¹ An alternative method of resolving disagreements between two houses in bicameral legislatures is the conference committee. Although conference committees have received substantial attention, especially in the United States, the navette system – which is also employed in the US Congress – has been overlooked. See, for example, Shepsle, Kenneth A. and Weingast, Barry R., ‘The Institutional Foundations of Committee Power’, American Political Science Review, 81 (1987), 85–104CrossRefGoogle Scholar; Shepsle, Kenneth A. and Weingast, Barry R., ‘Why are Congressional Committees Powerful?’ American Political Science Review, 81 (1987), 935–45Google Scholar; Krehbiel, Keith, ‘Why are Congressional Committees Powerful?’ American Political Science Review, 81 (1987), 929–35CrossRefGoogle Scholar, and Smith, Stephen S., ‘An Essay on Sequence, Position, Goals, and Committee Power’, Legislative Studies Quarterly, 13 (1988), 151–76CrossRefGoogle Scholar. This omission is surprising because over 30 per cent of the world's countries have bicameral legislatures (Inter-parliamentary Union, Parliaments of the World (Aldershot, Hants.: Gower, 1986))Google Scholar and in the overwhelming majority of these, agreement between the two houses is reached through the navette system (Money, Jeannette and Tsebelis, George, ‘Cicero's Puzzle: Upper House Power in Comparative Perspective’, International Political Science Review, 13 (1992), 25–42.CrossRefGoogle Scholar

² See Grangé, Jean, ‘L'Efficacité normative du Sénat’, Revue Française de Science Politique (1984), 32–84Google Scholar; and Mastias, Jean and Grangé, Jean, Les Secondes Chambres duparlemeni en Europe occidentale (Paris: La Documentation Française, 1987).Google Scholar

³ A recent example is the override of presidential veto concerning Chinese students in the United States. The Republicans in the House voted overwhelmingly to override, leaving to their colleagues in the Senate the unpleasant task of sustaining the presidential veto.

⁴ During the period 1959–80 the Gaullist party dominated the National Assembly and Gaullist allies the Senate, so there is sufficient continuity for the usual statistical assumptions to be met. For the Mitterrand period (1981–86), we do not perform statistical tests, but the model helps us form expectations about the interactions of the two chambers when the Left controlled the National Assembly and the Right, the Senate.

⁵ See Maus, Didier, ed., Les Grands Textes de la pratique institutionelle de la V^e République (Paris: La Documentation Française, 1987)Google Scholar, for the constitutional provisions and Bourdon, Jean, Les Assemblées parlementaires sous la V^e République (Paris: La Documentation Française, Notes et études documentaires No. 4463–64, 1978), pp. 109–45Google Scholar, among others, for a detailed description of the process.

⁶ See Luchaire, François and Conac, Gérard, La Constitution de la République Française, 2nd edn (Paris: Economica, 1987, p. 914.Google Scholar

⁷ In practice, the Senate selects its representatives through proportional representation, while the National Assembly is represented by its own majority. After the victory of the Left in 1981, the National Assembly regulations were modified to resemble those of the Senate (Grangé, Jean, ‘L'Éfficacité normative du Sénat’).Google Scholar

⁸ The government, if displeased with the joint committee compromise, is not required to submit the joint text to Parliament for final approval. After fifteen days, the house which last read the bill can restart the legislative process (see Bourdon, , Les Assemblées parlementaires, p. 125).Google Scholar

⁹ See Stone, Alec, The Birth of Judicial Politics in France: The Constitutional Council in Comparative Perspective (Oxford: Oxford University Press, 1992).Google Scholar

¹⁰ For a more detailed analysis of the way Article 49.3 operates, see Tsebelis, George, Nested Games: Rational Choice in Comparative Politics (Berkeley: University of California Press, 1990), chap. 7Google Scholar. The government has used this method to impose its will against the majority of both houses five times over the course of the Fifth Republic (Maus, , Les Grands Textes, pp. 193–264).Google Scholar

¹¹ For a discussion of the role of government in the selection of procedures to influence legislative outcomes, see Huber, John, ‘Restrictive Legislative Procedures in France and the United States’, in American Political Science Review, 86 (1992), 675–87CrossRefGoogle Scholar. For a discussion of the role of the Constitutional Council see Stone, , Judicial Politics in France.Google Scholar

¹² Grangé, Jean, ‘Attitudes et vicissitudes du Sénat (1958–1980)’, Revue Française de Science Politique (1981), 32–84CrossRefGoogle Scholar; Grangé, , ‘L'Efficacité normative du Sénat’Google Scholar; Mastias, and Grangé, , Les Secondes ChambresGoogle Scholar; Tardan, Arnaud, ‘Le Rôle législatif du Sénat’, Pouvoirs, 44 (1988), 104–10Google Scholar; Maus, Didier, ‘Le Sénat’, Pouvoirs, 44 (1988), 119–30Google Scholar; and Mastias, Jean, ‘Histoire des tentations du Sénat de la v^e République’, Pouvoirs, 44 (1988), 15–34Google Scholar. For a more detailed history with additional sub-periods, see Mastias, Jean, Le Sénat de la V^e République: réforme et renouveau (Paris: Economica, 1980).Google Scholar

¹³ The legislative analyses begin in 1959 because the first legislature under the constitution of the Fifth Republic was elected in late November 1958 and only began its work in 1959.

