolled comparisons of means. Franklin and Mackie (1984) were the first to perform a multivariate test of some of the size and ideology hypotheses. They lacked the statistical techniques, however, with which they could properly model the choice situation—a problem they explicitly recognized and incorporated into their conclusions, stating that their findings, as well as findings of earlier studies using bivariate regression, rested crucially on the specific samples chosen and the specific data manipulations used to fit this qualitative choice problem into a regression framework. The central problem with their approach is that in a regression framework (or even a dichotomous choice framework) each potential coalition in a formation opportunity enters the estimation as a separate case. Thus, if one includes countries such as Italy or Denmark, with a large number of parties at any given time, then thousands of cases enter the estimation and completely swamp out relationships in other countries. All credible attempts to use regression analysis to analyze this question have recognized this problem and have tried to circumvent it by creating elaborate schemes for weighting cases in the regression (Browne 1970; Franklin and Mackie 1984). Unfortunately, as Franklin and Mackie’s (1984) work demonstrates, the choice of a particular weighting scheme largely determines the results. Therefore, to evaluate the various theoretical hypotheses reviewed in the sections 12 Warwick (1996) has recently done a multivariate analysis in which he examines the chances that an individual party will get into the government. However, since he does not focus on characteristics of potential governments as a whole, but on those of individual parties, he cannot directly examine many of the prominent theoretical propositions in the literature that 15 above, we abandon the regression framework altogether and turn instead to a maximum likelihood model that more directly captures the structure of the choice problem. In particular, we model the formation of a government as an unordered discrete choice problem where each formation opportunity (not each potential coalition) represents one case and where the set of discrete alternatives is the set of all potential combinations of parties that might form a government. Specifically, the number of potential coalitions in any formation opportunity is equal to 2p-1, where p is the number of legislative parties. Thus, in a bargaining situation in which five parties have representation in the legislature, any one of thirty-one potential coalitions can form a government, ranging from a single-party government comprising any one of these parties to a grand coalition comprising all of them. For example, in the case of the formation opportunity that occurred in Germany following the election of 1972, three parties were represented in the Bundestag (none of which had a majority of seats)—the Christian Democrats (CDU/CSU), the Social Democrats (SPD), and the Free Democrats (FDP). Consequently, any of seven potential cabinets could form (a CDU/CSU single-party cabinet, a SPD single-party cabinet, a FDP single-party cabinet, a CDU/CSU-SPD two-party coalition, a CDU/CSU-FDP two-party coalition, a SPD-FDP two-party coalition, or a coalition of all three parties). Maximum likelihood estimation has become the standard approach to modeling these kinds of data, such as in much of the literature on consumer and occupational choice in economics (Greene 1993) and increasingly in political science (Whitten and Palmer1996). The reason for this is that the likelihood approach lets one choose a distribution for the dependent variable that is appropriate to the true form of that variable. For this problem the dependent variable is not a whether a specific potential coalition forms (the dependent variable analyzed in previous studies) but is instead a vector that gives a “0” to each potential coalition in a formation opportunity that does not form and a “1” to the coalition that does form. A univariate normal distribution (the assumption used if one models this as a regression) or even a univariate Bernoulli distribution (the assumption used in logit or probit) are inappropriate here because they assume that each potential coalition is independent of the others (this is clearly not true since one and only one of these coalitions can in fact be realized). Instead, one should rely on a multivariate distribution over the vector-valued dependent variable. The usual choice for discrete data such as are meaningless at the individual party level (e. g., minimal winning status). Still, some of his results are relevant, as we later show, especially with respect to the selection of the formateur and the ideological divisions within coalitions. 16 ours is the multinomial distribution. The specific specification of the discrete choice model we adopt is McFadden’s conditional logit model (McFadden 1973, 1974). Unordered discrete choice models posit that a particular form of the multinomial distribution characterizes the probability that an individual chooses a particular alternative from some set of all possible alternatives. The general form of the multinomial distribution is P(n1, n 2 , ... , n 3) = N! ∏n i=1 m ∏ π ni , i i =1 m where the m random variables ni indicate the number of occurrences in N trials of choice i and where πι is the probability that choice i is made in any one trial. The kernel of the log-likelihood function for data distributed in this way is simply L = ∑ n log π i i =1 mi ι , where the subscript on m indicates that the set of alternatives at each choice situation (formation opportunity) need not be identical. McFadden (1974) proposed a particular parameterization of πι that he called the conditional logit (because the model is specified conditional on mi) and that Amemiya calls the multinomial logit (we adopt McFadden's label). In this model the probability that individual i chooses alternative j is Pij = ∑ exp (x′ijβ ) mi exp (x′ijβ ) k=0 . Our application is somewhat unusual (but similar to McFadden’s (1974) analysis of highway route selection in California) in that the alternatives available in one choice situation (highway construction projects for McFadden and formation opportunities here) may not be substantively related to those available in any other choice situation in the data set. Thus, the usual conception of the discrete choice problem in which j signifies a certain common alternative for all individual decision-makers (for example, individuals voting for one of three candidates based on a vector of candidate characteristics) does not apply. As Amemiya has pointed out, however, since each alternative in a particular choice set is completely characterized by its vector of characteristics, labels 17 assigned to choices are arbitrary, and thus the distinction between these cases is simply semantic.13 The conditional logit model offers several advantages over the regression framework. First, since the unit of analysis in the conditional logit is the formation opportunity and not the potential coalition (as in the regression approach), adding a country with a very large number of potential coalitions per formation opportunity is not problematic. Consequently, schemes for weighting cases are not required. Further, the signs of the marginal effects from the conditional logit specification are consistent with the signs on the estimated coefficients. This is not a characteristic of unordered choice models in general, but follows from the specific properties of the conditional logit. This consistency between the estimated coefficients on covariates and the signs of the marginal effects is particularly useful in our case because of the difficulty in obtaining and interpreting marginal effects in this model. Specifically, the large number of alternative cabinets, as well as the fact that these alternatives are not the same in every formation opportunity, complicates the interpretation of marginal effects in this model. The reason is that the coefficients produce a marginal effect for each of the independent variables on the probability that each individual coalition forms. This means that to assess the marginal effects in general, we would have to look at the marginal effects for each potential coalition in the entire data set. While we have developed techniques to do this and have done it for these data, a lack of space prevents us from presenting these results. More importantly, after having examined these marginal effects, we find that the most important messages of this study can be communicated using only the estimated coefficients, which, fortunately, are sufficient to test the signs and significance of the hypothesized relationships.14 One potential drawback of the conditional logit model is that it assumes the independence of irrelevant alternatives (IIA), which implies that the odds of choosing alternative j over j0 do not depend on the other alternatives in the choice set or on the values of the covariates associated with those alternatives. This means that if an alternative is eliminated (or if a new alternative is made available) the odds ratios for the remaining (or original) alternatives would not change. While this implication may make theoretical sense if the eliminated (or 13 McFadden (1974) notes that one implication of the arbitrary labeling is that, within a given choice set, if two alternatives have identical vectors of covariates, they will have the same probability of being chosen. 14 Of course, this makes it difficult to speak of the relative size of the effects of different independent variables in the model, since the size of the marginal effects need not preserve the ordering of the size of the coefficients. 18 new) alternative is not substantively related to the other alternatives, it makes less sense if the reverse is the case (we might think, for example, that the odds of installing a socialist cabinet versus a Christian democratic cabinet should change if a new socialist party emerges). Whether this assumption causes problems for inference in any specific instance is essentially an empirical question—as Quinn, Martin, and Whitford (1999) demonstrate in several examples—and so statistical tests have been developed for determining whether the IIA assumption is a problem in any given application of conditional logit. These tests rely on the logic laid out above—specifically, if IIA holds, one ought to be able to drop an “irrelevant” alternative from the estimation and still consistently (but not efficiently) estimate the parameters of interest. Consequently, one can apply a standard specification test proposed by Hausman and McFadden (1984) to determine (essentially) whether the estimated coefficients differ between a model including all alternatives and a model in which some alternatives are dropped (Greene 1999, 865).15 If the coefficients do differ, then the estimates are sensitive to the IIA assumption, suggesting that we should carefully assess its theoretical plausibility. In our application of this test, difficulties arise because we have choice sets that are different for different formation opportunities. Consequently, we cannot simply choose one alternative to drop and then estimate the model without the alternative (since there is no specific coalition alternative that is common to all formation opportunities). Instead, we follow the suggestion of McFadden (1974) and drop a random set of alternatives from each formation opportunity (never dropping the government that actually formed) and then apply the Hausman test.16 We begin by dropping ten percent of the alternatives from each formation opportunity, and then, to be sure that the randomization procedure is not producing unusual answers, we repeat the sampling process and recalculate the Hausman test statistic twenty times. In each of the tables in the results section, we report the average p-value (over the twenty replications) for rejecting the null hypothesis that the IIA assumption holds. In every case, we are far from being able to reject the null hypothesis and thus conclude that the IIA assumption is not problematic in our application. The countries and years covered in our data set are Austria (1949-82), Belgium (1946-85), Canada (1945- 15 This test requires two estimates of the parameter vector: one that is consistent but inefficient under the null and one that is both consistent and efficient. The estimates before and after dropping alternatives fit these criteria. 