LEADER IMAGES AND THE PRESIDENTIAL VOTE IN FRANCE*

Transcription

1 LEADER IMAGES AND THE PRESIDENTIAL VOTE IN FRANCE* Richard Nadeau Department of Political Science University of Montreal Michael S. Lewis-Beck Department of Political Science University of Iowa * Prepared for presentation at the European Consortium for Political Research Workshop, Leaders and Democratic Elections, directed by Marina Costa Lobos and John Curtice, Lisbon, Portugal, April Please do not cite or quote without permission.

2 The decisive importance of presidential leadership in France stands as a given of the Fifth Republic. The legend of De Gaulle comes to mind unbidden. An early essay on French political behavior, by Converse and Dupeux (1966), registered his dominating stature, in the felicitious title De Gaulle and Eisenhower: The Public Image of the Victorious General. But, aside from popular impressions and the historic narrative, we in fact know almost nothing about the impact of leader images on the French presidential vote, from De Gaulle to Sarkozy. With the exception of an essay by Pierce (2002), in King s (2002) collection on leader personality and elections, the French case has not been systematically addressed. Such neglect seems strange, given the many studies on leadership and electoral choice that have been carried out in other Western democracies. [For samples see, on Britain: Crewe and King (1994); Clarke, Sanders, Stewart and Whiteley (2004); on America: Kessel (2004); Wattenberg (1998); on Portugal: Lobo (2008); on Australia: Bean and Mughan (1989); on a general treatment: Curtice and Holmberg (2005), Webb and Poguntke (2005).] Here we erase this neglect, through examination of leadership effects on vote choice in four French presidential contests 1988, 1995, 2002, and 2007 as measured in national surveys from each. Below, we begin with a discussion of issues surrounding the definition and measurement of leader image. Then, we propose measures from the French surveys, arriving at a general one the candidate feeling thermometer driven by specific dimensions, such as competence and integrity. These leader image measures are related to French presidential vote choice, in carefully specified models. Standard explanations of this vote choice, based on social cleavages, ideology, and issue preferences are incorporated. As well, ballot (first-round v. second) and party (right v. left, mainstream v. extreme) are taken into account. In conclusion, the relative strength of leader image, as a determinant of presidential candidate success, is assessed. As shall be seen, its impact is varied and strong. LEADER IMAGE: MEANING AND MEASUREMENT In their founding volume on political behavior, Campbell, Converse, Miller and Stokes (1960) make the pivotal impact of candidate perception clear. With regard to the public image of General Eisenhower in the 1952 presidential campaign, they report that 1

3 his most favorable perceived personal attributes were his record of service, and his integrity (Campbell, Converse, Miller and Stokes (1960, 55). These attributes continued at the top of his list before the 1956 landslide re-election, along with likeability ( I like Ike. ). Lewis-Beck, Jacoby, Norpoth and Weisberg (2008, 45), in their revisitation of The American Voter, find similar results in the 2004 contest, with President Bush seen, foremostly, as strong, decisive and honest, has integrity, ideals. In contrast, his to-bedefeated opponent, Kerry, was perceived, more than anything else, as weak, indecisive and dishonest. (Lewis-Beck et al. (2008, 49). These results suggest a leader s image can make a real difference at the ballot box. With respect to the theoretical literature, different dimensions of this image have been identified. Fiorina (1981, 151), looking at the candidate attribute data from The American Voter (Table 3.12) referenced above, sorted them into instrumental or affective, according to whether the trait aided in the task of governing, or was merely a personal characteristic. Using his schema, the first two Eisenhower attributes in the preceding paragraph can be classified as instrumental, the last affective. The most thorough sorting of these open-ended American National Election Study (ANES) questions, from 1952 forward, has been carried out by Kessel (2004), who grouped candidate qualities as follows: general, record-incumbency, experience- management, intelligence, trust, personality. Overall, in the American research literature, there is much agreement on the electoral relevance of two candidate traits: competence and integrity. (Interestingly, these are the two traits that Bush in 2004 scored the best in). The former relates to performance in office, or record of ability on the job. The latter relates to an image of honesty and fairdealing. Miller, Wattenberg and Malanchuk (1986), in a factor analysis of these likedislike responses from the ANES ( ) identify the following dimensions: competence, integrity, reliability, charisma, purely personal traits. Kinder (1986), in a different examination, of closed-ended items, identifies these dimensions: competence, integrity, leadership, empathy. Clarke, Sanders, Stewart and Whiteley (2004, 30-31), in their investigation of leader images in the United Kingdom (UK), find two dimensions: competence and responsiveness. The former includes keeps promises, decisive, principled ; the latter included caring, listens to reason, not arrogrant. This latter 2

