In France, as in other western democracies, the fortunes of parties and candidates wax and wan. These ups and downs are avidly watched by activists and pundits who wish to make some sense of them. The 2007 French elections, in particular, have raised an ever growing number of questions. Will Ségolène Royal run again? Is it over for the left? Is the right on the rise? Are the parties divorced from the people? Does the trajectory of such events reveal rhyme and reason, or is it simply random and ultimately inexplicable? Perhaps more importantly, has the outcome of the 2007 elections marked a significant turning point regarding the current state of democracy in that country?
This article sheds light on those issues through an examination of the evolution of the French party system and electorate throughout the Fifth Republic and aims at putting the 2007 elections in perspective. We look first at the pattern of party electoral movement over time. Then, we evaluate the cycle with respect to certain equilibrium conditions, namely stability, competition, and democratic rule. Based on these patterns, we arrive at a classification of French national elections. A key conclusion is that, in general, these Fifth Republic elections are in disequilibrium and often fail to express the majority partisan will. Finally, the 2007 results, which represent a return of the traditional right, are examined using the “Iowa” forecasting model (Lewis-Beck et al., Reference Lewis-Beck, Bélanger and Fauvelle-Aymar2008a) to see if they depart from these troubling conclusions.
Partisan Cycles
Partisan cycles have been the object of limited study at the national level, in France or elsewhere.Footnote 1 Examining the American case some time ago, Stokes and Iversen (Reference Stokes, Iversen, Campbell, Converse, Miller and Stokes1966) first explored the idea of equilibrium in electoral results. Fortunately, this tradition has been recently resuscitated by Norpoth (Reference Norpoth, Dorussen and Taylor2002), through an analysis of the Democratic presidential vote, with a time series (1828–2000) that he asserts has equilibrium characteristics (Norpoth, Reference Norpoth, Dorussen and Taylor2002: 128–29). First, the vote is rather stable, always staying within a 35 per cent-65 per cent range.Footnote 2 Further, the party system approximates a model of perfect competition, with the Democratic party repeatedly and regularly moving above and below the 50 per cent election winner's mark. More specifically, the average Democratic percentage share is 49.6 per cent, suggesting that races are very close, with no one party keeping the White House for many terms. Indeed, the expectation is that the winning party will not hold the office for more than two terms.
Curiously, outcomes are roughly the same for the example of the United Kingdom (Lebo and Norpoth, Reference Lebo and Norpoth2007; Norpoth, Reference Norpoth2004). The division of the vote between the two leading parties looks to be highly stable. In a time period of close to 200 years, the Conservative party lead never surpasses 20 percentage points. Further, contests seem very competitive, for the average Conservative margin is −2.2 percentage points, falling at the border of statistical significance. Practically speaking, the idea is that elections will often be real horse races. In fact, 12 out of the 44 elections from 1832 exhibited a telling change in partisan advantage. The full series appears in equilibrium, tracking an AR(2) process allowing for about two terms in power, then two terms out of power.
With respect to France, there has been no scientific study of its partisan electoral cycles. Perhaps this is due to the complexity of the French party system. In the Fourth and Fifth Republics, hundreds of political parties and movements have contested national elections. To impose some order on this vast array, we order coalitions ideologically, using the classic divide of left and right. For example, with respect to ruling parliamentary coalitions, before the 1981 election the government was on the right, mainly Rassemblement pour la République (RPR) and Union pour la Démocratie Française (UDF), while before the 1986 election it was on the left, mainly Socialist and Communist. In Figure 1 appears the left vote percentage of these two central parties—the Socialists and Communists—from 1924 to 2002, for first-round National Assembly elections.

Figure 1. Communist plus Socialist Vote Share (%), French Legislative Elections, 1924–2002*
* Percentage of valid votes. The data are assembled from Knapp (Reference Knapp2004: 62, Table 3.1).
