Old Wine in A New Bottle: MacRae s Assemblé Nationale Reconsidered

Old Wine in A New Bottle: MacRae s Assemblé Nationale Reconsidered Howard Rosenthal * and Erik Voeten ** Prepared for presentation at the annual meeting of the American Political Science Association, San Francisco, California, August, 2001. * Princeton University, rosentha@princeton.edu ** George Washington University and Stanford University, evoeten@princeton.edu This work was supported by National Science Foundation grant #973053. Any opinions, findings, and conclusions or recommendations in this work are those of the authors and do not necessarily reflect the views of the National Science Foundation (NSF). 1

Introduction This essay joins two threads. One is methodological. We argue that Poole s (2000) new optimal classification method for roll call analysis is likely to be preferable to older methods for studying legislatures other than the United States Congress. The example that supports the argument is an analysis of French National Assembly voting for 1946-58. We use the example for the other essay s other objective, which is substantive. We find that voting in the National Assembly was highly structured along ideological lines, in fact much more so than in Congress. Instability in the political system did not take the form of chaotic, non-spatial voting as occurred in the United States in the run up to the Civil War (Poole and Rosenthal, 1997). On the other hand, voting in the three legislatures of the Fourth Republic is best described by at least two spatial dimensions. One can readily be thought of as a classic economic left-right dimension. When a second dimension has arisen in the United States Congress, it also has a policy interpretation race. In contrast, in France the second dimension allowed parties of different economic ideologies to vote together in opposition to Fourth Republic governments. Indeed, particularly in the third, terminal legislature, a second dimension is largely unnecessary when one amends the basic spatial model to allow for both ends against the middle voting in a one-dimensional model. As events unfolded in the Fourth Republic, the roll calls that featured both ends against the middle voting increasingly concerned policy in North Africa. Thus, the demise of the French Fourth Republic is likely to have been in the interaction between ideological voting in a pure parliamentary system and the strong external threat posed by the colonial war in Algeria. 2

Methodological Issues Modern roll call analysis has undergone an increasingly sophisticated methodological development [Poole and Rosenthal (1985, 1991, 1997), Heckman and Snyder (1997), Snyder and Groseclose (2000), Londregan (2000), Poole (2000), and McCarty, Poole, and Rosenthal (2001)]. Almost all of the applications of the methods to legislative floor voting, however, have been to the United States Congress. 1 The data in this paper, in contrast, come from the three legislatures of the French National Assembly in the short-lived and not lamented Fourth Republic (1946-58). The Assembly was in many ways the antithesis of Congress. A large fraction of the deputies were members of a totally disciplined party that voted as a bloc, the Communists. As we move to the right in the political spectrum, we find the Socialists, only somewhat less disciplined than the Communists. Even the Christian Democrats (MRP) tended to vote as a bloc. The other members of the lower house, largely in the political center and right, were much more free wheeling in their behavior. Changes in party affiliation were common (Riker, 1962). Both discipline, making for more predictability, and vote trading, making for less, may have been influenced by the presence of proxy voting on most roll calls. Consequently, the French data is likely to be inappropriate for parametric and stochastic models like the NOMINATE methods of Poole and Rosenthal or the Heckman-Snyder method. These methods assume errors that are independent and identically distributed across both legislators and roll calls. These assumptions about errors, doubtful enough for Congress, are almost surely wildly inappropriate for postwar France. Moreover, identification of the parametric models relies heavily on the errors 3

being relatively substantial. 2 Applying the models to a strongly ideological, small error legislature will be problematical. The non-parametric optimal classification method of Poole (2000) might prove to be a more robust technology. 3 We here report that, in terms of face validity, the non-parametric method is much preferred. The methodological report is very much a tribute to Duncan MacRae, Jr., who produced seminal research in roll call analysis. We use the data he assembled for his award-winning, Parliament, Parties, and Society in France 1946-1958 (1967). MacRae, working at a time when the best university computers had a fraction of the power of today s bargain basement desktops, understandably chose to limit his analysis to the development of ordinal scales for four or five major party groupings in each of the three legislatures (MacRae, 1967, pp. 65-180). The voting patterns across, as against within, these groups were treated only informally. 4 Large numbers of deputies, including all Communists and Poujadists, were not included. MacRae s study inevitably emphasized internal party divisions rather than divisions between parties. 5 Internal party divisions do not necessarily indicate political instability. Internal party splits can occur with a stable, one-dimensional left-right system 1 For exceptions, see Voeten (2000), and the studies referenced in Poole and Rosenthal (2001). 2 When the error or noise level is small, the linear probability assumptions of the Heckman-Snyder method will be violated. As for NOMINATE, it will seek to maximize the likelihood for a perfectly classified legislator who is extremist by putting the legislature far from the center of the space. The method therefore constrains the position of such legislators to lie on the boundary of the unit hypersphere (circle in two dimensions) that defines the limits of the space. This will lead to a largely circular or spherical configuration of legislators. 3 To conserve space, we refer the reader to the published literature for description of the methods. 4 MacRae was previously criticized by Wood (1964, 1973) for considering only intra-party divisions. Wood himself, however, failed to provide an analysis of an entire legislature. Wood (1964) considers only 57 votes on European unification across all three legislatures of the Fourth Republic. Wood (1973) does consider all the L Année Politique votes in MacRae s data for the Second Legislature but then scales the roll calls in seven separate analyses corresponding to the seven cabinets of this legislature. 5 Two of the groups of deputies MacRae analyzed, the Radicals and the Right, were both amalgamations of two or more parliamentary groups. Consequently, it might be more appropriate to say that MacRae looked at internal divisions of parts of the ideological spectrum. 4