19 84), Denmark (1945-84), West Germany (1949-87), Iceland (1946-87), Ireland (1948-86), Israel (1949-84), Italy (1949-87), Luxembourg (1945-84), The Netherlands (1948-86), Norway (1945-86), Sweden (1945-86), and the United Kingdom (1945-87).17 The data set consists of all the governments that formed as well as all the potential governments that could have formed in these democracies during the periods indicated. All together, the data consist of information on a total of 33,256 potential governments for 220 formation opportunities, each with one potential government that formed. As we discussed above, the conditional logit model treats each formation opportunity as a separate and independent case and each potential coalition as a single choice within a given formation opportunity. Each potential coalition in any formation opportunity also has attached to it a set of institutional, ideological, and sizerelated characteristics that will serve as independent variables in our analysis. For example, in the German example given above, the percentage of seats in the Bundestag received by each party were 46.4, 45.4, and 8.3 for the SPD, CDU/CSU, and FDP respectively. In our data, then, we coded the variable for minority status as “1” for each of the three possible single party coalitions and “0” for each of the other four alternatives. Other variables coded at the level of potential coalitions were constructed from data on parties in a similar way. All of the variables that are constructed from legislative seat shares (minimal winning coalition, minimal connected winning coalition, ideologically-compact minimal winning coalition, median party in coalition, minority coalition, and largest party in coalition) are dichotomous and are based on the seat results reported by Mackie and Rose (1991), the corrections to these results listed in Appendix B of Lijphart (1994), and for later elections, the results reported in special issues of the European Journal of Political Research. All independent variables constructed from left-right ideology scores are based on the party ideological positions provided in the manifestos project of the European Consortium for Political Research, using the formula for determining party positions on the left-right scale from Laver and Budge (1992, 26-7). These include the dichotomous variables median party in coalition, minimal connected winning coalition, and ideologically compact 16 McFadden’s study was quite similar to this one in that he was modeling the “choice” of a complex organization over a large set of alternatives that was changing across individual observations. 17 Again, we exclude all formation opportunities in which a single party won a majority of legislative seats. Although in most countries such occurrences were rare, this qualification required us to exclude all bargaining situations in the United Kingdom, except for the one following the 1974 parliamentary election. 20 minimal winning coalition and the continuous variables government ideological divisions and potential opposition ideological divisions. We define these latter variables as simply the absolute distance between the parties at the left-right ideological extremes of the (government or opposition) coalition in question.18 For example, for the 1972 German formation opportunity, we assigned the SPD a score of -13.8 on the left-right scale according to the manifestos project, the CDU/CSU a score of -7.4, and the FDP a score of -9.8. This places the FDP a little to the right of the CDU/CSU, but relatively close. Consequently, (before rescaling for estimation) we would code the CDU/CSU-FDP coalition as having an ideological division of 2.4, the SPD-CDU/CSU coalition a score of 21.2, the SPD-FDP coalition and the coalition of all parties as having a score of 23.6, and all single-party coalitions as 0. The variable anti-system presence is a continuous measure corresponding to the anti-system score of the party having the greatest anti-system score within the potential coalition. This score is from the manifestos data project’s category anti-establishment views, as defined by Laver and Budge (1992, 24). The variables preelectoral pact and anti-pact are dichotomous indicator variables based upon the individual country surveys found in Laver and Budge (1992) and other historical descriptions of individual formation opportunities. For each formation opportunity we searched these historical materials for instances in which parties explicitly committed themselves to participate or not to participate in certain coalitions.19 For example, Strøm (1992, 71), in his discussion of constraints on coalition bargaining in Norway in the post-war period, cites the continual refusal of the Labour party to consider entering government with any other party. Based upon this observation, we have thus assigned any multiparty potential coalition in Norway containing the Labour party an anti-pact score of “1.” Similarly, in the case in Norway in 1969 in which the governing parties promised to renew their coalition if they retained their legislative majority following the election, we assigned a pre-electoral pact score of “1” to this coalition (as it did, in fact, narrowly retain its majority in Stortinget). Finally, the Portfolio Allocation variables (indicating whether parties are “very strong” or “merely strong”) are from Laver and Shepsle (1996) and have 18 To facilitate the estimation procedure, we re-scaled these continuous ideology variables, as well as the anti-system variable, to lie on a continuous [0, 1] interval by simply dividing them by their respective maximum sample values. This will make no difference to the interpretation of the results. Parties have occasionally made a commitment not to form a coalition with a party in all bargaining situations for the foreseeable future. In these cases, we did not require an explicit commitment in each election in to code an anti-pact. 19 21 been provided to us by Paul Warwick. Our versions of these variables code potential governments according to whether they contain such parties and whether they contain only such parties. Before presenting the results, one additional modeling issue remains. In hypotheses 10 and 11 above, we refer to the identity of the formateur. Specifically, we hypothesize that the formateur party, due to various procedural powers granted to it in the formation process, will be able to both guarantee its place in the cabinet that eventually forms and shape the ideological makeup of the cabinet to its liking. The problem with testing these hypotheses directly is that we do not have adequate data on the identity of the formateur. This is not a big problem in testing hypothesis 11 (about the formateur’s ability to shape the ideological makeup of the coalition) since we do know one thing about the identity of the formateur—that the last (by definition successful) formateur will be the prime minister in the new cabinet. We can test this hypothesis, then, by constructing a measure of the ideological “spread” of the coalition around the party that we know will be the new prime minister and estimating the effect of this variable in a sample in which we drop all potential coalitions that do not contain this party (called the “restricted sample” below and in the tables). This will allow us to test whether potential coalitions that are more ideologically favorable to the prime minister’s party are more likely to form, given that the identity of the new prime minister is known. Hypothesis 10 suggests that formateurs will be able to guarantee themselves a place in cabinet. Obviously, we cannot begin, as above, by taking advantage of the fact that the prime minister was the last formateur (since they all get into the cabinet by definition). In order to get at this hypothesis, then, we use proxies for the identity of the formateur in place of the variable itself. If these proxies prove to be important predictors of cabinet composition then we may be able to conclude that formateur-ship does have the effects predicted by the theory. 20 In identifying good proxies, there is both good news and bad news. The good news is that some very good proxies exist. Both Stevenson (1997) and Warwick (1996) find three variables that together predict the 20 Those familiar with the statistical problem of non-random assignment will have pause at this substitution. In general, as 22 identity of the formateur with near certainty. The bad news, however, is that these are the previous prime minister, the largest party in the legislature, and the median party. This is bad news because each of these variables is implicated in hypotheses other than those about the formateur. Thus, simply including them in an empirical model and finding that they are important will not permit us to distinguish between the formateur story and the other ways that these variables may be important in formation. One way that we may be able to get some leverage on this question is by finding a way to distinguish whether the effects of the various proxy variables are felt only in choosing the formateur or whether these effects are felt throughout the formation process (i.e., after a formateur has been identified). For example, if the median party variable were a good predictor of the identity of the formateur but not a good predictor of cabinet membership once the identity of the formateur was given, then we might prefer the institutional (formateur) hypothesis over the institution-free hypothesis that implies that median parties will always get into cabinets because of their centrality in the policy space. This would be a reasonable inference because the institution-free hypothesis regarding the median party does not imply that this effect should be restricted to parties that become the formateur—rather it should apply to all the parties equally. Conversely, if we find that the median party variable predicts both the identity of the formateur and the identity of the rest of the cabinet given the formateur, then we must conclude that at least part of the median’s effect is due to reasons other than simply its use in identifying formateurs. We could evaluate each of the other proxies in a similar manner. All that remains, then, is to construct a method for isolating whether the effects of our variables have a primary impact in formateur selection, in partner selection given a formateur, or in both stages. We do this by once again estimating the models not only on the full sample of cases but also on the restricted sample defined above. If our proxies are not important in the sample restricted to potential coalitions that contain the last formateur, but are important in the full sample, we know that their effect comes principally from the selection of the formateur.21 21 Clearly, in testing this hypothesis we must worry whether there are multiple formateurs in a single formation opportunity (i.e., one formateur fails to form a cabinet and resigns in favor of another). Although there are many cases of multiple formateurs in parliamentary democracies, Stevenson (1997) has shown that these are usually cases in which an individual fails, not a party. Consequently, the formateur-ship, while it may pass between leaders of the same party, does not usually fall to a new party. Indeed, cases of the parties of formateurs failing to secure their positions in the eventual cabinet are quite rare in parliamentary democracies. 