The mass strength of the core left has fluctuated dramatically over this extended time span, from a peak of 52.2 per cent (1981) to a low of 26.9 percent (1993). This range is large, with the high score about double the low one. Further, these extreme values are not mere outliers. If they are removed from the dataset, the next two extremes are 49.6 (1945) and 28.8 (1932). In other words, considerable variation in support for the left remains across the time period, even over the relatively short span of 78 years. Indeed, if the series is subjected to Auto-Regressive Integrated Moving Averages (ARIMA) modelling techniques, it is identified as AR(1) and not stationary.Footnote 3 The foregoing description thus reveals considerable fluctuation in the fortunes of the left. These swings have a rough pattern. Note in particular the up-down-up-down trajectory of the time line in Figure 1. There appears almost a generational periodicity of support, peaking in the 1940s and 1980s.Footnote 4
These periodic surges in the popularity of the left suggest it would be quite competitive in the electoral arena. But is it? The short answer is “no.” In the earlier Republics, the celebrated victory of the left is that of the Popular Front in 1936. But let us focus our analysis on the more recent Fifth Republic. The alternance of 1981, which brought Mitterrand into power and the first government of the left, was heralded as a political watershed. Since then, the left has taken control of parliament only twice, in 1988 and 1997. In Table 1 we see the outcomes for Fifth Republic legislative elections, for the total left (Communists, Socialists, extreme left, moderate left and Greens). All told, of the twelve National Assembly elections held between 1958 and 2002, the traditional right has gained the parliamentary majority nine times. Moreover, the average total vote for the left equals 46.3 per cent. This percentage is well short of the implicit, election-winning, 50 per cent-plus score. Thus, these French outcomes, unlike those from Great Britain and the United States, point to an electoral system that is unstable, uncompetitive, and out of equilibrium.
Table 1. Outcomes of Fifth Republic National Assembly Elections, 1958–2002*

* Total left = extreme left + Communists + Socialists + moderate left + greens, always for the first round, National Assembly contest (except 1986 when there was only one round). Source: Ministère de l'Intérieur.
a = left majority in the National Assembly.
A Classification of Partisan Outcomes
It has proven useful to classify US elections according to the vectors of partisan loyalty. The authors of The American Voter developed a four-fold partisan classification of elections: maintaining, deviating, reinstating, realigning (Campbell et al., Reference Campbell, Converse, Miller and Stokes1960, Reference Campbell, Converse, Miller and Stokes1961). Assignment of an election to one category or another depends on the relationship of two-party identification and two-party vote choice in the electorate, and whether they are in agreement. Specifically, an election is maintaining if the party with majority identification stays in office, for example, in 1964 the majority of party identifiers were Democratic, and Democrat Lyndon Johnson won the presidency. An election is deviating if the party with majority identification loses, for example, in 1956 the majority of party identifiers were Democratic, but Republican Dwight Eisenhower won. An election is reinstating if the party with majority identification wins after having lost, for example, in 1960 the majority of party identifiers were Democratic, and Democrat John F. Kennedy was elected, defeating Republican incumbent candidate Richard Nixon. Finally, an election is realigning if the party with majority identification changes, for example, in 1932 Franklin Delano Roosevelt became president and majority party identification changed from Republican to Democratic.
This system has never been applied to the French case, with the exception of Lewis-Beck (Reference Lewis-Beck1986), who argued that the 1986 National Assembly elections could be classified as something like reinstating, according to this schema. Certainly, it was true that the traditional right returned to parliamentary power after that contest. However, that is not sufficient evidence to make the classification. What needs to be known, and what Lewis-Beck (Reference Lewis-Beck1986) failed to report, are the relative measures of partisan identification in the electorate. Only by knowing the identification of the partisan majority, left or right, can the 1986 result be unambiguously categorized. For example, if in fact voters identified more heavily with the left, then it would have been more properly labeled as a deviating election.
In other words, the proper classification of the macro-partisan election outcomes rests on an understanding of the micro-foundations of partisan loyalty. For France, there has been a long-standing debate on the relative importance of party identification and ideological identification, as long-term psychological anchors of the vote choice (Converse and Pierce, Reference Converse and Pierce1986; Boy and Mayer, Reference Boy and Mayer1993; Evans, Reference Evans and Lewis-Beck2004; Fleury and Lewis-Beck, Reference Fleury and Lewis-Beck1993; Lewis-Beck and Chlarson, Reference Lewis-Beck and Chlarson2002; Pierce, Reference Pierce1995). The most sophisticated work, utilizing simultaneous equation techniques with panel data from the first round of the 2002 legislative elections, indicates that “ideological identification systematically outweighs party identification in shaping the vote” (Bélanger et al., Reference Bélanger, Lewis-Beck, Chiche and Tiberj2006: 514). Thus, we take ideological identification as our principal marker of partisan feeling in the mass electorate.