as in the contemporary two-party United States (Poole and Rosenthal, 2000) or in the contemporary roughly eight-party Czech Republic (Noury and Mielcova, 1997). A spatial analysis of the entire legislature is required to see if roll call voting can fit into a low-dimensional space. Our analysis provides snapshots of the legislatures. Since our worst classification in two dimensions for any of MacRae s party groups is 94% correct (Radicals in the third legislature), we claim that our spatial maps or snapshots capture most of the internal party divisions summarized in MacRae s more detailed analyses. Political Instability We use these maps to make our substantive points. These concern the stability of democracy and, in particular, the demise of the Fourth Republic. The Republic was notorious for instability. After running through 21 premiers and cabinets in just 12 years, the Republic was cashiered not by a democratic vote by the electorate but by the revolt of military officers and pieds noirs settlers in Algeria. In contrast, the United States has had largely democratic and regular transitions, albeit punctuated by a civil war, for over two centuries. Let us be clear about the concepts of stability we employ. One form of instability pertains to policy (Baumgartner and Jones, 1991). Policy instability, as the successive postwar nationalizations and privatizations of the British steel industry demonstrates, is not indicative of other forms of instability. This paper is not concerned with policy instability. Governmental instability might be defined as the frequency of change of either the head of government or the governing coalition. This type of instability was characteristic of the Fourth Republic. On the other hand, prolonged governmental 5

stability contradicts the notion that stable democracies have peaceful transitions in power. The election of Franklin Delano Roosevelt to four presidential terms triggered the adoption of the XX amendment in the United States. Incumbency bias in legislative elections triggered the term limits movement. True governmental stability in democracies might be associated with some long run rate of governmental duration that is far less than infinite. The consensus view would appear to be that the short durations of Fourth Republic governments were less than optimal. We do investigate whether the structure of roll call voting and strategic behavior of legislators contributed to governmental (cabinet) instability. Regime instability can be defined by the duration or survival of a form of government (Gurr 1974). Historically, regime instability followed the French Revolution. Noteworthy among the regime collapses, those of 1812, 1815, 1871, 1940, and 1942-1944 all took place when the incumbent regime suffered military defeat while that of 1958 occurred in the midst of a protracted and unsuccessful war using conscripts in Algeria. In particular, in the two democratic collapses 1940 and 1958 legislative activity could have been an indirect cause of collapse. It is also noteworthy that, while prone to cabinet instability, the Third Republic was the most durable regime. It spanned a 70-year period that included France s escaping defeat in World War I. Comparative studies have shown that genuinely effective and responsive legislatures enhance the stability of democratic regimes, but when societal cleavages are severe and complementary democratic institutions are lacking, legislatures may exacerbate political disorder (Mishler and Hildreth, 1984).We are concerned with how legislative activity related to the collapse of the Fourth Republic. 6

In addition to institutional features,ideologies are an important source for stability and instability in legislative activity. The spatial model of voting (Enelow and Hinich, 1984) is the standard approach to analyze legislative behavior. For the Euclidean preferences assumed by the spatial model, the Plott-McKelvey-Schofield theorems tell us that, once there is more than one dimension of preference, majority voting is generically unstable. The theorems led Riker (1980) to refer to political science as the dismal science since political processes would lack equilibria and be unstable. Consequently, the dimensionality of roll call voting might be a major marker for instability. In this paper, we compare the dimensionality and fit of spatial models for the Fourth Republic s National Assembly to similar measures for the United States Congress in the same historical period. In the United States, congressional roll calls have been studied intensively [See, for example, MacRae (1958); Poole and Rosenthal (1997)]. We find that the National Assembly was closer to a perfect model of spatial voting than was or is the United States Congress. On the other hand, at least a second dimension is required to obtain the near-perfect fit to the data. An important second dimension in the United States also existed, in the presence of the civil rights dimension in the 1940s, 50s, and 60s. This dimension is not directly associated with instability in the American political system. The American system, it can be argued, only fell apart once, in the Civil War. This was preceded by a period of chaotic, non-spatial voting following the collapse of the Compromise of 1850 and the emergence of a dimensional realignment (Poole and Rosenthal, 1997). The National Assembly did not provide an indication of chaotic voting. 7