23 Results In this section, we provide estimates of the coefficients from the conditional logit model as well as appropriate standard errors. Since these models are non-linear, however, these numbers provide information only about the direction and statistical significance of the relationships and not about the substantive magnitude of the effects. The conditional logit model is unusual among multichotomous choice models in that the signs of coefficients do accurately reflect the direction of the corresponding substantive effects (Greene 1993, 669). For example, a positive coefficient associated with a coalition characteristic means that if a coalition has “more” of that characteristic, then its odds of forming will go up and the odds of every other coalition forming will go down. The Effects of Size and Ideology As we pointed out previously, the goal of the size and ideology theories was to create widely applicable generalizations about the kinds of government coalitions that are likely to form in any parliamentary system. In Table 1, we begin to evaluate these theories using six different model specifications. These specifications begin with the pure office-oriented, or size, hypotheses (Model 1), add the hybrid office/policy hypotheses (Model 2), and then sequentially include the pure policy hypotheses (Models 3-5). We also include a final specification that we use as the basis for comparison with the institutional models reported in the next section (Model 6). The first important message to draw from Table 1 is that, with the exception of the minority cabinet variable (discussed below), the various specifications all produce similar estimates and tell the same substantive story—that both policy and office benefits play a significant role in the formation of governments. Besides this general conclusion, however, there are a number of more subtle messages in the results. First, since all coalitions must be either minimal winning, minority, or surplus majority, the results for Model 1 indicate both that minimal winning coalitions are more likely to form than the other two types of cabinets and that surplus majority coalitions and minority cabinets are not significantly more or less likely to form relative to one another.22 It is clear, then, that minimal winning theory is a significant improvement on the intuitive idea that majority cabinets are more 22 We code the majority status variable by giving a “1” to a minority coalition. This makes interpretation of the results more straightforward in the models that follow, which include several variables interacted with minority status, but makes the interpretation of Model 1 less intuitive, since minimal winning status is essentially an interaction between majority status (i.e., 1-minority status) and minimal size. Accounting for the various interactions, however, is not difficult and simply means the proper coefficient for surplus majority cabinets is just -1 multiplied by the minority estimate (0.10 in Model 1), while the coefficient for minimal winning coalition is actually the sum of this number and the coefficient for minimal winning coalition that appears in the table (i.e., .84). 24 likely to form than cabinets that control only a minority of seats in the legislature. In addition, the results provide support for the idea that smaller coalitions are more likely to form than larger ones and that the inclusion of the largest party in the legislature in a potential government coalition increases its chances of forming. With respect to the impact of policy differences on coalition formation, the other models in Table 1 send three messages. First, the “pure” policy hypotheses consistently seem to beat out the hypotheses that combine both office and policy motivations. For example, the minimal winning connectedness variable is positive and significant in Models 2 and 3; however, it falls to insignificance when the more general measure of policy divisions in the coalition (“ideological divisions within the coalition”) is included in the specification (Model 4). Likewise, the variable marking the least ideologically divided minimal winning coalition has the wrong sign and is not statistically significant from zero in some specifications. These results may suggest that the restriction of policy considerations to the set of minimal winning coalitions is not a good idea. Rather, policy divisions seem to be important for all potential coalitions, regardless of their size. 23 A second message concerns the role of the median party in coalition formation. The results clearly support the “pure” policy models that suggest that the median party should be in the cabinet regardless of size considerations. Consequently, an important conclusion to draw from the results in Table 1 is that both size and ideology play and important role in coalition formation. No theory of coalition formation that ignores policy would ever predict a significant role for the median party, while no theory based purely on policy considerations would suggest that minimal winning coalitions should form. This suggests that future research should strive to create theories that explicitly incorporate the concepts of size and ideological centrality. <> 23 An alternative reading of these results may simply be that attempts to distinguish empirically between very subtle variations in the way policy considerations enter into the coalition formation process are probably futile. Specifically, the three variables in the model that speak to the ideological profile of a potential government—its minimal winning connected status, its ideological compactness, and its ideological divisions—produce results consistent with the idea that they are either highly collinear (as one might expect) or that they are all measuring a single underlying phenomenon. For example, Achen (1986) shows that when two variables both measure the same underlying variable with different and imperfect degrees of accuracy, estimates for the variable that is measured least accurately will be biased (and will often have the wrong sign) if the measurement error in the two variables is correlated. Thus, rather than depending on subtle and potentially unreliable interpretations of these coefficients, it may be more prudent simply to take away the message that policy distance matters and then to measure it in the most obvious way—namely, by using policy distance directly (which performs as expected and is highly significant) rather than the minimal connected winning or minimal ideological range variables (which are not significant or weakly significant and inconsistent with theoretical expectations). 25 A final message in these results is that for potential minority governments, the characteristics of the opposition they will face can be as important in their chances of forming as their own characteristics. Specifically, the results for Model 5 show that minority cabinets are more likely to form when the ideological divisions within the majority opposition they face are greater.24 This is very much in line with explanations of minority cabinets that emphasize the inability of the majority opposition to overcome their ideological differences and remove the cabinet (or prevent it from forming). Model 6 provides a streamlined version of the size and ideology model that we employ in subsequent models incorporating institutional variables.25 The Effects of Institutions In contrast to the hypotheses tested in the previous section, new-institutionalist hypotheses do not focus on how a potential coalition’s size and ideology affect its chances of forming in majority-rule settings. Instead, they focus on how the details of the institutional process surrounding government formation may advantage or disadvantage potential government coalitions with particular characteristics. Most of these hypotheses concern the effects of the institutions and norms surrounding the process of government formation itself; however, Laver and Shepsle’s important new theory adds a role for a different set of institutions—those that structure how decisions in the government are made after the government has formed. We test the various hypotheses stemming from these two versions of new-institutionalist theory in Tables 2 and 3 below. <> Institutions Structuring Pre-Formation Bargaining. Table 2 provides the estimates without Laver and Shepsle’s variables. The first thing to notice is that the estimates for the size and ideology variables examined in the last section are very robust to the inclusion of these new-institutional variables. A comparison of the direction and statistical significance of these variables between Tables 2 and 3 reveals that the relationships reported earlier 26 remain strong even when controlling for institutional influences. This is an important result since it is the first to examine the impact of the institutional and the traditional size and ideology variables in the same statistical model. The first hypothesis we discuss concerns the effects of the investiture vote on minority government formation. Our findings strongly support this hypothesis. Indeed, these results show that the effect of minority status is twice as large in systems with an investiture vote than in those without (shown by adding the coefficients for minority status and for the interaction of minority status and investiture). This is an important result for newinstitutionalist theory if only because so many have cited the fact that minority governments are less common in countries with investiture votes as evidence for the claim that institutions are important in general. Until now, however, researchers have presented no evidence that this comparison would hold up when controlling for other factors. In addition, this result adds credence not only to an institutional approach to government formation but also to the importance of policy considerations in that process. Specifically, the logic behind the explanation for the effects of the investiture vote is that where it is not found, minority governments can make issue-by-issue coalitions with different opposition parties. This presupposes the