Measuring left–right self-placement has a long tradition in French political surveys. The first recorded instance we have uncovered comes from 1946, where the apparent distribution of public opinion on this scale was 35 per cent on the left, 39 per cent in the centre, 26 per cent on the right (Lewis-Beck, Reference Lewis-Beck, Dalton, Flanagan and Beck1984: 440). Michelat (Reference Michelat, Boy and Mayer1993: 67) graphs the left–centre–right distribution in surveys across the years 1964–1988. The question typically reads something like this one from the 1995 French National Election Survey: “One usually classifies French people on a scale of this type, going from left to right. Where would you personally place yourself on this scale?” [Interviewer shows respondent a seven-point scale running from left to right.] Out of a total N = 4,078 interviewees, 98 per cent were able to place themselves on this seven-point scale. In general, whenever such a scale has been used, the response rate exceeds 90 per cent (Lewis-Beck and Chlarson, Reference Lewis-Beck and Chlarson2002: 28). Clearly, the French do not hesitate to locate themselves along this ideological spectrum.
In Figure 2 we see a time series of aggregate ideological identification in France, from 1962 to 2002.Footnote 5 The solid line represents identifiers on the left, the broken line identifiers on the right. Centre identifiers are the residual category, 100 − (L + R). The left curve rises and falls in a manner reminiscent of the vote for the left in Figure 1. Indeed, across the series, the measure for identification on the left, by election year, yields a high correlation with the total vote share of the left for that year, r = 0.64 (p-value = 0.034). This is not surprising, since among individual-level vote predictors, none shows a greater correlation with choice than ideology. For example, in the 2002 French National Election Study, before the first-round National Assembly vote, the correlation between left–right self-placement and vote intention is 0.65 (Bélanger et al., Reference Bélanger, Lewis-Beck, Chiche and Tiberj2006: 506). This individual-level correlation has shown remarkable stability over time. Lewis-Beck (Reference Lewis-Beck, Dalton, Flanagan and Beck1984: 442), reporting on a series of survey analyses, indicates an average ideology–vote correlation of 0.66. In sum, the aggregate level ideology–vote correlation accurately reflects the individual-level ideology–vote correlation.

Figure 2. Left–Right Ideological Identification (%) in the Mass Electorate of France, 1962–2002*
*The lines track the percentage (out of 100) of respondents who identified with the left (or right) on a self-placement scale. The scale, which could range from three to ten points, was always collapsed into three categories: left, centre, right. See Footnote note 5 for data sources.
Given the strong link between ideology and vote, we would expect the dominant mass ideological identification to be congruent with the dominant political tendency at the ballot box. That is, given that parties on the right have almost always triumphed over parties on the left in these National Assembly contests, our expectation is that right-wing ideology would correspondingly dominate left-wing ideology in the mass electorate. But that is not what we find. Instead, in terms of ideological identification, the right has virtually never prevailed over the left during the entire course of the Fifth Republic. Below, we explore further the implications of this anomaly, with particular reference to the classification of French election outcomes.
With these results in hand, we are in a position to classify French national elections according to their partisan dynamics, following The American Voter type of schema outlined above. For each election since 1962, we examine the majority partisan identification, and the partisanship of the winning coalition. According to the rules, there are no maintaining elections, that is, the left majority never wins after having already been in office. In other words, it never manages to “maintain” its legislative power from one election to another. With respect to deviating elections, there are many, that is, while the majority of voters virtually always identify themselves with the left, it frequently loses (1962, 1967, 1968, 1973, 1978, 1986, 1993, 2002). In terms of reinstating elections, there are three, 1981, 1988, and 1997. For these contests, the left majority made a comeback, after losing their majority in the previous legislative contest. There appear to be no realigning elections. While in recent years the gap between left and right may be closing, it was still around ten points going into the 2002 election, suggesting that no partisan realignment in the electorate had yet taken place.
