¹ Klingemann, Hans-Dieter, Hofferbert, Robert and Budge, Ian, Parties, Policies, and Democracy (Boulder, Colo.: Westview, 1992).Google Scholar

² Two of the most prominent efforts are Budge, Ian, Robertson, David and Hearl, Derek, eds, Ideology, Strategy and Party Change: A Spatial Analysis of Post-War Election Programmes in Nineteen Democracies (Cambridge: Cambridge University Press, 1987)CrossRefGoogle Scholar; and Laver, Michael and Hunt, W. Ben, Policy and Party Competition (London: Routledge, 1992).Google Scholar

³ A minimal winning government is a government that would lose its parliamentary majority if it lost any of its member-parties. This hypothesis is most closely associated with Riker, William, The Theory of Political Coalitions (New Haven, Conn.: Yale University Press, 1962).Google Scholar

⁴ Axelrod, Robert, Conflict of Interest (Chicago: Markham, 1970).Google Scholar

⁵ See, for example, Baron, David, ‘A Spatial Bargaining Theory of Government Formation in Parliamentary Systems’, American Political Science Review, 85 (1991), 135–64.CrossRefGoogle Scholar

⁶ Another strategy, adopted by Budge, Ian and Keman, Hans, Parties and Democracy: Coalition Formation and Government Functioning in Twenty States (Oxford: Oxford University Press, 1990)Google Scholar, is to attempt to account for the emergence of certain broad types of coalition governments (rather than the precise composition of individual governments).

⁷ Laver, Michael and Shepsle, Kenneth, Making and Breaking Governments: Cabinets and Legislatures in Parliamentary Democracies (Cambridge: Cambridge University Press, 1996).CrossRefGoogle Scholar

⁸ Among the systems to be considered here, the exceptions are West Germany and Ireland, where the government leader is elected by parliament, and Sweden since 1975, where the leader is appointed by the Speaker of the lower house.

⁹ Laver, and Hunt, , Policy and Party Competition.Google Scholar

¹⁰ The median voter theorem was established in Black, Duncan, The Theory of Committees and Elections (Cambridge: Cambridge University Press, 1958)Google Scholar; the voting-cycles or ‘chaos’ theorem is attributable to McKelvey, Richard, ‘General Conditions for Global Intransitivities in Formal Voting Models’, Econometrica, 47 (1976), 1085–111.CrossRefGoogle Scholar

¹¹ Laver, and Shepsle, , Making and Breaking Governments.Google Scholar

¹² Laver, Michael and Shepsle, Kenneth, ‘Coalitions and Cabinet Government’, American Political Science Review, 84 (1990), 873–90, at p. 888.CrossRefGoogle Scholar

¹³ Laver, and Shepsle, , Making and Breaking Governments, pp. 66–9.Google Scholar

¹⁴ Laver, and Shepsle, , Making and Breaking Governments, pp. 69–78.Google Scholar

¹⁵ The systems are Austria, Belgium, Denmark, Finland, France (Fourth Republic), Iceland, Ireland, Italy, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Britain and West Germany. Following standard practice, a change in government is considered to have taken place whenever an election has been held, a government has resigned or been defeated in parliament, or the prime minister or party composition of the government has changed. Basic information on governments and parties is derived primarily from Keesing's Contemporary Archives (London: Keesing's Publications, 1945–1990).Google Scholar

¹⁶ The exception is the portfolio allocation hypothesis. Since strong-party status is a function of size and ideological location, it should occupy an intermediate position between these factors and government membership. This expectation will be examined.

¹⁷ This proportion was not calculated for the first five governments in each system, since proportions based on so tew cases could be highly misleading. The consequence is that the first five governments in each system are missing for all analyses in which this variable appears.

¹⁸ Parties not in any previous government are treated as being out of government from the beginning of the period covered by the data set (the first election since the Second World War). An unfortunate consequence is that parties in the early post-war period tend to receive relatively low values even if they had never had government experience. This is controlled to some extent, however, by the fact that the Experience variable excludes the first five governments in each system.

¹⁹ Note that this is not simply a question of the head of state choosing a member of the outgoing government to form the new one, since we are considering successful formateurs only.

²⁰ Absence is removed in order to augment the number of cases: if left in, it remains insignificant with each addition to the model's independent variables.