4 UK dimension, of responsiveness, seems to have been relatively neglected in the US literature. However, the UK dimension of competence (which appears to include integrity), finds its practical US counterpart in their emphasis on competence and integrity. With respect to France, there have been no formal studies of leader images and their electoral impact. However, in the surveys available, items relevant for the measurement of presidential candidate perceptions have been posed. Unfortunately, expect for one to be mentioned, they are not posed with any consistency. Nevertheless, we do manage items on the following dimensions: competence, integrity, and personality. Competence includes whether the candidate has the right stuff and a satisfactory record. Integrity includes honesty, and keeping promises. Personality items include appearance on television. Our expectation is that these components of candidate perception relate to, indeed help determine, the leader s overall image. Certainly, this is what the British election survey data show (Clarke et al, 2004, chp.2). They find that party leader images are structured in terms of two strongly interrelated competence and responsiveness dimensions noted above (Clarke et al., 2004, 92). Their dependent measure of leader image is a 0-10 self-placement scale, from strongly dislike to strongly like. We employ a similar scale, consisting of a feeling thermometer on each leading candidate (also initially scored 0-10, or 0-100). Fortunately, we have the feeling measure for the four most recent elections (1988, 1995, 2002, 2007) for the leading major party candidates (Socialist, RPR-UMP, UDF, National Front). Table 1 contains illuminating descriptive statistics on the public perception of these candidates. In Panel A are the mean scores for each, in each contest. We observe that, in general, the majority show sympathy for the leaders of the traditional left or the traditional right. (Le Pen, leader of the extreme right, regularly exhibits a distinct minority appeal). Further, that majority race is not always won by the leading left (i.e., Socialist) or leading right (i.e., RPR-UMP) party. Note, for instance, that in 1988 feelings for Barre surpassed those for Chirac, and in 2007 feelings for Bayrou paralleled those for Royal. [TABLE 1 ABOUT HERE] 3

5 These thermometer measures, while obviously reflecting vote preference to some degree, are tapping into something else, namely candidate appeal. In Panel B (Table 1) are sample correlations of our scattered candidate perception items with the thermometers, for these leading candidates. For instance, personal image (as measured by how the candidate looks on TV), has an average correlation, across all the 1995 candidates, of r =.44. Regarding competence, one sees, for example, that thinking Jospin has the right stuff for leadership in 2002 correlates.50 with his thermometer rating. With respect to the dimension of integrity, observe that judgments of Chirac s honesty in 2002 yields a feeling thermometer correlation, r =.49. One observes, encouragingly, that all these candidate image items generally relate well to their respective thermometer, but not perfectly. In fact, the median correlation, among all these items, is a respectable r =.46. An immediate criticism is that the feeling thermometers measure more than leader image. To quote Fiorina (1981, 154): No one knows what thermometer scores measure. 1 Put another way, they may tap into social structure, partisanship, or issue positions, as well as candidate traits themselves. For this reason, it is very important to be sure these competing explanatory variables, which are undoubtedly correlates of the thermometer itself, are fully controlled in carefully specified voting models. To this task we now turn. FRENCH PRESIDENTIAL VOTE CHOICE: A MODEL INCLUDING LEADERS The French tradition in political behavior studies is similar to, and different from, the American one. [On France, see the review by Mayer and Perrineau (1992)]. On the US, see especially the funnel of causality discussions in Campbell, Converse, Miller, and Stokes (1960, chp.2); Lewis-Beck, Jacoby, Norpoth, Weisberg, (2008, chp.2).] Consider the division forces acting on the voter: long-term (socio-demographics and partisanship) and short-term (issues and candidates). With respect to the former, both traditions share an attention to social cleavages, what the French term les variables lourdes. (Boy and Mayer, 1993; Cautrès, 2004; Michelat and Simon, 1977). However, considering the other long-term force partisanship they greatly differ, with American research emphasizing the role of party identification, French research the role of left-right ideological 4