idea that parties are concerned about issues and not simply with capturing office benefits. Another hypothesis suggested above was that incumbent governments would enjoy an advantage in subsequent coalition negotiations due to their (supposed) position as the reversion point in bargaining. Despite our misgivings about the theoretical soundness of these arguments, the empirical results do support the conclusion that incumbent cabinets are more likely to form than other cabinets. More impressively, this is true even controlling for the fact that some incumbent governments commit themselves to re-form if they win the elections (accounted for by the pre-electoral pact variable) and for those variables (such as minimal winning status, etc.) that got them there in the first place. It seems, then, that incumbency is important for some reason. Our message to theorists would thus be that is probably worthwhile to tighten up the reversion point argument or to seek other arguments that would explain this result. Another new-institutionalist hypothesis discussed earlier was that that previous prime ministers, by virtue of their control over the timing of coalition formation, should be more likely to get into subsequent cabinets after a governmental crisis. Our analysis finds no support for this hypothesis, beyond the obvious fact that previous prime ministers are part of the incumbent governments that as a whole are likely to form again. Thus, we find no 27 advantage for the party of the incumbent prime minister that does not apply to all other parties in the cabinet. The next set of hypotheses is concerns the ability of the party selected as the formateur to use its procedural powers in the formation process itself (by making coalition proposals and otherwise setting the agenda for coalition negotiations) to guarantee its place in the cabinet and perhaps to shape the cabinet to its liking. As discussed above, in the absence of data on the identity of the formateur in each coalition formation opportunity, the argument that formateurs should be able to guarantee their place in the cabinet leads to three hypotheses that connect variables likely to be related to the selection of the formateur to the eventual choice of the government. Specifically, the hypotheses are that the largest party, the median party, and the party of the previous prime minister are all likely to be chosen as the formateur and so are likely to be included in the coalition that forms. Of course, since each of these variables is implicated in other theoretical arguments that do not use them as proxies for the formateur, simply including them in our empirical models does nothing to support or refute the role of the formateur in the process. Indeed, since the median and largest party variables are also rooted in the size and ideology tradition, results suggesting they are important cannot even help us distinguish the relative usefulness of these two traditions. As discussed above, however, conditioning on the identity of the final formateur (the prime minister in the new government) may help us isolate the source of the effects that these variables may have on coalition formation. For the largest party variable, the estimates in Models 7 and 8 indicate that while potential coalitions including the largest party are more likely to form in the full sample, this is no longer true once the model is made conditional on the identity of the formateur. This finding clearly indicates that being the largest party is important only in the selection of the formateur and not in the later selection of government partners. In other words, if the appointed formateur party were not the largest legislative party, then it would not then be more likely to bring in the largest party as a coalition partner. From these results, then, we are persuaded that the reason the largest party variable is important in the empirical explanation of cabinet composition is that it is a proxy for the party of the formateur, which can usually get into the cabinet. This stands in contrast to the explanation for the importance of the largest party provided by institution-free theorists such as Peleg (1981) and van Deemen (1989). The results for the median party variable provide a nice contrast to those for largest party variable. In both the full and restricted samples, the coefficient on the median party variable is positive and significant. Unlike 28 the largest party variable, then, the median party variable is important throughout the processes of cabinet formation.26 Such a finding does not refute the hypothesis that median status has some role in the designation of a formateur, but it also does not allow us to restrict the effect of the median party to that role. Instead, the result is consistent with the years of work by spatial modelers who have found that the median party is difficult to exclude from cabinets—apparently whether it is the formateur or not. Above we reported that there was little evidence that the party of the previous prime minister enjoys any particular advantage in getting into the government, aside from its status as a member of the incumbent cabinet. In the restricted sample, the situation is even worse for the party of the previous prime minister. Specifically, the significant negative coefficient suggests that if this party does not become the final formateur, it is actually less likely to be included in the cabinet that forms. This intuitively pleasing result suggests that a party controlling the office of prime minister in one administration that is not able to hold on to it in the succeeding administration is unlikely to enter the new government as a partner. What is interesting about the results for party of the previous prime minister in general is that our empirical analysis indicates no special role for this (ostensibly) procedurally advantaged party. Another result concerning the role of the formateur party in the process of government formation relates to its ability to shape the membership of the government to its liking. Again, one might expect that the formateur party will use its procedural powers in the coalition bargaining process to bias the ideological complexion of the government towards its own position. We test this expectation in Model 9, replacing the government ideological division variable with a variable measuring the ideological distance between the formateur (or the party of the prime minister, the final formateur) and the most ideologically distant government partner in each potential coalition. As expected, this variable is negatively signed and strongly significant. Thus, as the ideological division of a coalition, from the perspective of the prime minister, gets larger, the coalition is less likely to form.27 26 Technically, this result can only tell us that the variable is important in the second stage of the coalition formation process (that is, when the formateur is not the median party, it tends to select median parties to join it in government), but it does not directly show that median parties are important in the selection of the formateur in the first place. Other work, however, shows that median parties are more likely to be appointed as formateurs(Warwick 1996; Stevenson 1997). 27 This is consistent with both Warwick (1996) and Stevenson (1997), who show in multivariate analyses that the more ideologically distant a potential partner party is from the formateur, the less likely it is that this party will be included in the government. 29 Unfortunately, this result must remain partial due to limitations in the data. Namely, because of very high collinearity between this variable and the variable measuring the ideological division of the cabinet as a whole (a correlation coefficient above .9 in all specifications), it is impossible to include both variables in the same model in any meaningful way. Consequently, while we can be sure that ideological divisions in the cabinet are important to formation, we cannot be sure whether there is an independent effect for the divisions between the formateur party and its partners as opposed to divisions within the cabinet as a whole. The final results to examine from Table 2 concern the role of anti-system parties and pre-electoral commitments to exclude or include certain parties in the government. The effect of pre-electoral commitments is in line with theoretical expectations. When one party publicly excludes another party or parties from consideration as a coalition partner, it is very unlikely that a potential coalition containing these parties will form. Similarly, all else equal, a potential government that is announced before an election in the form of a pre-electoral coalition will be more likely to form than a similar potential government that is not announced. With respect to the anti-system status of parties, the effects are also in the expected direction and highly significant. This is an interesting result not only because it shows anti-system parties are excluded but also because this variable remains significant even when controlling for the anti-pact variable discussed above. Specifically, the anti-pact variable includes cases in which parties have gone on record that they will not coalesce with certain “anti-system” parties. Indeed, this variable captures most of the standard set of “non-coalitionable” parties in the party systems examined. This means, then, that the anti-system result depends mainly on “coalitionable” parties that take some anti-system stances but that do not go so far as to be categorically excluded from government. Institutions Structuring Post-Formation Government Decision-Making. The brand of newinstitutionalism pursued by Laver and Shepsle (1996) emphasizes a different set of institutions than the models discussed in the last section. Again, they suggest that the way that cabinets make decisions largely determines the set of feasible coalitions that can possibly come out of the cabinet formation process. Table 3 presents the results from the analysis of Laver and Shepsle’s theory.28 Again, the first thing to note is that, despite the differences in 28 In the empirical analysis that Laver and Shepsle provide in their book, they find strong support for these hypotheses, but their purpose was not to construct a fully controlled statistical model nor to evaluate the strength of their model against the various alternatives in the literature. We have tried to do that here by including their identification of very strong and merely strong parties (the definition of which they discuss in their book) in our empirical model. 