²¹ Dodd, Lawrence, Coalitions in Parliamentary Government (Princeton, NJ: Princeton University Press. 1976).Google Scholar

²² Browne, Eric C., Gleiber, Dennis W. and Mashoba, Carolyn, ‘Evaluating Conflict of Interest Theory: Western European Cabinet Coalitions, 1945–80’, British Journal of Political Science, 4 (1984), 1–32.CrossRefGoogle Scholar

²³ Castles, Francis and Mair, Peter, ‘Left-Right Political Scales: Some “Expert” Judgments’, European Journal of Political Research, 12 (1984), 73–88.CrossRefGoogle Scholar

²⁴ Party positions on the three sets of scales intercorrelate at above the 0.9 level. For France, Luxembourg and Iceland, all of which Castles and Mair omit, party positions are derived from the other two sources. All three sources omit Portugal, which becomes missing in subsequent analyses.

²⁵ Powell, G. Bingham, Contemporary Democracies (Cambridge, Mass.: Harvard University Press, 1982), pp. 233–4.Google Scholar

²⁶ Laver and Shepsle limited the application of the Laver-Hunt data to periods in which they judged that party positions had not changed significantly from the time (1989) of the survey. This practice, which is followed here, costs a substantial amount of data, including the entire French Fourth Republic.

²⁷ Median, empty-winset median, and strong parties were determined by means of the WINSET program and associated data files, supplied by Ken Shepsle.

²⁸ If Median (LR) is excluded from the equation, then Median (LH1) enters in its place. This reflects the fact that LH1, based as it is solely on the issue of public spending, is a more narrowly-defined version of the left-right dimension.

²⁹ The second dimension concerns foreign policy in all countries except Austria, where permissiveness is the issue, and Belgium, where it relates to decentralization of decision making.

³⁰ The coefficients are 1.20 (SE = 0.52) and 0.37 (SE = 0.42) for VSP and MSP, respectively, based on 938 cases. Further confining the analysis to the relatively few situations where a VSP exists (n = 480) shows VSP status to be barely significant in a one-tailed test (t = 1.72).

³¹ The %² associated with the improvement in the model's log-likelihood brought about by the addition of VSP to the model is 5.75, p = 0.017. The proportion of parties correctly classified by the model increases 0.4 per cent.

³² Inflation variables were calculated from data on consumer prices in the international Labour Organisation's International Labour Review (1945–1963)Google Scholar and Bulletin of Labour Statistics (1964–1970)Google Scholar, as well as the International Monetary Fund's International Financial Statistics (1970–1990)Google Scholar. Unemployment data came primarily from the UN Monthly Bulletin of Statistics (1947–1990)Google Scholar, supplemented by the OECD Historical Statistics 1969–88 (1990)Google Scholar. The GDP data are from the OECD National Accounts Vol. I: Main Aggregates 1950–78 and 1960–88 (1980, 1990)Google Scholar and the OECD Economic Outlook (1990)Google Scholar. Further details are available on request.

³³ Data are frequently absent in the early post-war years; in addition, changes in the method of calculating the unemployment rate often made earlier values non-comparable with later values. A second source of missing data is definitional: parties that were never previously in government are missing on all of the economic variables.

³⁴ Without the causal factors, there are significant differences across systems. Further examination reveals that this is largely due to variations in party-system fragmentation: other things being equal, the more fragmented the party system, the lower the chances that any one party will become a formateur. Other things are not equal, however. In fact, many parties in fragmented systems stand no chance of becoming formateurs. The consequence is that once the size and experience variables are introduced into the model, any effects due to party-system fragmentation as well as any differences across systems become insignificant. Fragmentation and other party-system attributes will be discussed in the next section, where they play a more prominent role.

³⁵ Strictly speaking, this is only true of the system dummies; the decade dummies can be eliminated without affecting the model in any appreciable way.

³⁶ The economic variables are excluded because they would have eliminated too many cases and play no role in the final model.

³⁷ Without knowledge of the independent variables, it is possible to guess formateur status correctly in 83.8 per cent of cases by choosing the modal category (the party is not a formateur). The model thus achieves a proportional reduction in error of 51.3 per cent.