6 identification. [On the first perspective, see Converse and Pierce, 1986; Pierce, On the second, see Boy and Mayer, 1993, 1997; Mayer, For a blend of the two, see Bélanger et al., 2006; Evans, 2002; Lewis-Beck and Chlarson, 2002.] 2 With respect to the short-term forces, each tradition includes issues. But regarding the other short-term force candidates - again they greatly differ, Americans emphasizing their role, and the French neglecting it. Ysmal (1994, 379), criticizing the 1988 French National Election Study, laments the dearth of questions about the candidates, concluding the authors could argue that the old forces of social class, religion and left-right identification had not declined in importance but the questionnaire design limited their ability in analyzing and contesting Indeed, the French inattention to the impact of candidate characteristics on presidential vote motivates this paper. At the same time, it creates obvious data limitations, which we attempt to overcome. In order to assess candidate impact, we need to first establish a baseline model of the vote choice, upon which to build. We propose the following: Vote = f (social cleavages, ideological identification, issues) Eq. 1 where vote = presidential vote (first-round choice among the four leading candidates of the Parti socialiste, the UDF, the RPR-UMP, and the Front National, or second-round choice between the final two); social cleavages = class, age, gender, education, income, religiosity; ideological identification = self-placement on a left-right ideology scale; issues = immigration, the economy. [On the enduring importance of these issues, see Lewis-Beck, 1997; Mayer and Tiberj, 2004; Whitten, 2004.] All these variables can be measured in the four most recent presidential elections contests -1988, 1995, 2002, and (Measurement detail on each is provided in the appendix). Such a model specification has performed well in previous analyses of French presidential elections, yielding satisfactory goodness-of-fit statistics. [For example, in 1988 and 1995 (the second-round) presidential contests, the respective R-squared =.55 and.54, the respective percent vote correctly predicted = 89.6% and 86.9 %; Lewis-Beck and Nadeau (2000, 179); see also Lewis-Beck (1993, 10).] Besides its comprehensiveness, this equation has the virtue of being estimable across all four contests, thus holding constant the fundamental model specification. 5

7 To this recognized baseline model, we simply add our candidate variable leader image as measured by the feeling thermometers. With regard to the analysis strategy, for each election a binominal logistic regression will be run on each of the two leading candidates (government v. opposition, first-round) and the final two candidates (secondround). Then, these results (first-round) will be compared to a multinominal logistic regression for each election, always with the incumbent party candidate as the reference category. Once the foregoing findings are fully interpreted, the feeling thermometers themselves will be explained, in regression models for leading candidates in each election. LEADER EFFECTS IN FOUR FRENCH PRESIDENTIAL ELECTIONS In Table 2, appear estimates (binomial logits) of the fully specified model of French presidential vote choice, for the four most recent elections. Observe, first, that the model generally does well in accounting for variance in choice, with the McFadden pseudo-r-squared averaging.43 on the first-round, and.78 on the second-round. Second, the signs of the coefficients are generally as expected, and there are multiple statistically significant effects. Consider first the long-term force of social cleavages. The strict demographic measures (age, gender, education, income) achieve statistical significance eleven times. The class measures (professional, white-collar, blue-collar, or private sector status) achieve statistical significance six times. Religiosity is statistically significant twice. Now consider the long-term force of ideological identification. Not surprisingly, it has a compelling impact. In addition to claiming conventional statistical significance in all cases (save one), it generally manages statistical significance at unconventionally high levels. (For example, the ideology coefficient for Sarkozy, first-round, yields a z-score of 10.65). With respect to short-term forces, consider the impact of issues. The economy and immigration, together, exert influence six times, according to the standard of statistical significance. [TABLE 2 ABOUT HERE] These positive results, rendering in themselves a substantive explanation of presidential vote behavior in France, provide the framework for our central aim: evaluation of the force of candidate appeal. To carry out this assessment, we turn to the 6