¹⁴ Our analysis ends with the 1986 National Assembly election when the Socialist majority was replaced by a conservative coalition of Gaullists and Giscardians. This period marks the so-called cohabitation of the socialist president with a prime minister and National Assembly of the Right.

¹⁵ Grangé, , ‘L'efficacité normative’Google Scholar; and Mastias, and Grangé, , Les Secondes Chambres.Google Scholar

¹⁶ During the period 1946–72, 5.6 per cent of the bills became laws after five or more iterations of the navette system (Trivelli, Laurent, Le Bicamerisme-institutions comparées–étude historique, statistique et critique des rapports entre le Conseil national et le Conseil des États (Lausanne: Payot, 1975))Google Scholar. Bills that are introduced in both chambers more than three times exist even in France (Bourdon, , Les Assemblées parlementaires, p. 124).Google Scholar

¹⁷ Grangé, , ‘L'Efficacité normative du Sénat’.Google Scholar

¹⁸ Quoted in Grangé, , ‘L'Efficacité normative du Sénat’Google Scholar;, from Debu-Bridel, J. in Notre République of 29 04 1966.Google Scholar

¹⁹ Some bills took as little as ten days for passage (such as the 1974 reform of the Office of French Radio and Television – ORTF), whereas others took several months. The Senate won many concessions on the ORTF legislation, whereas it was unable to obtain its preferences in the 1973 bill on commerce and crafts. For a detailed discussion of several bills and the influence of the Senate in the process, see Money, Jeannette and Tsebelis, George, ‘The Political Power of the French Senate’ (Journal of Legislative Studies, forthcoming).Google Scholar

²⁰ The reader may object to the heroic simplification of houses as unified players. Indeed, McKelvey and Schofield have shown that most of the time under majority rule there is no ideal point for a group of rational players (McKelvey, Richard, ‘General Conditions for Global Intransitivities in Formal Voting Models’, Econometrica, 47 (1979), 1085–112CrossRefGoogle Scholar, and Schofield, Norman, ‘Instability of Simple Dynamic Games’, Review of Economic Studies, 45 (1978), 575–94)CrossRefGoogle Scholar. However, the assumption of unified players can be relaxed if we assume that the problem is decomposed to its different dimensions, and a decision is made on one dimension at a time (see Shepsle, Kenneth A., ‘Institutional Arrangements and Equilibrium in Multidimensional Voting Models’, American Journal of Political Science, 23 (1979), 27–57CrossRefGoogle Scholar; Cox, Gary W. and McKelvey, Richard, ‘A Ham Sandwich Theorem for General Measures’, Social Choice and Welfare, 1 (1984), 75–83)CrossRefGoogle Scholar. In this case, the ideal point of each house coincides with the corresponding median voter's.

²⁰ The reader may object to the heroic simplification of houses as unified players. Indeed, McKelvey and Schofield have shown that most of the time under majority rule there is no ideal point for a group of rational players (McKelvey, Richard, ‘General Conditions for Global Intransitivities in Formal Voting Models’, Econometrica, 47 (1979), 1085–112CrossRefGoogle Scholar, and Schofield, Norman, ‘Instability of Simple Dynamic Games’, Review of Economic Studies, 45 (1978), 575–94).CrossRefGoogle Scholar However, the assumption of unified players can be relaxed if we assume that the problem is decomposed to its different dimensions, and a decision is made on one dimension at a time (see Shepsle, Kenneth A., ‘Institutional Arrangements and Equilibrium in Multidimensional Voting Models’, American Journal of Political Science, 23 (1979), 27–57CrossRefGoogle Scholar; Cox, Gary W. and McKelvey, Richard, ‘A Ham Sandwich Theorem for General Measures’, Social Choice and Welfare, 1 ( 1984), 75–83).CrossRefGoogle Scholar In this case, the ideal point of each house coincides with the corresponding median voter's.