30 sample size due to limited data on the Laver and Shepsle variables, little change occurs in the estimates of most the variables already discussed.29 The only exception is that the strongly significant median party variable, as well as the majority opposition divisions variable, fails to reach conventional levels of statistical significance. Below, we discuss the importance of these findings, but for now it is enough to note that in general the relationships we have already explored are robust to the inclusion of the Laver and Shepsle variables. We operationalized hypotheses 15 through 18 by including four additional variables in our models. The first is simply a dummy variable indicating whether a potential coalition includes a very strong party. A second variable provides an even stronger test of the VSP hypothesis since it singles out potential governments in which a single party rules alone as a minority government. We also calculated similar variables for coalitions containing merely strong parties.30 <> The findings in Table 3 provide strong support for the importance of very strong parties and less convincing evidence for the special role of merely strong parties. Very strong parties do tend to get into government and, even more, to rule alone. Further, the difference between the results for the full and restricted samples make it plain that very strong parties tend to hold the office of the prime minister in the government that forms.31 It also seems clear that the concept of the strong party beats out the uni-dimensional median as a way to capture the significance of a party’s weight and spatial centrality in coalition bargaining. This is an important finding and goes a long way towards showing that the Laver and Shepsle approach provides added value beyond the simpler hypothesis that central (in particular, median) parties are strategically well placed to get into the government. The evidence for the importance of merely strong parties is weaker. Specifically, while the coefficient on the merely strong party variable is positive, indicating that potential coalitions containing one of these parties are more likely to form, it is not statistically significant at conventional levels. More informative, perhaps, is the 29 The reduction in sample size, however, does lead to insufficient variation in the number of anti-pacts, thereby no longer permitting us to obtain reliable estimates of the anti-pact coefficient. 30 By definition, a potential coalition cannot contain both a very strong party and a merely strong party since they cannot occur in the same formation opportunity. In the restricted sample, the insignificant coefficient on VSP and the significant one on VSP forming alone indicate that most VSPs are the final formateurs and that the particular coalition they are likely to form is the one in which they rule alone. 31 31 almost significant negative coefficient on the interaction term that identifies potential governments in which merely strong parties would rule alone. When combined with the previous coefficient (as it must be since it is an interaction), the proper coefficient capturing the effect of being a merely strong party in a single-party coalition is indistinguishable from zero.32 This suggests, then, that when one controls for other factors, being merely strong may (weakly) help a party get into the government but apparently does not help it consistently to win the kinds of “standoffs” that would enable it to rule alone. Another interesting aspect of these results is that the effect of the investiture vote on the formation of minority governments remains strong and significant. This is interesting because, by definition, single-party potential governments are minority governments (since we consider only non-majority situations). From the set of minority governments, then, the strong party variables, when interacted with single-party government status, pull out a subset of minority governments that are very well placed and are probably the most likely sort of minority government to form. Even after controlling for these particularly likely minority governments, however, the effects of the investiture vote remain. One may conclude, therefore, that the effect of the investiture vote is not simply a reflection of a spurious correspondence between certain well-placed parties capable of forming minority governments and countries with investiture votes. Finally, the effects of the Portfolio Allocation variables remain strong after controlling for the ideological divisions within the opposition for a potential minority government. In the sample without the variables from the Portfolio Allocation approach, as shown in Model 7, the majority opposition divisions variable is strongly significant and positive, which is in line with theoretical expectations. In Models 10 and 11, however, in which we control for strong parties, the variable drops below standard levels of statistical significance. This is interesting because it mirrors the results above, suggesting that the strong party concept does a good job accounting for how the relative spatial positions of all the parties in a formation opportunity affect which governments form. In other words, not only are the ideological properties of a potential government important to its chances of formation, but the ideological properties of the legislature as a whole are important as well. 32 It is probably worth emphasizing that these coefficients should not be interpreted as showing an almost significant negative effect since when properly combined the effects are quite close to zero, given their standard errors. 32 Overall Predictive Accuracy One common critique of the cross-national quantitative or systematic theoretical approach to studying government formation is that the predictive accuracy of such approaches is rather poor (von Beyme 1983). Such critiques, however, seem to be somewhat biased against the quantitative approach because they are not based on results that take advantage of the full predictive power of a controlled multivariate model but on the admittedly uninspiring results of bivariate comparisons. To correct this deficiency, we have generated predicted probabilities of formation for each potential government in each formation opportunity in the data set (based on the coefficients of some of the models given above) and then compared these predictions with the actual governments that formed. Assessing the predictive success of models such as the ones we have estimated, however, is not particularly straightforward. One obvious option is to generate predicted probabilities from the model and then predict that the coalition with the highest probability will form. We can then easily compare these predictions with the coalitions that actually formed. Table 4 provides these comparisons for a selection of the models discussed above. << Table 4 about here >> One immediate conclusion we may draw from these results is that the traditional size and ideology variables (Model 6) appear to be rather limited predictors of government composition, as critics of coalition theory have long contended. Potential coalitions predicted by this set of variables as the most likely to form actually go on to form only eleven percent of the time. If we expand the prediction set to include the second most likely coalition to form, these variables perform somewhat better by predicting the actual government one out of five times. Still, the prediction rate of the size and ideology variables is substantially less than those from models incorporating institutional factors. A comparison with the predictions from Model 7 illustrates this point. Accounting for the impact of pre-formation institutions on coalition bargaining as well as the size and ideology variables increases the prediction rate by almost thirty percent. Together, these variables predict the government correctly a respectable forty percent of the time, if we consider only the potential coalition predicted with the highest probability. If we again expand the prediction set to include the next most likely potential coalition to form, the prediction rate jumps to over fifty percent. Finally, the prediction rates from Model 10, which incorporates the Portfolio Allocation variables, are the highest of all those considered thus far. The inclusion of these variables improves the predictions of Model 7 by approximately four percent. All said, then, the full model, 33 accounting for size and ideology influences and pre-formation and post-formation institutions, predicts the correct government approximately half the time. This is a real improvement of the usual predictive success of empirical models of cabinet formation. Indeed, given that in most of the bargaining situations for the countries in this study hundreds—and frequently, thousands—of coalitions could potentially form a government, and given the rather unsuccessful efforts of previous studies that have not relied on a multivariate statistical model, this prediction rate is extraordinary. We should note that other models (see Laver and Budge 1992) have reported higher success rates (seventy percent in some cases) but unlike our predictions, they gain this success by predicting a large number of coalitions and then finding that one of these predicted coalitions did in fact form. The important statistic for comparing the accuracy of our models with previous efforts, then, is the predictive efficiency of the models rather than a simple rate of success. This statistic is simply the proportion of cabinets correctly predicted out of all the cabinets that the theory predicted could have formed. For example, a prediction that a cabinet with the median party will form predicts half the possible coalitions, and so even if one were usually correct, predictive efficiency would be very low. If we focus on the first column of the table (in which we make a unique prediction), the efficiency rate for each of our models is the same as the accuracy rate. In contrast, Laver and Budge show seven general models for predicting cabinet formation with mean efficiency of .08, .25, .06, .10, .10, .07, and .22. Only in less general models that cannot make any prediction at all in 116 of the 209 cases they consider do their efficiency rates approach ours (three models with efficiency rates of .33, .40, and .46 for the cases in which they can make predictions). As revealing as the numbers in Table 4 are, however, they leave much of the predictive content of the empirical models unexplored. In particular, our models can produce predicted probabilities for every potential coalition and can therefore give us a complete ranking of the coalitions in each formation opportunity with respect to their relative likelihood of forming. We can then use this information to get a better idea of the relative success of the various models in moving the entire predictive distribution rather than just its maximum. Using the same three models from Table 4, Figure 1 presents the results of this exercise. To create these graphs, we first generated predicted probabilities from the coefficients in each of these empirical models. Then, within each formation opportunity, we ranked the potential coalitions according to their predicted probabilities of forming— 34 giving a “1” to the potential coalition with the highest predicted probability, a “2” to the second, and so forth. Next, we normalized the predictions by the number of potential coalitions in the formation opportunity using the following formula: (number of potential coalitions - rank of actual governing coalition) / (number of potential coalitions - 1). This provides a number that is equal to 1 if the actual coalition that formed was ranked first in the predictions and 0 if it is ranked last, varying appropriately between these extremes. The vertical bars in each panel of Figure 1 represent the frequency with which the coalition that actually went on to form the government (the “realized” coalition) fell within the corresponding predictive range indicated beneath each bar on the horizontal axis. <