In addition to looking for possible instability in the overall pattern of roll call voting, we study instability from the perspective of strategic behavior both of important members of the legislatures and on important roll call votes. The experience of the Fourth Republic is particularly fascinating to political scientists interested in the strategic incentives that institutions create for legislators. Under the rules of the Republic, a cabinet that lacked sufficient (majority) support in the Assembly would resign. This would typically not lead to new general elections, but to the formation of a new cabinet from within the Assembly. Thus, legislators who were not part of the current cabinet had the opportunity to obtain ministerial positions when a cabinet would lose majority support. Moreover, if they voted not to support a cabinet, this had no personal implications in that they did not face the immediate prospect of new elections. This institutional setting created clear incentives for enterprising legislators to undermine a cabinet order to satisfy personal ambitions. Those with the prestige or power to obtain a cabinet position were known as the ministrables. One group of scholars point to opportunistic behavior by ministrables as the main source of cabinet instability (Wood, 1973). Others point to more substantive and ideological differences between parties and legislators (MacRae, 1967). In this view, cabinet instability stems mostly from variations in opinions on central issues such as the European Defense Community, Indochina and North Africa. These external events would determine whether legislators defected from the governing coalition. In examining the behavior of ministrables, we find largely in favor of the ideological hypothesis against the opportunistic hypothesis. 8

Of course, the ministrables might look unexceptional if they succeeded in persuading a number of their supporters, those who were not ministrables, to joint in strategic manipulation. Such manipulation might be most expected to occur on the handful of votes that represent votes of confidence (for sitting prime ministers) and investiture (for candidate prime ministers). We therefore analyzed these votes separately. Confidence and investiture votes, we find, are at least as ideological as other roll call votes in the Assembly. Most governments rose and fell, therefore, because of changes in allegiance of deputies that largely respected the ideological lineup of the Assembly. Outof-character shifts were quite rare. As with our findings on ministrables, our analysis of confidence and investiture votes does not turn up indications of instability in spatial voting. On the other hand, we also find that opposition to the Fourth Republic regime in itself always constituted one of the two primary dimensions of the space. Consequently, it might be claimed that ideology per se was an important source of instability. Ideology, in the technical sense of being a systematic pattern of spatial voting, spelled instability because opposition to the regime was deep and systematic. This is most dramatically evidenced in the regime, rather than left-right, constituting the primary dimension in the first legislature. The regime was not sabotaged because of tricky, outof-character, voting on a few roll calls. Au contraire, the regime was constantly under attack. Consequently, both conflicst over institutions, particularly the electoral law, and over policy, particularly Algeria, fit into a more general pattern of pro-regime, antiregime voting. Of course, there are many other explanations for instability: 9

?? Those that go under the rubric of culture, national character, etc., can be quickly disregarded since France has been stable since 1958 under a democratic regime that enjoys a high degree of legitimacy. Despite De Gaulle s famous crack about governing a people that makes hundreds of cheeses, France is not inherently unstable.?? Another possible explanation would be institutional factors in addition to the rules, discussed above, that directly affect cabinet duration. In contrast to the United States, which has a president, two coequal chambers in a bicameral legislature, and an independent judiciary, that is, separation of powers and checks and balances, the Fourth Republic was very much a pure parliamentary government dominated by the legislature. Moreover, whereas in the United States the constitution fixes elections at regular intervals and Congress long ago imposed the single-member district system, the electoral law in France is entirely endogenous. The Fourth Republic legislature devoted substantial energy, in 1951, to imposing the complex apparentement system (Lee, McKelvey, and Rosenthal, 1980) on top of a proportional representation system, and, in 1956, to failing to revise the law. Roll calls on the electoral law in fact accounted for much of the both ends against the middle voting in the first two legislatures. In contrast, the Fifth Republic has less power in the legislature (Assembly), and more power in an elected president and the cabinet or Conseil des Ministres. A single-member district electoral system has been relatively stable. 6 The contrast between the Fourth and Fifth Republics suggests that institutions matter. That institutions did it, however, can be questioned by comparing France to Italy. Postwar Italy had similar institutions and similar internal divisions to those of France, 6 Proportional representation was used in one legislative election, 1993, when the Socialists tampered with the system.. There is also some degree of malapportionment. 10

but the Italian system, despite similar cabinet instability, was always one of democratic transitions.?? Where France and Italy differed, of course, is that Italy did not have to deal with colonial revolts after the war. Indeed, the main charge against the Fourth Republic governments had been their impotence ( immobilisme ) to act in foreign and military affairs, in particular in the Algerian crisis. Thus, external threat may have been the source of instability. In turn, our two major findings highly spatial, ideological voting and the absence of clear strategic behavior by ministrables allow us to suggest that instability resulted from external threats that, given ideological differences, could not be dealt with by a strong majority government. Summary On balance, our reanalysis of the Fourth Republic data suggests that the Republic s problems were more ones of irreconcilable ideologies than of excessive concern of the deputies with their private parliamentary game and their careers, without sufficient care about principles or policy (MacRae, 1964, 4). The main piece of evidence here is the extremely high fit of one and two-dimensional spatial models of roll call voting. A huge chunk of irreconcilable ideology was contributed by the Communists, who represented about one-fourth of the electorate and the Parliament. The external event here was the Cold War. The main objective of this Soviet-financed party was to take France behind the Iron Curtain. Governments and policies would therefore have to be formed by a supermajority of the remaining deputies. 11