²¹ Rubinstein, Ariel, ‘Perfect Equilibrium in a Bargaining Model’, Econometrica, 50 (1982), 97–109CrossRefGoogle Scholar; and Rubinstein, Ariel, ‘A Bargaining Model with Incomplete Information About Time. Preferences’, Econometrica, 53 (1985), 1151–72.CrossRefGoogle Scholar

²² More precisely our representation is isomorphic, that is, there is a correspondence between the two models, so that one can solve any problem on one model and then transpose the solution to the equivalent solution of the other in a unique way.

²³ There is, however, one major difference between the bargaining game of ‘divide the dollar’ and ‘agree on a bill’. While both games assume the initial position of each player known, the ‘divide the dollar’ game is safe in assuming that each player's initial position is to want the whole dollar, while in our game there is room for strategic misrepresentation of (initial) preferences. For example, if the Senate wants a smaller (but positive) deficit than the National Assembly it could start the bargaining process by claiming that it wants no deficit at all. We shall discuss this problem in more detail in the incomplete information problem that follows.

²⁴ Grossman, Sanford J. and Perry, Motty, ‘Sequential Bargaining Under Asymmetric Information’, Journal of Economic Theory, 39 (1986), 120–54.CrossRefGoogle Scholar

²⁵ If the number of rounds is not an integer, that is, if one house introduces the bill and the other applies the stopping rule, the advantage depends on the time discount factors of both houses.

²⁶ Grossman, and Perry, , ‘Sequential Bargaining’.Google Scholar

²⁷ Alternatively, one can make use of the results of the complete information model and consider the time discount factor of the National Assembly as unknown by the Senate. The National Assembly could be patient (have a high discount factor, in which case we would call it strong) or impatient (have a low discount factor, in which case we would call it impatient) or have any value between the two extremes. We shall make use of this interpretation in the empirical section.

²⁸ A Bayesian perfect equilibrium is an equilibrium where both players play equilibrium strategies in every subgame and update their beliefs along the equilibrium path using Bayes's rule. This equilibrium concept generates an infinity of equilibria all of which have the properties described in the text. Grossman and Perry use an additional criterion to define a perfect sequential equilibrium. For further information the technically inclined reader should consult Proposition 3.1 of Grossman, and Perry, , ‘Sequential Bargaining’Google Scholar, which provides the basis for the next paragraph.

²⁹ See Proposition 3.1 in Grossman, and Perry, , ‘Sequential Bargaining’.Google Scholar

³⁰ Since they are not proven formally, but are observations generated by their simulations. Both conjectures are cited from Grossman, and Perry, , ‘Sequential Bargaining’, p. 146.Google Scholar

³¹ For a test of this conclusion through a series of case studies, see Money, and Tsebelis, , ‘The Political Power of the French Senate’.Google Scholar

³² It would have been desirable to have data concerning bills which were not voted into laws broken down into similar sub-groups (bills aborted before, during or after the joint committee). Unfortunately, such data are not available.

³³ The major part of these allies belong to the party of Valéry Giscard d'Estaing. In 1958 he was a young but prominent leader of the Centre National des Independants et Paysans (CNIP). In subsequent elections Giscard became the leader of the Independent Republicans (RI), who were allied with the Gaullists. In 1978 the Independent Republicans merged with other centre-right parties and formed the ‘Union for French Democracy’ (UDF). There were other centre-right parties that joined the Gaullist majority over time. The party names change while leadership and political platforms are more stable, so we refer to all the parties that joined the Gaullist majority as ‘allies’.

³⁴ Obviously, this is a simplification because discount factors are also bill specific. The same National Assembly facing the same Senate may be in a big hurry to adopt one bill and indifferent about another; or one particular project may be very important for the Senate while another may not be. However, considering discount factors as a function of the composition of the two houses will give us some information about the average interaction between the two houses in the course of the legislative period.

³⁵ An alternative, and, in our opinion less convincing, hypothesis would be that intermediate levels of strength would not increase uncertainty about N's discount factor but produce an intermediate discount factor. Although we find the reasoning developed in the text more convincing, we test this hypothesis too. It produces expectations identical with the ‘linear composition models’.

³⁶ In other words, the Senate and the National Assembly agree on the legislation after one or two rounds of the navette, making government intervention and further negotiations unnecessary. We resort to this indicator because the data on French legislation do not indicate the number of rounds of the navette. Rather they indicate whether agreement was reached fast (in one or two rounds); whether the legislation was referred to a joint committee (two or three rounds); and ultimately whether the joint committee failed to reach an agreement and the government asked the National Assembly to make the final decision. While these distinctions do not replicate the actual number of rounds in the navette, they are an indicator of the length of the process ranging from few to many rounds. See Table 1 for actual data.