What do these findings tell us about French electoral democracy? They imply a deficit. Almost all these legislative elections can be described as deviating, none as maintaining. That is, the ideological majority in the mass electorate seldom attains parliamentary control and never keeps it, from one National Assembly election to the next. Borrowing from Rousseau, the “general will” of the French people, at least those with a political point of view, appears routinely thwarted at the ballot box. That majority, always on the left throughout the Fifth Republic, must vote in overwhelming numbers in order for it to gain legislative power.Footnote 6 Specifically, the left needs about 40 percent support in public opinion in order to win (that is, when ideological identification on the left stands at 40 per cent, the predicted first-round vote share is 50 percentage points or more: V = 14.97 + 0.88 (40) = 50.17).
As can be seen from Figure 2, when support for the left is this high, at 40 percentage points, support for the right is only around 25 percentage points. The workings of the political system, then, appear to seriously handicap the left, for it must have a big advantage over the right in order to win. Further, this handicap is made worse by a slight anti-left bias in the translation of votes into seats. In particular, according to the swing ratio of first-round vote share to legislative seats, a vote share of 50 per cent has an expected yield of only a little over 48 per cent of the seats (that is, left seats = −84.40 + 2.66 (50) = 48.6).
One may object to the conclusion that the system disadvantages the left on grounds that the centre identifiers are ignored. Clearly, some of these centre identifiers vote for right-wing parties. However, this argument seems neither necessary nor sufficient, as the American case illustrates. In the foregoing classification of US elections, the centre is also ignored, but without the same ill effects for the left. More precisely, in the US, there exists a group of independents, identified with neither major party. The proportion of these independents fluctuates but is normally around one-third of American National Election Study (ANES) respondents. For example, in 1984 it was 34 per cent, in 2004, 39 per cent. Across the entire period of studies, from 1952 to 2004, the percentage of Democratic identifiers has exceeded the percentage of Republican identifiers (Lewis-Beck et al., Reference Lewis-Beck, Jacoby, Norpoth and Weisberg2008b: 111–27). In other words, the majority partisan identification in the US has always been Democratic during this period, and the Democrats have managed to translate this into White House victories on at least five occasions (out of 14).
Contrast these US results with the French results. For France, election surveys also reveal a group of respondents in the ideological centre, hovering near 30 per cent, who reject identification with both left and right. For example, in 1985 it was 28 per cent, in 2002 it was 26 per cent. Again, across the entire period for which we have data, 1962–2002, the percentage of identifiers on the left has exceeded the percentage of identifiers on the right. Despite the fact that the left has always been in the majority ideologically, it has managed to take control of parliament only three times (out of 11). With respect to presidential elections, it has been successful in only two (out of eight). Both systems, then, the US and France, have a democratic deficit, measured by the extent to which the partisan majority is denied the power to govern. Of the two democracies, the deficit appears greater in France.
Democracy in Disequilibrium?
The partisan outcomes of the national elections of Fifth Republic France are characterized by considerable instability, lop-sided competition and noteworthy right-wing bias. As a consequence, the governing desires of the partisan majority are seldom realized. Democracy, defined simply as the expression of majority partisan preference, seems to fall short of being totally fulfilled in France.
Democracy, of course, also contains other elements, such as a healthy competition between balanced rival party coalitions, which regularly share and exchange power. In French legislative contests, these elements are also lacking. These conclusions square with the recent broad-scale critique of Knapp: the “party system—and particularly the system as manifested, through elections, in France's parliamentary and governmental élites—has become increasingly disconnected from the voters… there is no sign of an end to France's malaise” (Reference Knapp2004: 342). Altogether, we must conclude that France was a democracy in disequilibrium, at least heading into the 2007 elections. The question we turn to now is how those elections affected this disequilibrium, if they did at all. Or, in Knapp's words, does 2007 show any “sign of an end to France's malaise”?
The Meaning of the 2007 Elections
Does the Sarkozy presidential victory, followed by a comfortable legislative win for the right, signal a break with the past? First, we look at raw election results for the presidential and legislative elections of the Fifth Republic. The left's share of the vote for the 2007 presidential contest in the first round is the third smallest of the series, at 36.4 per cent (only in the Gaullist elections of 1965 and 1969 was it smaller). In the legislative area, the picture is even worse. In the first-round National Assembly elections of 2007, the left garnered just 40.3 per cent of the vote, the smallest share ever registered from 1958–2007. Clearly, the raw results of the 2007 contests are extreme, not to say outlying, values. They are nowhere near the outcomes which are more representative of the Fifth Republic.