> The results here clearly mirror those from Table 4 with respect to the relative predictive power of the three different model specifications discussed above. The pure size and ideology specification performs fairly well, but the inclusion of pre-formation bargaining constraints increases predictive accuracy substantially, with another (small) increase made evident by the inclusion of the Portfolio Allocation variables. In Model 6, for example, the realized coalition was ranked in the top five percent of potential coalitions predicted to form approximately twenty-five percent of the time (shown in the “Predictive Accuracy: Model 6” panel of Figure 1 by the far-right vertical bar). The distribution in this model is clearly skewed to the right and thick in the right tail, indicating that the relative predictive power of the size and ideology variables may have been somewhat underestimated by the results in Table 4, which took account only of the two highest predicted probabilities. Figure 1 (Model 6) reveals that the realized coalition was ranked in the top twenty-five percent of potential coalitions predicted to form by these variables approximately sixty percent of the time (shown by adding together the frequency of the five rightmost bars). Furthermore, only one out of four actual governments was predicted with low probability (which we define as less than fifty percent frequency). Figure 1 also demonstrates very clearly how the inclusion of institutional variables increases the predictive success of the size and ideology specification. The frequency of actual governments predicted with low probability falls to twelve percent in Model 7 and ten percent in Model 10. Moreover, the rightward skew of the distribution becomes much more pronounced in these latter models. The frequency of actual governments ranked in the top five percent of potential coalitions predicted to form increases by approximately thirty-five percent over 35 that of Model 6. Interestingly, this increase is almost equal to the density of the seventy-five to ninety-five percentile range in Model 6. Thus, it appears here that the most visible effect of adding institutional variables to the size and ideology variables has been to shift the frequency of correct predictions from the “fairly high” predictive accuracy range to the “very high” range. Conclusion The goal of this paper has been to subject the theoretical work in the coalition formation literature to a controlled statistical test. Having done this, we first should recognize that, despite a number of quite interesting and clarifying results (which are summarized below), our analysis has produced no “bombshells.” That is, we find, as the literature contends, that size and ideology are important determinants of cabinet composition when controlling for institutional variables and that institutional variables themselves contribute a lot to predicting and explaining cabinet formation independent of size and ideology. To some, the lack of a “bombshell” conclusion (e.g., institutions do not matter or perhaps that they are all that matters) may be a reason to discount the contribution of the analysis. To this, however, we have three responses. First is the obvious point that since no properly specified controlled statistical analysis has ever been done in this literature, we (and the discipline) could not know there would be no big surprises until the analysis was actually done and the results became known. In this respect, then, a confirmatory result is as valuable as a surprise. Second, one could argue that it is unrealistic to expect (or demand) such "bombshell" empirical results in a literature in which the theory is so well developed. Rather, what we should expect empirical work in this literature to contribute is not a complete revision of this carefully developed theoretical literature, but rather, within the context of broadly confirmatory results, more subtle findings that serve both to extend our knowledge of the details of the coalition formation process and to address the subtle distinctions in our numerous theories of this process. In this paper, we provide a wealth of such findings, each an extension of our empirical knowledge. Indeed, because of the large scope of our effort, these many findings constitute, in our opinion, a real step forward in the empirical understanding of cabinet formation.33 Finally, some of our findings, while they confirm the overall importance of political institutions or size and 33 Further, because (1) we use an appropriate empirical methodology, (2) we deal with a sample that includes almost all the cases available for analysis, and (3) the operationalization of most of the variables is uncontroversial (indeed, many of the concepts, like minimal winning status, are essentially measured without error), the more subtle conclusions are also 36 ideology, suggest important revisions to the standard thinking about what drives cabinet formation. We recognize that because we are dealing with numerous theories and findings that each contend for attention, the significance of our results can get lost in the clutter. To try to help navigate these results, then, we have bulleted some of them below: • In systems without an investiture vote, the impact of majority status on the chances a potential cabinet will form is negligible. Therefore, all else equal, majority cabinets and minority cabinets have the same chance of forming in these systems. A party is more likely to get into the government if it controls the largest number of legislative seats, just as several size-oriented theories have suggested, but only if it is successful in gaining control of the premiership. All else equal, then, the largest party is no more or less likely than any other party in the legislature to enter the government as a partner of the prime minister. Incumbency (in terms of being the incumbent coalition) has an independent, positive effect on the chances that a cabinet will form. This is true even when controlling for the factors that made the coalition an incumbent in the first place, including any pre-electoral commitments between incumbents to rule together. The above conclusion does not extend to incumbent prime ministers, apart from their role as members of the previous cabinet. Moreover, if the party of the prime minister in the outgoing government fails to become the formateur of the new government, it is actually less likely to enter the coalition. Laver and Shepsle's theory suggests that Merely Strong Parties should be able to rule alone in minority cabinets, but only if they can win "standoffs" with other parties. Our analysis suggests that while MSPs do tend to get into cabinets, they do not usually win such standoffs and therefore do not rule alone. Laver and Shepsle's notion of Very Strong Parties "beats out" median status and the ideological spread of the opposition as a way to predict whether minority cabinets will form. This is the only result from a controlled statistical analysis that supports the usefulness of their model.34 While we think that the real value of these results are in the specific details, and we hope that theorists will make use of them individually, there are some general messages that one can take away from the analysis as a whole. One is simply that current research on the way cabinets develop seems to be progressing in an empirically sound direction. New theories adding institutional detail to earlier considerations of size and ideology do a lot to help the predictive content of models of coalition formation and should continue to provide the focus for theoretical efforts. More than this, however, this work suggests that it is useful for institution-oriented approaches • • • • • dependable. That is, it is unlikely that possible changes in the data sample, the measurement of the theoretical variables, or other modeling issues will alter these conclusions. 34 Warwick (1996) included VSP and MSP variables in his analysis of the selection of the formateur and found that VSP was statistically significant and dampened the effect of median status. His interpretation of these results, however, was essentially negative because the size of the effect was small. 