In the course of the first legislature, De Gaulle decided to reenter politics and challenge the regime. The electoral success in the 1951 elections of his RPF party resulted in two-dimensional voting in the second legislature. The Communists and Gaullists would often vote together to topple governments and block policies such as the European Defense Community. This second dimension resulted from the political opportunism and strategy of the Gaullists, not the career concerns of the regime parties. The Gaullists also exploited another aspect lay vs. church of ideology in France but conflict here was subservient to the Gaullists political ambitions rather than the direct cause of conflict. The Gaullists were defeated. They were unable to use the career concerns of the regime deputies to bring about the fall of the Fourth Republic during the second legislature. De Gaulle withdrew from politics; his deputies largely were absorbed into the regime. He returned to French politics when the third legislature was terminated by another external event, the Algerian war. The elections to the third legislature produced a new challenge to the regime, the Poujadists. These anti-tax, populist deputies (Jean-Marie Le Pen was one of them) joined the Communists in opposing the regime. The Poujadists and remaining Gaullists are largely responsible for a second dimension in the last legislature of the Republic. On balance, then, elections, not internal games, are the cause of important shifts in voting patterns within the Assembly. Within each legislature, voting is highly ideological (spatial). Of course, the inability of the legislature to make policy (immobilisme) contributed to the electoral outcomes. 12

Methods We carry out separate spatial analyses of the three legislatures of the Fourth Republic. We compare the W-NOMINATE method of Poole and Rosenthal (1997) to the optimal classification algorithm of Poole (2000). For data from the first 100 United States Congresses, the W-NOMINATE results are highly similar to the results from the Heckman-Snyder method (Poole and Rosenthal, 1997). Thus, we take W-NOMINATE as illustrative of a parametric method that, as said in the introduction, assumes that the errors in the spatial utility functions are i.i.d. In contrast, the optimal classification method simply seeks to find ideal points for legislators and separating hyperplanes for roll calls such that the number of classification errors is minimized. A classification error for a legislator on a roll call occurs when the legislator s ideal point is such that his or her vote is inconsistent with the separating hyperplane for the roll call. This procedure is likely to be highly robust to the stochastic nature of the data. Note that for optimal classification all errors are weighted equally. No single vote decision is likely to make a large difference in an estimate. In contrast, the parametric methods will adjust estimates based on the most serious errors, such as a Communist voting in favor of the European Defense Community. There is, of course, a price to be paid for using the non-parametric method. Metric information about ideal point locations and separating hyperplanes can not be identified. In one dimension, this problem is a serious one. Only a rank order of legislator ideal points (and roll call cutting points) can be recovered. For the United States, the metric recovery generated by W-NOMINATE appears to have face validity in that it shows a bimodal distribution of ideal points corresponding to the two-party system. 13

But, as we show later, in two dimensions W-NOMINATE does not have face validity for the French data. In contrast, optimal classification does generate a reasonable map. How does optimal classification recover metric information in more than one dimension? The two dimensional case illustrates the basic answer. If there is not much error to voting and we have many unique separating hyperplanes from the roll calls, each legislator s ideal point will typically be pinned down to a small cell generated by the intersections of hyperplanes. This cell does not generate a precise point estimate (Poole s method picks a point in the cell), but if the cells are small, the imprecision is small. Moreover, as Londregan (2000) has shown, any roll call method, including a parametric one, will suffer from this imprecision because of the coarseness or granularity of roll call data. The coarseness of the data appears to be much less problematic for the nonparametric method than for parametric methods. On the basis of our experience with the Fourth Republic data, we advocate the Poole non-parametric method as the preferred method for roll call analysis in large, multi-dimensional legislatures. Data The sample of roll calls consists of two parts [see MacRae (1964, Appendix A) for more detail]: 1. All 739 roll calls mentioned in L Année Politique are included. Sampling in this way is roughly similar to constructing a sample for Congress using only key votes reported in Congressional Quarterly. L Année Politique includes important votes, such as investitures, confidence votes, important budget votes, and other policy votes. 2. A random sample of 50 votes per legislature was taken from the remainder of all roll call votes. 14

The total of 889 roll calls in MacRae s sample are a little over one-tenth of the total of 8550 roll calls taken during the Fourth Republic. The full sample include votes on topics such as NATO passage, Algeria, Indo-China, Italy treaty, Paris accords, military budget, atomic energy development, labor strike regulations, taxes, public health, Parisian transportation, colliery schools, National Assembly reform, agriculture and the economy, land reform, fiscal reform, anti-inflation, salary scale, and amnesty. We will report on results for the random sample, to check if our results are strongly influenced by the entire sample being heavily weighted by the L Année Politique votes. Following Poole and Rosenthal (1997), we exclude all roll calls with fewer than 2.5% of those voting on the minority side. This reduces the sample from 889 to 855 votes. We also exclude all legislators with fewer than 25 votes. Abstentions of any form are disregarded. 7 One overall characteristic of the roll calls is strikingly similar to that for the United State Congress throughout its history the average majority is a little under two-thirds. The data is summarized in Table 1. 7 MacRae, in contrast, included abstentions. On the basis of pre-scaling adjustments, he would often treat abstentions as equivalent to No or Yes votes. The analysis of abstentions, including strategic abstentions, is an important topic for future research. 15