8 feeling thermometer coefficients. The vote choice for each candidate is held to be a function of the powerful control variables just reviewed, plus the leader image of the four principal candidates in the contest. Below, we first review the thermometer results from each first-round contest, beginning with the most recent. First-Round 2007 Election. Vote support for Sarkozy, as the incumbent UMP candidate, receives shape from the voter s feelings about Sarkozy himself, plus feelings about the other candidates (column 1, Table 2A). Observe that the strongest leader image is made by candidate Sarkozy in his own right (for the Sarkozy thermometer, b = 8.33). Unexpectedly, feelings about his chief rival, Royal, do not register a significant effect. However, sympathy for Bayrou, his centrist challenger, does. Those who found Bayrou an appealing candidate tended to turn away from Sarkozy, with a defection rate even greater than that from sympathy with Le Pen. (Compare the respective thermometer coefficients, b = and b = -2.28). Finally, in his public draw as leader, Sarkozy held a great image advantage, because a sympathetic view of his Socialist opponent costs him no votes. The popular perception of Royal the candidate simply gave her no traction against Sarkozy. In contrast, Royal s candidate appeal strongly motivated her own supporters. (column 2, Table 2A). Interestingly, Royal s supporters appear as intensely loyal to her as Sarkozy s were to him, as the respective coefficients (8.79 and 8.33) show. This comparison documents the journalistic attention, at the time, to the provocative personalities of these candidates, something rather unusual in French presidential campaigning. Bayrou s appeal cost her about as many votes as it cost Sarkozy, underlining the notion that, as a moderate, he was drawing votes from each of the leaders. Finally, Le Pen s evaluation made no difference, one way or another. Obviously, that does not mean that pro-le Pen supporters did not turn away from Royal; instead, it simply means they operate in an essentially different political universe, where the Royal candidacy is not a relevant choice, unlike that for Sarkozy Election. Viewing this competition, we see that Chirac support, while fed by own personal draw, was significantly undercut by the attractiveness of his Socialist rival, 7

9 Jospin (b = -2.40). (See column 3, Table 2A). This contrasts with Socialist Royal in Another contrast is the relatively weak impact of Bayrou sympathy, suggesting the relatively feeble appeal of the center in that race. Again, as in 2007, the traditional right candidate loses votes to Le Pen backers, and to the same extent. (The coefficients, respectively, are and -2.23). This indicates that one Gaullist leader image was not more successful than the other, in staunching the flow of votes to the extreme right. The pattern of sympathy for Jospin has similarities and differences. (See column 4, Table 2A). Note that both the leading candidates Jospin and Chirac elicit about the same level of passion. (Respectively, the coefficients are 7.24 and 6.98). On that score, they both seem equally strong candidates, a finding that contradicts dominant press commentary of the time regarding Jospin s weakness as a candidate. Further, the notion that Jospin was, in fact, a capable enough leader reinforces itself in the face of the absence of significant vote losses as a function of sympathy for Chirac or Bayrou Election. In this election, Jospin was the standard bearer for the incumbent Socialists. His own candidate traits registered positive enough (b = 5.74), although not to the extent they would in his 2002 race (column 6, Table 2A). Further unlike 2002, Chirac, on grounds of his own charisma, led a significant number of voters away from Jospin (b = -1.15). Le Pen, with his version, did the same (b = 1.74). Thus, the picture emerges of a relatively weak incumbent, in terms of candidate appeal, especially compared to the perceived attractions of traditional and extreme candidates on the right. His chief opponent, the Gaullist Chirac, evoked a lively battle of leadership images. His personal appeal produced a very positive effect, b = 7.33 (column 5, Table 2A). Still, this was balanced in the negative direction by the pull of his rivals on the right Balladur (b = -2.08), Le Pen (b = -1.72). These conflicting candidate appeals, on both left and right, suggest a close race, which in fact it was Election. In 1988, we see a candidate with a paramount leader image Mitterrand. Undoubtedly, the power of incumbency, combined with a long career as a skillful politician, helped elicit strong feelings from the voters, b = 8.21 (column 7, Table 2A). Moreover, he was able to consolidate his appeal, because none of his major opponents Chirac, Barre, or LePen - managed to diminish into this affective base. (Observe that none of their feeling thermometers achieve statistical significance). His 8