Bolstering the raw results, forecasting models suggest the outlier status of the 2007 election. Table 2 displays two forecasting models, one presidential and the other legislative. The models, based on what has come to be known in France as “le modèle de l'Iowa” (Fauvelle-Aymar and Lewis-Beck, Reference Fauvelle-Aymar and Lewis-Beck1997, Reference Fauvelle-Aymar and Lewis-Beck2002; Lewis-Beck et al., Reference Lewis-Beck, Bélanger and Fauvelle-Aymar2008a), are parsimoniously specified, taking into account the scarce number of cases. Further, their specification is dictated by well-established political economic theory, that is, the incumbent is rewarded for good economic and political policies, and punished for bad (Lewis-Beck et al., Reference Lewis-Beck, Bélanger and Fauvelle-Aymar2008a; Lewis-Beck and Stegmaier, Reference Lewis-Beck, Stegmaier, Dalton and Klingemann2007). In particular, the model reads as follows: Left Vote (first round)t = Unemploymentt−1 + Executive Popularityt−1
Table 2. French Election Forecasting Models (OLS), with and without the 2007 Results

* = absolute value of the t-ratio (in parentheses) is greater than 2.00.
a These presidential results, for the first round (percentage votes for total left) are from Lewis-Beck et al. (Reference Lewis-Beck, Bélanger and Fauvelle-Aymar2008a). The first observation is 1965. The economy variable is the unemployment rate, logged, measured about three months before the election; the popularity variable is for the president, measured in an IFOP poll about three months before the contest (multiplied by +1 or −1 depending on whether the president is on the left).
b These legislative results forecast the first-round, National Assembly vote share for the total left. The first observation is 1962. The economy variable is measured the same as in the presidential model; the popularity variable is for the Prime Minister, measured in an IFOP poll about three months before the election (multiplied by +1 or −1 dependent on whether the PM is on the left).
As an experiment, the 2007 data are excluded, then included. In Table 2, columns 1 and 2, compare the ordinary least squares (OLS) estimations of presidential model A (the excluded condition) and presidential model B (the inclusive condition). With model A, the coefficients are easily significant, and the R-squared satisfactory. Moreover, the Durbin-Watson test indicates no significant autocorrelation, the Cook's distance measure reveals no troublesome cases, and the residuals are evenly balanced between positive and negative values. Comparing model B to model A, the coefficients and the t-ratios have deteriorated substantially, as have the R-squared and the standard error of estimate (SEE). Further, the Durbin-Watson test indicates possible autocorrelation and the Cook's distance measure shows that 2007 is a serious outlier. Finally, the within sample prediction error for 2007 is a whopping 8.6 percentage points. The 2007 presidential result, then, clearly stands out as unusual. In addition, its uniqueness cannot simply be attributed to the small sample size. A parallel analysis, using regional instead of national data, allowed an N = 81 and yielded the same outlier status for 2007 (Lewis-Beck et al., Reference Lewis-Beck, Bélanger and Fauvelle-Aymar2008a).
The legislative models demonstrate the same conclusion. Compare the estimations of model C (2007 excluded) to model D (2007 included). The coefficients for economics and popularity each lose their statistical significance.Footnote 7 Further, the R-squared is halved, while the SEE increases greatly. In addition, the Durbin-Watson statistic is more problematic. Particularly telling is investigation of the within sample prediction errors of model D. Of the 12 elections, the largest error of all is for 2007, an 8.2 percentage point overestimation of the left's share of the vote. In terms of the standardized (or studentized) residuals, as well as the Cook's D measure, the 2007 election displays the most extreme values of any in the series.
Obviously, these 2007 elections are not elections “comme les autres.” Instead, they are special. In them, the left suffered historic, unprecedented losses, which traditional explanations can only partially account for. What did happen? What is “the rest of the story”? Recall Figure 2 and its curve of left-wing ideological identifiers. Across the period, the right-wing identifiers have never surpassed the left-wing identifiers in number. In fact, on average, the left-wing identifiers have been 10.6 percentage points ahead of the right-wing identifiers. Before the 2002 elections, the gap was a yawning 11 points, that is, 40 − 29 = 11. Using a comparable left–right self-placement scale, and applying it in April 2007, the gap is now −3, that is, 32–35 = −3 (Michelat and Tiberj, Reference Michelat and Tiberj2007: 384). In other words, for the first time ever in the Fifth Republic, right-wing partisans weighed in with more force than left-wing partisans. Of course, the possibility that this result is merely artifactual, the product of survey item reactivity bias because of its closeness to the election itself, cannot be totally discounted. Nevertheless, on the face of it, the French electorate has experienced not a “cultural shift,” in Inglehart's (Reference Inglehart1990) phrase, but rather an “ideological shift.” The stunning defeat of the left, and the Sarkozy victory wave, may be laid at its feet.