Table 1. Summary Statistics 1 st 2 nd 3 rd Total Valid Votes 191036 195719 80705 Minority Votes 71943 69630 26959 Average Majority % 62.3% 64.4% 66.6% Legislators a 645 647 599 Roll Calls b 346 346 163 a Total number of legislators voting at least 25 times. b Total number of roll calls with at least 2,5% on the minority side The data contains coding for the political party (technically, parliamentary group) of each deputy. In our analysis, we combine, in a straightforward manner, some of these codes. The details appear in the Appendix. Results Fit to Spatial Models The overall fit of the spatial models is presented in Table 2. This shows results for estimated one, two, and three-dimensional models. Very small additions to fit were provided by considering four or more dimensions. Indeed, to two decimal places, with three dimensions we correctly classify 98% of the individual voting decisions in each of the three legislatures. Any addition beyond this point is likely to be simply an over fitting of noise in the data. As summary statistics, we present, for both types of models, the classification percentages, and the Average Proportion Reductions in Error (classification gains relative 16

to the marginals). For the W-NOMINATE model we also present the geometric mean probability of the observed choices (GMP). 8 Table 2. Fit Statistics For W-Nominate and Optimal Classification 1 st Legislature 2 nd Legislature 3 rd Legislature W-Nom Optimal W-Nom Optimal W-Nom Optimal 1 Dimension Class. % 93.2 95.1 89.0 91.6 92.3 93.7 APRE 81.7 86.8 69.2 76.4 77.0 81.0 GMP.849 -.784 -.835-2 Dimensions Class. % 94.0 97.6 92.9 95.7 95.7 97.6 APRE 83.9 93.7 80.0 88.2 87.2 92.8 GMP.895 -.844 -.890-3 Dimensions Class. % 95.9 98.1 95.3 97.5 96.6 98.4 APRE 89.0 94.9 86.7 92.9 89.7 95.3 GMP.910 -.873 -.905 - APRE2-APRE1 a 2.2 2.6 10.8 11.8 10.2 11.8 APRE3-APRE2 b 5.1 1.2 6.7 4.7 2.5 2.5 a Difference between APRE in two-dimensional model and one-dimensional model. b Difference between APRE in three-dimensional model and two-dimensional model. For reasons to be made clear shortly, we focus on the Optimal Classification results. The first legislature looks to be predominantly one-dimensional. It has the highest one-dimensional classification of the three legislatures, the smallest gain for a second dimension, and a truly negligible gain for a third. Both the second and third legislatures appear to have an important second dimension. A third dimension may have had some significant role in the second legislature; the third dimension captured internal 8 These measures are explained in detail in Poole and Rosenthal (1997, pp. 30-31). 17

divisions of the MRP, Gaullists, and Radicals, both on colonial policy and institutions.. At any dimensionality, however, the classifications are exceptionally strong by comparison to the United States. The one-dimensional fits are as strong or stronger than any found in United States history, including the sharply ideologically divided Congresses at the ends of the nineteenth and twentieth century. In the United States, when a second dimension has been important, it has served only to bring overall classification up to that found with one dimension. More precisely, in the first 100 Congresses, the average two-dimensional D-NOMINATE classifications for the House of Representatives are only 85%, the average APREs only.56, and the average GMPs.71. 9 Between 1945 and 1958,, classifications in two dimensions were always below 90%. 10 In contrast, adding second and third dimensions in France essentially explains what is left over from the one-dimensional model and results in a near perfect fit. One might think that our results largely reflect an ability to correctly classify the Communists and other parties exhibiting discipline. That is not the case. Table 3 presents average classifications (each deputy is weighted equally, regardless of number of votes actually cast) by party. Classification is high for all the parties, exceeding 90% for all parties in the first legislature in only one dimension, and 91% for all parties in two dimensions in all three legislatures. What is also immediately clear from Table 3 is that the second dimension is most important to classification of the two non-communist groups, Gaullists and Poujadists, that most strongly opposed the regime. The second dimension appears largely to be a matter of allowing left and right to vote against the 9 Poole and Rosenthal (1997, p. 28). We use the results for the cubic specification, which is the specification that is most similar to Congress by Congress W-NOMINATE scalings. 10 Poole and Rosenthal (1997, p.32). Poole (2000) reports that optimal classification for the Senate for the period of our study is about 85% in one dimension with another 5% added by a second dimension. 18

middle. This dimension was not necessary in the first legislature, where, as we shall see shortly, the first dimension directly incorporates anti-regime voting. The classifications in table 3 are impressive even when one takes into account party cohesion or discipline. This is especially true for the Radicals, who voted together only 86% of the time in th1e first and third legislatures, and 90% in the second. The spatial model also captures the internal splits of the Right and Gaullists, who voted together about 90% of the time. The other parties were more cohesive, so there was limited room for improvement. The spatial model, however, has a much tighter theoretical focus than does a party cohesion model. Party cohesion says nothing about which way a party will vote on a particular roll call. The spatial model says that party voting must obey a consistent relationship to how other parties vote, the relationship being defined by the location of the ideal points of the party members. For example, in a one-dimensional model for the second legislature, one could not have the Socialists and Christian Democrats voting against all other parties. 19