10 challenger, in contrast, recorded a somewhat weaker sympathy score for himself (b = 7.48), plus was significantly undermined by supporters of Le Pen and Barre (column 8, Table 2A). In particular, Barre, his rival for the leadership of the RPR, generated a substantial thermometer coefficent (b = -3.59), almost half the size of Chirac s. Clearly, Chirac faced serious rivals for leadership on the right. Mitterrand, by way of contrast, had no one taking away his spotlight. These first-round, binomial logit results are fascinating. The discussion, across these four elections, conveys the complicated dynamics of candidate perceptions in multicandidate contests. However, complicated as things may appear, they contain a simplification. Technically, the strategy of binomial logistic regression analysis assumes the implied voter choice model Candidate X v. all others is the functional equivalent of the implied multinomial voter choice model Candidate X. v. Y, controlling on others (Menard, 1995, chp.5). Fortunately, in our case these two strategies yield approximately the same empirical results, with respect to the two leading candidates (See the discussion in the footnote). 3 Thus, we retain the benefit that comes from the clarity of presentation the binomial results of Table 2 afford. Clearly, first-round candidates with better images received more votes. But, does image count for much, after all, in the second-round? Below, we look at candidate appeal, and how it affects this decisive ballot. Second Round The second-round competition drastically simplifies the candidate array, making it easier to generalize the results. In each of the four contests, we see that candidate image played a large role, independent of the other numerous and powerful forces at work (See Table 2B). Of all the explanatory variables, none register higher levels of statistical significance (as judged by the z-scores) than the feeling thermometers of the two candidates in competition. Further, the effects of candidate sympathy on the vote choice tend to be not far apart in magnitude, for the two contenders. Here are the comparative thermometer coefficients, for each election: Sarkozy = 13.27, Royal = (2007); Chirac = 6.96, Jospin = (2002); Jospin = -7.78, Chirac = 7.27 (1995); Mitterrand = 14.06; Chirac = (1988). This suggests that the radical reduction in the number of candidates from the first- to the second-round, coupled with the mobilizing effects of the 9

11 campaign, push the two aspirants toward an equilibrium. Note, relatedly, that unlike the first round, feelings toward the past candidates, with a few exceptions, cease to matter. Thus, if each finalist plays the game according to plan, campaigning in the usual way, leadership sympathies for them both tend to converge, in terms of sheer magnitude. The electorate comes to see them each as more (or less) acceptable, and to a roughly equal degree. Nevertheless, the paired thermometer scores are not identical. And, the candidate with the larger coefficient usually wins. The most dramatic example comes from 1988, where Mitterrand had a lopsided advantage over Chirac, in terms of his image as a leader. Also, for 2007, we see Sarkozy s candidate advantage over Royal. For 2002, we observe that winner Chirac holds the barest of margins over Jospin. (For comparative purposes, the dependent variable here is vote intention, assuming Jospin had made it to the second round. Chirac s leadership advantage was actually much greater, in the real contest against Le Pen in the second round). The anomalous case is that of 1995, where Chirac registers a thermometer coefficient just under that of Jospin. The explanation, which our analysis reveals, seems to lie with the intra-party rivalry inside the RPR. Recall that Gaullist Balladur, as of 1993, was serving as Prime Minister, and his faction wanted him to stand as the RPR presidential candidate in However, Chirac, as RPR party head, won out. In the first-round of balloting Balladur made a strong showing, just behind Chirac. In the second-round, Balladur supporters united behind their party, and threw their votes to Chirac. Thus one sees that the Balladur thermometer registers a positive, statistically significant effect on Chirac support, b = (Observe that, in all the other second-round results, the also-ran candidates fail, with one exception, to register a significant impact on the vote). In the end, then, it was Chirac s candidate appeal, combined with that of his fellow Prime Minister Balladur, that put it over the top for him. (Imagine adding their two thermometer scores, both positive, together, = 8.70). The message, an old one, is when two attractive candidates unite behind the party, the day may still be saved. 10