Conclusion
Given that the right, rather than the left, now forms the partisan majority in the mass electorate, one must ask if 2007 represents a realigning election. By definition, in a realigning election the partisan majority, within the public, switches. In the US case, such a realignment has occurred rarely, and there is controversy over which elections pass the test (see for example Mayhew, Reference Mayhew2002). Commonly mentioned are the Civil War election of 1860 and the decisive regional contest of 1896. The New Deal elections of the 1930s, where the majority shifted from Republican to Democratic, is the most recent case in point. Obviously, identifying a realignment can be difficult. A common problem is the time requirement, that is to say, it may take several years, and several elections, to establish that the partisan majority has really changed.
Clearly, it remains to be seen whether France has entered a realigning era. Hellwig, in a current commentary, expresses his doubts, saying “that changes signaled by 2007 do not amount to the arrival of an all-out realignment” (Reference Hellwig2007: 17).Footnote 8 In general, we have sympathy with such a view. Some years ago, Lewis-Beck expressed the idea that the French electorate was “‘stalled’ in past habits” (Reference Lewis-Beck, Dalton, Flanagan and Beck1984: 446). He went on: “What impresses me most about mass attitudes and behaviour in France during the Fifth Republic is the persistence of the old political cleavages… Perhaps, then, the ‘doctrinal’ or ‘ideological’ quality of French politics has prevented any lasting electoral realignment” (Lewis-Beck, Reference Lewis-Beck, Dalton, Flanagan and Beck1984: 446; see also Lewis-Beck, Reference Lewis-Beck and Lewis-Beck2004: 11). By and large, the old cleavages, or variables, and their relative weights, have persisted. For example, left–right ideology is as important a predictor of vote choice as ever. However, what may be changing, and what we had not foreseen, was a fundamental change in the ongoing distribution of the cleavages, specifically, the possibility of movement from a left-dominant to a right-dominant mass ideology. This is the shift that may well be taking place now, and that explanatory models of French electoral behaviour such as the one presented herein may have not been able to pick up. If the 2007 outcome proves to be part of a growing change in France's traditional societal cleavages, then these models, including the “Iowa model” presented herein, are doomed to fail in the coming years and will be in need of significant revision. But if 2007 proves to be a mere outlier, and the French electorate still remains stalled, then the models should likely continue to perform relatively well in predicting future contests.
The question as to whether the French political system is in a state of disequilibrium ought to interest non-French scholars, and particularly Canadians who are often concerned about the seemingly disequilibrating and volatile features of the Canadian party system. In Canada, the federal Liberal party has been largely dominant over the past century (Clarkson, Reference Clarkson2005; Blais, Reference Blais2005) even while the electorate has appeared to become permanently “dealigned” (LeDuc, Reference LeDuc, Gagnon and Tanguay2007). What this study of the French case suggests is that, in order to make some sense of the evolution of the Canadian political system, it might be interesting to look more closely at the electorate's ideological leanings, especially as this system seems to have taken on a more ideological character of late.
How does the 2007 French election result bear on the more general state of “democratic disequilibrium” that France has found herself in? In the sense that the partisan majority in the electorate matches the partisan majority in government, at least one particular democratic aspiration is fulfilled. What remains to be seen is whether there will be more balanced competition between left and right, with more alternation in power. Ironically, the right had managed to govern France most of the time without a mass partisan majority. If the mass partisan majority remains on the right, does that mean that the right will have an even easier time holding onto power? This is a special concern, given the anti-left bias in the translation of votes to seats. It seems clear that a France governed by a “permanent majority” of the right (or the left) would be ill served. However, it is in the nature of democratic systems, and France is no exception, to bring down such attempts at permanent majorities. Rulers as they rule accumulate enemies and sooner or later they are brought down.