Table3. Percentage Correct by Party For One- and Two-dimensional Optimal Classification. a 1 st Legislature 2 nd Legislature 3 rd Legislature Party 1 dim 2 dim 1 dim 2 dim 1 dim 2 dim Communist 99.9 99.0 95.3 99.9 100 100 Socialist 93.1 97.8 99.2 99.8 98.9 99.5 Radical 90.8 94.6 89.0 93.7 90.8 93.6 Right 90.2 96.4 91.4 93.9 91.9 96.0 Christian Democrat 95.5 97.8 91.2 93.7 97.3 97.4 Gaullist - - 83.8 93.8 84.2 91.1 Poujadist - - - - 69.9 99.2 a All entries are party means Preliminary Interpretation of the Dimensions Interpretation of the dimensions is facilitated by the scatterplots of legislator positions shown in figure 1-3. We will subsequently use alternative presentations of the data since the scatterplots obscure the distributions of party ideal points as a result of the overstriking that occurs with more than 600 legislators. The coordinates that are produced by Poole s optimal classification program have all been scaled by a constant such that the first dimension standard deviations of the W-NOMINATE and Optimal Classification coordinates are equal. We focus first on the Optimal Classification plots. In all three legislatures, opposition to the government or regime generates an important dimension. In the first legislature, which has the highest one-dimensional classification and APRE of the three legislatures, the first dimension is the regime dimension. Surprisingly, the first dimension is not a classic left-right dimension. The 20

order of party means on the dimension (see below) is Communist-Right-Radical- Socialist-Christian Democrat. The relatively left (anti-regime) placement of the Right is consistent with the Right never having held the premiership during the first legislature. In the last two-legislatures, the second dimension, as indicated by our analysis of gains to classification by party, was the regime dimension. The other dimension (second in the first legislature and first in the last two legislature) is a classic left-right dimension for the non-communist parties. The order of the parties is Socialist-Communist-Christian Democrat-Radical-Right in the first legislature, Communist-Socialist-Gaullist-Radical- Christian Democrat-Right in the second, and Communist-Socialist-Radical-Poujadist- Gaullist-Christian Democrat-Right in the third. Note that, while the Poujadists anchor the anti-regime pole on the second dimension in the third legislature, they are relatively left on the economic dimension, consistent with the Populist aspects of their ideology. 21

Figure 1. Legislator Coordinates, First Legislature W-NOMINATE 1.0.5 Left-Right 0.0 party Non-Affiliated Other -.5 Right Christian Democrat Radical Socialist -1.0 Communist -1.0 -.5 0.0.5 1.0 Anti-Pro Regime Optimal Classification 2.5 1.9 1.3 Left-Right.7.1 party Non-Affiliated Other -.5 Right Christian Democrat -1.1 Radical Socialist -1.7 Communist -1.1 -.7 -.3.1.5.9 1.3 Anti-Pro Regime 22

Figure 2. Legislator Coordinates, Second Legislature W-NOMINATE 1.0.5 party Anti-Pro Regime 0.0 Non-Affiliated Other Gaullist -.5 Right Christian Democrat Radical Socialist -1.0 Communist -1.0 -.5 0.0.5 1.0 Left-Right Optimal Classification.8.6.4.2 party Anti-Pro Regime 0.0 -.2 Non-Affiliated Other Gaullist -.4 Right Christian Democrat -.6 Radical Socialist -.8 Communist -1.6-1.2 -.8 -.4 0.0.4.8 1.2 1.6 Left-Right 23

Figure 3. Legislator Coordinates, Third Legislature W-NOMINATE 1.0.5 party Non-Affiliated Anti-Pro Regime 0.0 Other Poujadist Gaullist -.5 Right Christian Democrat Radical Socialist -1.0 Communist -1.0 -.5 0.0.5 1.0 Left-Right Optimal Classification 1.2.7 party Anti-Pro Regime.2 -.3 Non-Affiliated Other Poujadist Gaullist Right -.8 Christian Democrat Radical Socialist -1.3 Communist -1.4-1.0 -.6 -.2.2.6 1.0 Left-Right 24

Comparison of the Two Methods Visual comparison of the Optimal Classification results with the W-NOMINATE results in figures 1-3 discloses two important differences. First, whereas the Optimal Classification coordinates show a relatively smooth distribution throughout the space, the W-NOMINATE coordinates are concentrated at or near the unit circle that defines the boundary of the space for the W-NOMINATE algorithm. In fact, the percentage of legislators placed on the rim of the space is 41.9, 36.0, and 48.7 in the first, second, and third legislature, respectively. 11 This is largely because classification is high and there is little error to voting. In order to maximize the log-likelihood, W-NOMINATE will place a perfect (errorless) legislator as far from the roll call cutting lines as possible. Such a circular distribution of ideal points seems implausible. Second, particularly in the second and third legislatures, there is an enormous gap between the Communists and all other parties in the W-NOMINATE results, less so in Optimal Classification. This largely reflects the Communists voting in a disciplined and errorless way. 12 W-NOMINATE captures the perfectness of the Communists by making them very distant from everyone else. The gap seems implausible, particularly on the left-right or economic dimension where the Socialists, particularly in the 1940s and 50s, shared goals of redistribution and public ownership with the Communists. 11 More precisely, these are the percentages of legislators for whom the sum of squares of the coordinates exceeded 0.99. A legislator exactly on the rim would have a sum of squares of 1.0. We use.99 as the cutoff, because the coordinates are only precise up to two decimals. 12 The fact that the Communists do not all have exactly the same ideal point reflects variations in abstention. Each deputy s position is determined by placing the deputy relative to the cutting lines for the roll calls on which he or she in fact voted. 25