³⁸ This may even be the case if the formateur party is a core party or a very strong party. The existence of either signifies that no majority favours some other policy point to that party's ideal point, but this need not mean that other parties will join it and help implement its policies. The advantages of office may be more than offset by the liabilities, electoral and otherwise, of a policy sell-out.

³⁹ Laver, Michael and Budge, Ian, eds, Party Policy and Government Coalitions (New York: St Martin's, 1992)CrossRefGoogle Scholar. The European Manifestos Project is described in Budge, , Robertson, and Hearl, , Ideology, Strategy and Party Change.Google Scholar

⁴⁰ The formateur distance variable is available for a total of 976 non-formateur parties. The size and experience variables in the model reduce the number to 850, while Formateur Distance (LR) drops it to 813. The Formateur Distance variable based on the Laver-Budge data remains significantly related to government membership when either (or both) of these sets of independent variables is excluded to increase the number of cases.

⁴¹ The number of cases would be reduced to 565. For what it's worth, none of the Laver-Hunt ideological variables enters the model significantly in this scenario.

⁴² The sample reduction is partially responsible for weakening PSPs' role in this model, but further analysis shows that the presence of Distance-Formateur (LH2) alone eliminates the PSP effect without producing a significant loss of cases. If MSP is included in the analysis (on the argument that the hypothesis allows for merely strong parties to be coalition partners), it, too, becomes insignificant when the other independent variables are added to the model.

⁴³ This choice has no substantive effect on the results to be presented.

⁴⁴ The coefficient associated with this variable is −0.30 (SE = 0.14), indicating an effect that is fairly strong but not highly significant. As with formateur status, the absence of stronger evidence of a connection between past economic performance and the odds of becoming a coalition partner is not due to an intervening effect by the size variables.

⁴⁵ Systems where the prime minister is chosen by the legislature are not considered to have an investiture requirement, since the vote is for an individual, not a coalition. The results do not change, however, if these systems are classified as having investiture.

⁴⁶ Laakso, Markku and Taagepera, Rein, ‘“Effective” Number of Parties: A Measure with Application to West Europe’, Comparative Political Studies, 12 (1979), 3–27CrossRefGoogle Scholar. The measure is defined as the inverse of one minus the sum of squared proportions of party seats.

⁴⁷ Strom, Kaare, ‘Party Goals and Government Performance in Parliamentary Democracies’, American Political Science Review, 79 (1985), 738–54CrossRefGoogle Scholar; King, Gary, Alt, James, Burns, Nancy and Laver, Michael, ‘A Unified Model of Cabinet Dissolution in Parliamentary Democracies’, American Journal of Political Science, 34 (1990), 846–71.CrossRefGoogle Scholar

⁴⁸ Strom, , ‘Party Goals and Government Performance’, p. 742.Google Scholar

⁴⁹ Note that although polarization is measured in terms of the anti-system presence in the parliament, its effect cannot be attributed to the fact that anti-system parties are generally non-coalitionable. This is indicated by the earlier finding that the anti-system status of a party is not significantly related to coalition partnership, once other relevant factors are taken into account.

⁵⁰ Note that only governments that began as caretakers are included in this definition; regular governments that terminated but continued in office until new governments could be formed are not treated as caretaker governments. In addition, caretaker governments that do not have a party composition were excluded from the data set.

⁵¹ Laver, and Shepsle, , Making and Breaking Governments, pp. 262–6.Google Scholar

⁵² The regression coefficients are −0.222 (SE = 0.078) for Polarization and 0.029 (SE = 0.021) for the variable representing a VSP formateur. The analysis was based on 256 governments with valid data for all variables.

⁵³ Guessing the modal category (non-participation) without knowledge of the independent variables would produce a success rate of 70.1 per cent; the proportional reduction in error of the model is therefore 48.3 per cent.

⁵⁴ Confining the analysis to first post-election governments shows five of the independent variables in the formateur model still significant (n = 568). These live do not include First Party.

⁵⁵ Part of the problem may be that PA theory aims to predict which parties will hold key portfolios, not which parties will be in government. Nevertheless, one would expect that merely strong parties and their partners would be more likely to enter governments than other parties, other things being equal. Laver and Shepsle themselves test their theory with government membership as the dependent variable (Making and Breaking Governments, pp. 174–92)Google Scholar. Incidentally, these tests show stronger relationships when performed on the present data set, which indicates that it does not incorporate any systematic bias against the PA approach.