12 HOW STRONG ARE LEADER EFFECTS? SOME ESTIMATES The influence of candidate perception pervades vote choice in French presidential elections. At the first-round, the feeling thermometers of the four leading candidates register statistically significant effects 29 out of 32 times. At the second round, the feeling thermometers of the two competing candidates always register statistical significance. But we know, especially with survey data, that statistical significance does not mean substantive significance. In principle, a feeling thermometer coefficient could be statistically significant at some conventional level, but exercise only a trivial impact on the probability of voting for the candidate in question. Therefore, to assess substantive significance, we pose the traditional question what happens to Y when X changes? For logistic regression, in contrast to ordinary least squares regression, the answer to this question depends on where the values of the other X are set. Following convention, let us set them at typical (mean or modal) values. The next question considers the unit of change in the thermometer score to be evaluated. One possibility is to set it at the mimimum (0) and maximum (1) values; this generates estimates of very large effects. However, most voters do not fall at these extremes. Further, for voters who actually change their candidate sympathies, the movement along the scale would undoubtedly be much less than this absolute switch. Therefore, let us postulate a reasonable change, say one-half of a standard deviation. In terms of an original 100-point self-placement scale, this translates into a thirteen-point move, on average. 4 This plausible magnitude, a modest sort of sympathy shift, could potentially be experienced by many voters. Table 3 shows how such a change would impact the probability of voting for the incumbent candidate, at the second-round. Regardless of the contest, we see the impressive effects of leader image. The average effect, across all candidates and races, is.51, which is substantial. The most dramatic is that of 2007, where the probability of a Sarkozy vote goes up.78, in the face of such a change. Similarly, a like shift toward Royal reduces the probability by.70. These results show, in general, how important candidate traits can be. The recent 2007 competition shows, in particular, how personalized a contest can be. [TABLE 3 ABOUT HERE] 11

13 LEADER IMAGES: HOW TO EXPLAIN THEM? A voter s feeling thermometer score represents a summary measure of how he or she feels about the candidate. As we have seen, these feelings play a vital role in candidate choice. Where do they come from? In other words, what are the component parts of these perceptions? Early on (recall Table 1), we examined their simple relationship to various candidate traits, such as competence and integrity. We would expect these variables to help determine, in a serious way, candidate sympathies. However, in addition, we would expect that other forces, both short-term and long-term, would also mold these impressions. In particular, many variables that shape the vote choice, which we have already discussed, undoubtedly exercise a prior influence on candidate perception itself. Combining these notions, we posit the following general model for explaining the feeling thermometer scores: Thermometer = f (social cleavages, ideological identification, issues, candidate traits) Eq. 2 where the feeling thermometer, social cleavages, ideological identification, issues are measured as in Table 2, and candidate traits are measured as in Table 1. (See also the appendix). This model is estimated in Table 4. Because of data limitations, explanations are not offered on all sixteen thermometers (four elections times four candidates each). Candidate trait questions vary from election to election, and candidate to candidate. We employed all those available. Thus, we can offer more fully specified equations on the thermometers of the leading two candidates (or three for the case of 1995). These equations, then, contain extensive baseline variables as controls, plus candidate traits. (See Table 4). Note that the models generally do a satisfactory job of soaking up the variance in the feeling thermometers, with a median pseudo-r-squared =.50. [TABLE 4 ABOUT HERE] Social cleavages make a difference for candidate sympathies, upon examination of the patterns of statistical significance (denoted in what follows with + or -). Observe that candidates on the traditional right tend to be rated significantly higher by older people (+ Sarkozy, + Balladur, - Mitterrand), the more religious (+Sarkozy, +Chirac 95 12

14 and 02, + Balladur), and the more affluent (blue collar + Royal; private sector Royal, + Balladur; professional Jospin 02 + Chirac 88). Taken together, more established, welloff, church-going French citizens tend to lay their sympathies with conservative candidates. Ideology itself, over and above these structural features, plays a vital role. The magnitude of its coefficent has increased steadily, with the exception of Here are the ideology coefficients for the leading right-wing candidate: 2007 =.60, 2002 =.25, 1995 =.49, 1988 =.34. And, for the leading left-wing candidate: 2007 = -.55, 2002 = -.15, 1995 = -.41, 1988 = We see that, for both sides of the political spectrum, ideology is coming to assume a greater position, suggesting perhaps a greater ideological polarization of underlying candidate perception. (This point in itself is especially interesting, since it undercuts the popular view that the importance of the gauche-droite divide is waning in France). With respect to issues, the findings also engage. French voters seem to take into account two types of economic evaluations: valence and position. [See Stokes (1963), and Stokes and Dilulio (1993) on this distinction).] Valence issues refer to those on which there is an opinion consensus, e.g., economic prosperity, lower crime, less corruption. In both 1988 and 1995, the economic survey question used asked the respondent to evaluate national economic performance over the past year. In other words, it is the standard sociotropic retrospective question widely-employed elsewhere (Lewis- Beck and Stegmaier, 2002). In 1988 it exercised a significant impact on feelings about the incumbent Prime Minister, Chirac. The same pattern happened in 1995 for Balladur with spillover positive effects for Chirac. Position issues refer to those on which there is an opinion dissensus, e.g., abortion, tax policy, marijuana. The economic item used for the 2002 and 2007 surveys is positional, asking respondents whether they give priority to the unemployment problem, as opposed to some other problem. It exerts a significant impact on attitudes toward the Socialist candidate in both races. Turning to immigration, while it did not register significant effects in 1988, since then it has impacted sympathy (pro or con) for both candidates. Moreover, the Sarkozy coefficient (b =.07) hints that it is growing in importance for traditional right candidates 13