While Optimal Classification thus appears to have greater external validity than W-NOMINATE, the two methods both capture the main features of the data. If we can interpret optimal classification results as metric, we can look at Pearson correlations with W-NOMINATE coordinates. In Table 4, we report Pearson correlation coefficients for two-dimensional models. We report Spearman correlations in brackets. The correlations are quite high. Table 4. Comparison of W-NOMINATE and Optimal Classification Pearson (Spearman) Correlation Legislature 1 st dimension 2 nd dimension 1 st (1946-51).984 (.926).887 (.905) 2 nd (1951-55).895 (.956).931 (.938) 3 rd (1956-58).978 (.915).883 (.910) Party Positions on the Dimensions An analysis of party means and standard deviations on the dimensions, presented in Table 5 provides another, more precise view of the results seen in the scatterplots for the Optimal Classification results. For the first legislature, the first dimension basically separates the Communists, with the only negative, left mean from everyone else. The gap between the Communists and the closest of the four non-communist groups is 1.25, while the total span of these four groups is only 0.26 What seems anomalous is that the Right is the most left of the four anti-communist party groups. But the result is understandable in terms of regime support. The governments of the first legislature where basically center-left. All the prime ministers were Socialists, Radicals, or Christian Democrats, and a large chunk of 26

Right deputies always voted against investiture of these governments. The Right also, by shifting from abstention to opposition, as between Ramadier s investiture vote in Jan. 1947 and the vote to reorganize his cabinet in October (MacRae, 1967, p. 72), showed only limited support for the regime. Similarly, after voting for Schuman s investiture in November, 1947, the Right either abstained or voted against on all other major votes during Schuman s premiership (MacRae, 1967, p. 74). The Right also joined the Communists on votes related to finance, the budget, and taxes. The Right was objecting to government intervention in the economy and redistribution; the Communists were systematically attempting to undermine the government. The second dimension in the first legislature corresponds to a standard left-right order of the non-communist parties. These parties are now differentiated. The Right opposes all the other parties on some economic issues and on the release of Petain. The only anomaly is the Socialists being to the left of the Communists. There are only three roll calls where the Socialists oppose all other parties, including the Communists. This observation suggests that we should not lose sight of the basic finding that, for the first legislature, the second dimension is weak and unimportant. The basic conflict concerned the regime itself. On both dimensions, we obtain the expected result that the Communists and Socialists are more homogeneous than the other parties. In the second legislature, the first dimension resembles a standard left-right alignment. Note that the Gaullists, in line with their relative leftism on social and economic policy as against Communism and foreign and military policy locate close to the Radicals. They are still sharply different from the Socialists and, a fortiori, the 27

Communists. A majority of Gaullists nonetheless vote with the Socialists and Communists to bring down the Pinay government over the budget in December, 1952 and the Mayer government over tax reform in May, 1953; Pinay and Mayer had majority support from all the other parties (MacRae, 1967, pp. 119, 121). The Right is as expected, furthest to the right. In contrast to the first dimension, on the second dimension, the Gaullists and Communists have nearly identical positions opposition to the regime. The Socialists and Radicals are furthest to the right. They vote en bloc, with some Christian Democrat support, for Mèndes-France s failed investiture attempt in June, 1953. Similarly, the Socialists, Radicals, and Christian Democrats are the only parties to support Pineau s failed investiture attempt in February, 1955 (MacRae, 1967, p. 130). It should be noted that the lay-catholic conflict does not appear to generate a separate dimension. Votes on both the Marie education bill and the Barangé bill in September, 1951, have 45 o cutting lines in the space. As can seen by inspection of Figure 2, such cutting lines allow a fraction of the Radicals, the Socialists, and the Communists, to oppose all other deputies. If education was a wedge issue (for the Gaullists), the wedge was simply to pick an issue whose cutting line was interior to the governing coalition. The second legislature marked a sharp shift to the right. (See the party Ns in table 5B.) The prime ministers were either from the Right or the Radicals. Nonetheless, the Socialists and Christian Democrats continued to support the regime. Socialist support was forthcoming on all but one investiture vote except Mayer s; similarly, the Christian Democrats supported all prime ministers other than Mèndes-France. In contrast, the 28

Gaullists supported only Mayer and the Communists only Mèndes. The Right only voted against Pineau s investiture, but a majority of the Right voted against the government on 11 of the 51 votes of confidence during the second legislature. Variations in party homogeneity in the second legislature were similar to those in the first. The new party, the Gaullists, was dispersed, reflecting their breakup during the legislature. (Recall that the party codes reflect the code applied by MacRae to each deputy s first vote.) The first dimension in the third legislature looks similar to that of the second. The ordering on the first dimension is the same. It is noteworthy, however, that, in their quantitative placement, the Socialists have moved to the right, perhaps because of the need to support some policies during the premiership of Guy Mollet, a Socialist. Also, only some of the Socialists join the Communists in refusing to invest De Gaulle in June, 1958 (MacRae, 1967, p. 163). The rump Gaullists have also moved to the right. The Poujadists are not as far to the right as the classical Right, perhaps because of their support for some populist economic measures. The Poujadists do join the Communists in opposition to the regime, forming a pole on the second dimension. The Poujadists distinguish themselves from the Communists by, for example, being the only party to solidly oppose Mollet s Algerian policy in March, 1956 (MacRae, 1967, p. 159). Once again, the Right shows itself as not fully supportive of the regime. They, for example, join the Communists and Poujadists in voting to bring down the Mollet government over a Finance bill in May, 1957 (MacRae, 1967, p. 159) and in voting not to invest Mollet in October, 1957. (MacRae, 1967, p. 163). A majority of the Right voted against the government on 17 out of the 44 votes of confidence in the third legislature. Many of these 29