37 to look at the effect of the specific institutions that govern cabinet formation as well as the effect of institutions that govern what cabinets do after they form. Laver and Shepsle have begun well by looking at the rules governing cabinet decision-making, but there is much left to do in this area. For example, one could explore the effects of the rules governing the influence of the legislature on policymaking or of rules (such as confidence procedures) that govern the termination of cabinets.35 Finally, we now know that controlled statistical analysis is not going to “knock out” either of the main theoretical contenders for explaining coalition formation—size and ideology or institutions and norms. Further, within each of these general theoretical camps different models are supported—both policy and office-oriented theories as well as pre-formation and post-formation institutionalist approaches. This leaves us with the question of whether a single theory of coalition formation exists that is consistent with all these empirical findings. While we present this as a challenge to theorists, our own opinion is that researchers have overemphasized the differences in coalition theories and have been slow to recognize their compatibilities. Coalition formation is a complex political outcome that is no doubt the result of variously motivated politicians bargaining in institutionally rich environments. Like blind men, each touching different parts of the same elephant, each of the theories in the cabinet formation literature focuses on only part of the whole picture and thus has only limited explanatory power. We have made progress. Indeed, when we take all the theories as a group, we have at least explored much of the elephant. The challenge now is to integrate that knowledge into a single coherent theory. What is clear from our analysis is that such an integrative theory will necessarily allow for multiple kinds of institutions (i.e., both pre- and post-formation) and multiple kinds of goals for politicians (i.e., both office and policy goals). While the complexity of such models has stymied previous integrative efforts, we are encouraged by the very recent work of some formal theorists, such as Baron and Diermeier (1999), who have put forward more simple models that nonetheless include both office and policy motivations and explicitly model both the coalition bargaining process and post-coalition policymaking in an institutionally complex environment.36 35 At least one very recent paper (Baron 1999) has examined in a game-theoretic model how different confidence procedures might affect cabinet formation. The key to making these models tractable is the assumption of “efficient bargaining.” That is, politicians can freely trade office benefits for policy concessions such that all coalition bargaining is efficient. 36 38 References Achen, Christopher H. 1986. The Statistical Analysis of Quasi-Experiments. Berkeley: UC Press. Austen-Smith, David, and Jeffrey Banks. 1988. “Elections, Coalitions, and Legislative Outcomes.” American Political Science Review 82 (June): 405-22. Axelrod, Robert. 1970. Conflict of Interest. Chicago: Markham. Baron, David P. 1998. “Comparative Dynamics of Parliamentary Governments.” American Political Science Review 92 (September): 593-610. Baron, David P. 1993. “Government Formation and Endogenous Parties.” American Political Science Review 87 (March): 34-47. Baron, David P. 1991. “A Spatial Bargaining Theory of Government Formation in a Parliamentary System.” American Political Science Review 84 (September): 891-906. Baron, David P., and Daniel Diermeier. 1999. “Elections, Governments, and Parliaments in Proportional Representation Systems.” CMS-EMS Discussion Paper, No. 1215. Kellogg Graduate School of Management, Northwestern University. Baron, David P., and John A. Ferejohn. 1989. “Bargaining in Legislatures.” American Political Science Review 83 (December): 1181-206. Bartolini, Stefano, and Peter Mair. 1990. Identity, Competition and Electoral Availability: The Stabilisation of European Electorates 1885-1985. Cambridge: Cambridge. Browne, Eric C. 1970. “Testing Theories of Coalition Formation in the European Context.” Comparative Political Studies 3 (January): 391-412. Budge, Ian, and Hans Keman. 1990. Parties and Democracy: Coalition Formation and Government Functioning in Twenty States. New York: Oxford University Press. de Swaan, Abram. 1973. Coalition Theories and Government Formation. Amsterdam: Elsevier. Diermeier, Daniel, and Randolph T. Stevenson. 1999. “Cabinet Survival and Competing Risks.” American Journal of Political Science 43 (October): 1051-68. Franklin, Mark N., and Thomas T. Mackie. 1984. "Reassessing the Importance of Size and Ideology for the Formation of Governing Coalitions in Parliamentary Democracies." American Journal of Political Science 28 (November): 671-92. Gamson, William. 1961. "A Theory of Coalition Formation." American Sociological Review 26 (June): 373-82. Greene, William H., 1993. Econometric Analysis. 2d ed. New York: Macmillan. Huber, John D. 1996. “The Vote of Confidence in Parliamentary Democracies.” American Political Science Review 90 (June): 269-82. Klingemann, Hans-Dieter, and Andrea Volkens. 1992. “Coalition Governments in the Federal Republic of Germany: Does Policy Matter?” In Party Policy and Government Coalitions, eds. M. J. Laver and Ian Budge. New York: St. Martin’s Press. Laver, M. J. 1992. “Coalition and Party Policy in Ireland.” In Party Policy and Government Coalitions, eds. M. J. Laver and Ian Budge. New York: St. Martin’s Press. Lieserson, Michael. 1968. “Factions and Coalitions in One-Party Japan: An Interpretation Based on the Theory of Games.” American Political Science Review 62 (March): 70-87. Luebbert, Gregory. 1986. Comparative Democracy: Policy Making and Governing Coalitions in Europe and Israel. New York: Columbia University Press. Lupia, Arthur, and Kaare Strøm. 1995. “Coalition Termination and the Strategic Timing of Parliamentary Elections.” American Political Science Review 89 (September): 648-65. McFadden, D. 1973. "Conditional Logit Analysis of Qualitative Choice Behavior." In Frontiers in Econometrics, ed. P. Zarembka. New York: Academic Press. McFadden, D. 1974. "The Measurement of Urban Travel Demand." Journal of Public Economics 3 (Nov): 303-28. Mackie, Thomas T., and Richard Rose. 1991. The International Almanac of Electoral History. 3d ed. London: Macmillan. Martin, Lanny W., and Randolph T. Stevenson. 1995. "A Unified Model of Government Formation and Survival." University of Rochester. Typescript. Mastropaolo, Alfio, and Martin Slater. 1992. “Party Policy and Coalition Bargaining in Italy, 1948-87: Is There Order Behind the Chaos?” In Party Policy and Government Coalitions, eds. M. J. Laver and Ian Budge. New York: St. Martin’s Press. Peleg, Bazalel. 1981. “Coalition Formation in Simple Games with Dominant Players.” International Journal of Game Theory 1: 11-13. Quinn, Kevin M., Andrew D. Martin, and Andrew B. Whitford. “Voter Choice in Multi-Party Democracies: A Test of Competing Theories and Models.” American Journal of Political Science 43 (October): 1231-47. Riker, William. 1962. The Theory of Political Coalitions. New Haven: Yale University Press. Strøm, Kaare. 1990. Minority Government and Majority Rule. Cambridge: Cambridge. Stevenson, Randolph T. 1997. “How Parties Compete: Electoral Performance and Government Participation in Parliamentary Democracies.” Ph.D. dissertation, University of Rochester. Strøm, Kaare. 1994. “The Presthus Debacle: Intraparty Politics and Bargaining Failure in Norway.” American Political Science Review 88 (March): 112-27. Strøm, Kaare, Ian Budge, and Michael Laver. 1994. "Constraints on Government Formation in Parliamentary Democracies." American Journal of Political Science 38 (May): 303-35. Tops, Pieter, and Karl Dittrich. 1992. “The Role of Policy in Dutch Coalition Building, 1946-81.” In Party Policy and Government Coalitions, eds. M. J. Laver and Ian Budge. New York: St. Martin’s Press. van Deeman, A. M. A. 1989 “Dominant Players and Minimum Size Coalitions.” European Journal of Political Research 23: 1-33. von Beyme, Klaus. 1983. “Governments, Parliaments and the Structure of Power within Political Parties.” In Western European Party Systems, eds. H. Daalder and P. Mair. London: Sage. von Neumann, I., and O. Morgenstern. 1953. Theory of Games and Economic Behavior. Princeton: Princeton University Press. Warwick, Paul. 1994. Government Survival in Parliamentary Democracies. New York: Cambridge. Warwick, Paul. 1996. “Coalition Government Membership in Western European Parliamentary Democracies.” British Journal of Political Science 26 (October): 471-99. Whitten, Guy D., and Harvey D. Palmer. 1996. “Heightening Comparativists' Concern for Model Choice: Voting Behavior in Great Britain and the Netherlands.” American Journal of Political Science 40 (Feb): 231-60. 