³⁷ The dependent variable in all the subsequent models is not dichotomous, but continuous (frequency of early agreement). The results reported subsequently are calculated by OLS. However, logit models produced similar findings.

³⁸ The minimum of a quadratic function y = ax² − bx + c is at the point x = b/2a.

³⁹ Depending on the estimated model this percentage is 67 per cent (calculated from the esti mated coefficients of Equation (3)) or 44 per cent (calculated from the estimated coefficients when (2) and (3) are merged into the same equation) or 52 per cent when one includes time in the equation (see p. 119). The maximum strength of the Left during the examined period was 40.5 per cent in 1978.

⁴⁰ Marichy, Jean-Pierre, La Deuxième Chambre dans la vie politique française depuis 1875 (Paris: R. Pichon et R. Durand-Auzias, 1969), p. 431.Google Scholar

⁴¹ Grangé, , ‘L'Efficacité normative’, p. 970.Google Scholar

⁴² Grangé, , ‘L'Efficacité normative’, p. 970.Google Scholar

⁴³ See also Marichy, , La Deuxième Chambre, p. 431.Google Scholar

⁴⁴ Georgel, Jacques, Le Sénat dans l'adversité (1962–1966) (Paris: Editions Cujas, 1968), p. 102.Google Scholar

⁴⁵ Also see Mastias, , Le Sénat de la V^e République, p. 49.Google Scholar

⁴⁶ Since the correlation between the age of the National Assembly and its composition is zero, the estimated coefficients of Equation (2) are unbiased even if one assumes that the fully specified model is Equation (3) or (4).

⁴⁷ Without 1978 the mean is 10.5 per cent, the standard deviation 0.063. It can be calculated that the point 32.2 per cent lies 3.44 standard deviations away from the mean, and, therefore, assuming a normal distribution, with probability more than 95 per cent, 1978 lies outside the area where at least 99 per cent of the points of the distribution are (see Walpole, Ronald E. and Myers, Raymond H., Probability and Statistics for Engineers and Scientists (New York: Macmillan, 1985), p. 223Google Scholar, on tolerance intervals).

⁴⁸ In 1978 the Parliament voted the annual budget and two additional financial laws; two laws modifying the penal procedure and penalties; a law reorganizing radio and television; a law regulating the relations between citizens and the bureaucracy; a law on financial reorganization of local government; laws on unemployment, employment of youth, businesses; laws concerning savings and investment; a law on rents; laws on ocean pollution, etc. (Jean Grange, personal communication).

⁴⁹ Lange, Peter, Ross, George and Vannicelli, Maurizio, Unions, Change and Crisis: French and Italian Union Strategy and the Political Economy, 1945–1980 (London: George Allen & Unwin, 1982), pp. 39 and 57.Google Scholar

⁵⁰ Jean Mastias, personal communication.

⁵¹ This is only one possible assumption. One could argue that discount factors decline with the importance of bills, because the two houses are under stronger pressure. However, we defend our assumption not only because it is reasonable (‘good’ results for each house are more important than expedience), but also because it is congruent with experiences in other countries. For example, in the United States, the joint committee is used more frequently for important bills (Krehbiel, Keith, Information and Legislative Organization (Stanford, Calif.: Graduate School of Business, Stanford University, 1990)).Google Scholar

⁵² A strict test would require the introduction of one additional variable (percentage of politically important bills) to our model.

⁵³ Tardan, , ‘Le Rôle législatif du Sénat’, p. 100.Google Scholar

⁵⁴ Huber, , ‘Restrictive Législative Procedures’.Google Scholar

⁵⁵ Maus, , ‘Le Sénat’, pp. 122–3.Google Scholar

⁵⁶ The French Parliament cannot sit in normal sessions more than six months a year. The extraordinary sessions cannot last more than two weeks each (Wright, Vincent, The Government and Politics of France (New York: Holmes and Meier, 1989)).Google Scholar

⁵⁷ Grangé, , ‘L'Efficacité normative du Sénat’.Google Scholar

⁵⁸ The technical dimension includes time constraints that lead the government to declare urgency and reduce the number of rounds and excess load of the lower house, which leads the government to introduce a bill in the upper house first.

⁵⁹ Grangé, , ‘L'Efficacité normative du Sénat’, p. 956.Google Scholar

⁶⁰ Money, and Tsebelis, , ‘Cicero's Puzzle’.Google Scholar

⁶¹ Money, and Tsebelis, , ‘Cicero's Puzzle’.Google Scholar

⁶² See Selten, Reinhard, ‘Reexamination of Perfectness Concept for Equilibrium Points in Extensive Games’, International Journal of Game Theory, 4 (1975), 25–55.CrossRefGoogle Scholar