votes came on socio-economic issues, in particular the Old Age Fund, the issue of paid vacations and the civil budget. On other issues, including the Algerian question, they were mostly loyal to the government. Table 5. Party Means and Standard Deviations, Optimal Classification Scaling. A. First Legislature 1 st Dimension 2 nd Dimension Party Mean Std. Dev. Mean Std. Dev. Communist (N=177) -.99.08 -.44.12 Socialist (N=108).43.14 -.79.19 Radical (N=79).37.30.61.46 Christian Democrat (N=169).52.21.07.37 Right (N=77).26.24 1.18.45 B. Second Legislature 1 st Dimension 2 nd Dimension Party Mean Std. Dev. Mean Std. Dev. Communist (N=98) -1.01.06 -.32.02 Socialist (N=108) -.65.03.37.03 Radical (N=107).33.21.39.18 Christian Democrat (N=92).38.07.05.05 Right (N=116).59.21 -.03.18 Gaullists (N=117).26.14 -.33.13 C. Third Legislature 1 st Dimension 2 nd Dimension Party Mean Std. Dev. Mean Std. Dev. Communist (N=144) -1.19.11 -.41.06 Socialist (N=93).01.10.66.13 Radical (N=93).14.23.23.29 Christian Democrat (N=73).30.02.26.03 Right (N=107).54.14 -.22.22 Gaullist (N=21).28.13.00.45 Poujadist (N=45).20.15-1.01.13 Stability of the Legislator Positions Table 6 gives the correlation coefficients between the ideal point estimates of legislators that served in more than one legislature. There are 234 legislators for whom we have ideal point estimates in all three legislatures, 358 legislators for whom we have 30

estimates in both the first and the second legislatures, and 340 in both the second and third legislatures. The correlations between the ideal point estimates measure the extent to which the ideological positions of these legislators have been stable over time. Table 6. Correlation Coefficients between Coordinates from Two-Dimensional Optimal Classification in Different Legislatures (Spearman rho above diagonal, Pearson R below diagonal) Dim 1 1 st Leg Dim 1 2 nd Leg Dim 1 3 rd Leg Dim 2 1 st Leg Dim 2 2 nd Leg Dim 2 3 rd Leg Dim 1, 1 st Leg 1.00.57.60.26.57.69 Dim 1, 2 nd Leg.67 1.00.59.69.06.09 Dim 1, 3 rd Leg.85.82 1.00.59.02.16 Dim 2, 1 st Leg.20.67.47 1.00 -.18 -.18 Dim 2, 2 nd Leg.72.07.36 -.22 1.00.75 Dim 2, 3 rd Leg.68.03.29 -.23.75 1.00 The Pearson correlation between the 1 st dimension coordinates in the first and second legislatures is.67. Nevertheless, once we exclude the Communists, the correlation is only.08. On the other hand, the first-dimension coordinates correlate strongly with the second (anti-regime) dimensions in both the second and third legislatures. These findings corroborate the conclusion that the first dimension in the first legislature captures antiregime voting, whereas it captures left-right voting in the second and third legislatures. The Pearson correlation between the first dimension coordinates in the second and third legislatures is strong even when excluding the Communists from the analysis (.64). These results show that much of the left-right and the anti-regime conflict carried over between the three legislatures. The ideological stability, however, is less than in Congress. Poole and Rosenthal (1997, p. 72) report House t to t+1 correlations (for W- NOMINATE) averaging 0.96 in the 1945-80 period, and t to t+2 correlations averaging 31

0.89. The corresponding figures for the Senate are 0.87 and 0.85. In making the comparison, it should be remembered that the first legislature and the second both lasted more than four years as against two years for United States Congresses. The French National Assembly had higher turnover than recent United States Congresses, of the 645 we included from the first legislature, only 234 served in both the second and the third. But similar turnover can be found in the late nineteenth century for the United States, the twentieth century United Kingdom, and, last but not least, the Fifth Republic without any suggestion of political instability. Another way to assess the stability in the legislator coordinates is to apply the Optimal Classification algorithm to all three legislatures simultaneously. Doing this assumes that the legislators that serve in more than one session have fixed ideal points that anchor the space for the remaining legislators. In two dimensions, this pooled model classifies 96,1% of all vote choices correctly. This is only 0,7% less than if we estimate each legislature separately. Overall, these results are not indicative of political instability. Explaining Within-Party Differences The primary purpose of MacRae s study was to investigate the dimensions of cleavage within the major parties that caused political instability. To meet this objective, he classified clusters of roll calls that exhibited similar divisions within a party. He then created ordinal issue scales that capture the most important intra-party divisions. 13 Table 3 demonstrates that a two-dimensional spatial model of the entire legislature generally explains intra-party differences well. Table 7 evaluates whether the ideological positions 32