Shared by: Myrna Carlson

Myrna Carlson

retired

counselor/professor

contact me

In order to send a message you need to be logged in.

About

Home-schooling my youngest child (16). Small on-line bookseller. Unpublished writing.

Other docs by Myrna Carlson

Politics+of+India

Views: 86  |  Downloads: 1

Politics+of+globalization+2

Views: 85  |  Downloads: 1

Politics+and+Government+of+Africa-1

Views: 75  |  Downloads: 0

Todays+Agenda+Political+Parties+It+is+a+Party+Up+in+here

Views: 76  |  Downloads: 0

Nonnested+Model+Testing+for+World+Politics

Views: 82  |  Downloads: 0

Executive+Popularity+in+France+Version

Views: 88  |  Downloads: 0

Political+Marketing-1

Views: 91  |  Downloads: 3

RELIGION+AND+POLITICS+IN+TWENTIETH+CENTURY+AUSTRALIA-1

Views: 98  |  Downloads: 2

The+Political+Philosophy+of+Deception+Types+and+Consequences+of+Political+Lying

Views: 76  |  Downloads: 0

The+implications+of+taking+on+a+political+economy+approach

Views: 51  |  Downloads: 0

TRANSPARENCY+INTERNATIONAL+CORRUPTION+IN+EDUCATION

Views: 55  |  Downloads: 0

Policies+Prototypes+and+Presidential+Approval

Views: 43  |  Downloads: 0

What+Are+Your+Political+Beliefs

Views: 55  |  Downloads: 0

Internet+Political+Surveys

Views: 52  |  Downloads: 0

Mexican+presidential+politics+Will+a+dinosaur+win

Views: 40  |  Downloads: 1

Related docs

Government Formation in Parliamentary Democracies

Views: 40  |  Downloads: 1

About:

What is Docstoc? | Docstoc Terms of Service | DocStore Terms of Service | Privacy Policy | FAQs

Links:

Docstoc Blog | Join Pro | Requests | Docsters | Upload | DMCA Guidelines | DMCA Notification | Docstoc API | RSS Feeds

Contact:

Contact Us | Suggest Features | Join Our Team

Share:

Embed Documents | Docstoc OneClick | Docstoc Sync 

© Docstoc 2009. All rights reserved.