15 Intriguing as the foregoing is, our great interest lies in the impact of candidate traits. These items are varied and scattered. However, organizing them into our theoretical categories of competence, integrity, and personality helps. Competence is measured in 2002 (right stuff), and 1988 and 1995 (satisfaction). In 2002, those who felt that Chirac or Jospin had the right stuff were more likely to perceive them favorably. In 1988 and 1995, being satisfied with the candidate s record was a significant factor in candidate support in all five cases, seeming especially so for Mitterrand in That last point is theoretically reinforcing, since President Mitterrand had established a clear, seven-year long record in office. Similarly, in 1995, the fact that Balladur s satisfaction coefficient was higher that than of Jospin and Chirac makes sense, since he was serving as Prime Minister. We have integrity measures from 2002 (honesty, promises) and 2007 (confidence, proximity). The former are more specific, the latter more general. In the 2002 contest, while being perceived as honest was important for the images of both Chirac and Jospin, it was more important for Chirac. In contrast, keeping promises was important for both, but more so for Jospin. Clearly, for either candidate, the character trait of integrity mattered in terms of their popular appeal. In 2007, confidence in the candidate significantly influenced backing for both candidates, especially Sarkozy. Besides, in that contest, a feeling of closeness to his or her ideas had a significant impact, especially for Sarkozy. With respect to the last dimension, that of personal image, we have measures for 2007 (good candidate), and 2002 and 1995 (TV appearance). In both the latter instances, respondents are asked if their opinion of the candidate got better or worse after they saw him on television. This TV image variable has significant effects, at high levels, for all five of the candidates. It was strongest in 1995 and, within that three-way contest (perhaps the first seriously fought on the TV screen), strongest for Chirac (b =.30). Undoubtedly, this aided him in his party struggle with Balladur, and his final competition with Jospin. Indeed, the effects of this personality measure seem hearty, i..e, a unit change in opinion after TV viewing, form better to worse, would produce a 30-point rise in support on the feeling thermometer. Turning to 2007, where the respondent was 14

16 asked if he or she is a good candidate, we also find potent effects, especially for Sarkozy (b =.23). These measures, as disparate as they are, do seem of worth. If, despite their limitations, they clearly enhance overall model performance, further empirical reassurance of their value would be provided. A demanding test concerns how much they can add to variance explained, after the baseline variables have been allowed to explain all they can. Here are the increments (Full Model Adjusted R-Squared Baseline Model Adjusted R-squared): 1988, Mitterrand =.26; Chirac =.15; 1995, Jospin =.16; Chirac =.10, Balladur =.17; 2002, Jospin =.26; Chirac =.18; 2007, Sarkozy =.15, Royal =.10. These are large increments. Obviously, character traits, so measured, matter in the formation of candidate feeling. 5 CONCLUSIONS Leader effects have been an object of serious study in leading Western democracies. France has stood out as an exception here. We explore the impact of leader images in the four most recent French presidential elections (1988, 1995, 2002, 2007). The general measure of leader image employed is a candidate feeling thermometer. Our findings reveal pervasive and strong leader effects, net of other important influences on the French voter. Consider highlights from the first-round contests. In 2007, Sarkozy was greatly aided by the strength of his own leadership appeal, and by the weakness of Royal s leadership appeal. For 2002, both Chirac and Jospin were viewed very sympathetically by the public. Ironically, Le Pen, with his own base of sympathizers, derailed Jospin s chances there. In 1995, Jospin did not have the same candidate appeal he had in 2002, but neither did Chirac, who was undercut within his own political family by rival Balladur. The overall result was a close race. For 1988, the story is about the success of an effective and emotionally attractive president-in-office, Mitterrand. At the second-round, in contrast to the first-, candidate sympathies become more closely balanced, and are mainly motivated by attitudes about the two front runners. Generally, the one able to elicit a greater leader effect goes on to claim presidential victory. Still, feelings about the also-rans can make a difference, as they did in 1995 (with Balladur sympathy ultimately benefitting Chirac). 15

17 In general, how strong are these leader effects? First, note that they usually manage statistical significance, regardless of candidate, contest, or balloting-round. Beyond that, interpretation of the coefficients shows they are substantively significant. These strong effects can be demonstrated in different ways. We show that, following a modest change of plus one-half a standard deviation in a candidate s feeling thermometer, the average vote probability increase would be a substantial.51. Shifting candidate sympathies seemed especially vigorous in the personal contest between Sarkozy and Royal in 2007, where this average number was well exceeded, for both second-round candidates. What determines sympathy for a candidate? Many factors go into forming candidate perception, as measured by a feeling thermometer. Enduring elements of the voter s social and political situation count, namely cleavages and ideology. The cleavage effects might be summarized by saying that the bourgeoisie are more likely to sympathize with right-wing candidates. With respect to ideology, a long-standing attachment to the left or the right colors candidate perceptions, in the expected way. This is especially noteworthy, since we detected increasing polarization of candidate attitudes in the French mass public, along the left-right divide. More immediate concerns, such as the abiding issues of the economy and immigration, also count. To the extent the candidate aligns with the voter on these issues, they will be perceived more positively. These things are worth knowing. But the big question is the role of specific candidate traits in the formation of these attitudes. What difference does a candidate s perceived competence, integrity, or personal magnetism make? We find that these characteristics, variously measured, almost always matter, sometimes considerably so. Chirac s having the right stuff for leadership, at least in the mind of the electorate, definitely helped him carry the day in 1988 and For Jospin, in 2002, his image as someone who keeps is promises, brought him closer (but not close enough) to victory. The candidate s TV face appears pivotal, across candidates and races. These are bits and pieces of evidence. But considering candidate traits taken together, they stand as a decisive force. Without these trait bundles in the thermometer models, they would manage the explanation of vote choice much less well (as the large R-squared increments from their addition make clear). Looking at these overall 16

18 differences in candidate appeal, from contest to contest, they could be used to account for Mitterrand s easy victory in 1988, and Royal s easy defeat in Be that as it may, the extensive analysis here demonstrates leader image has been too long neglected in explanations of French presidential vote choice. Hopefully, future surveys will contain systematic batteries of items measuring candidate competence, integrity, and personality. 17

19 FOOTNOTES 1 One concern is that feeling thermometers are mere surrogates for vote choice. This does not appear to be the case for these French data. If the thermometers were actually just another measure of vote intention, then we would expect the thermometer scores to be highly correlated with each other. For example, if my high Jospin thermometer score merely registers my intention to vote for Jospin, then that thermometer score should be highly (negatively) correlated with my thermometer scores on the right-wing candidates (assuming that those scores are merely registering my intention of not voting for these right-wing candidates). However when we actually examine the inter-correlations of the thermometer scores, we are struck by how low they are, the average r being equal to.40,.35,.18, and.25 for 1988, 1995, 2002, and 2007, respectively. The highest correlations for those years are: Chirac-Barre (.58) and Chirac-Mitterand (-.52) for 1988, Chirac-Balladur (.70) and Chirac-Jospin (-40) for 1995, Chirac-Bayrou (.43) and Jospin-Le Pen (-.29) for 2002, and Royal-Sarkozy (-.43) and Royal-Bayrou (.36) for The lowest correlations are Le Pen-Barre (.17) for 1988, Le Pen-Chirac (.02) for 1995, Le Pen-Chirac (.17) for 2002, and Le Pen- Bayrou (-.10) for Overall, these results suggest that these thermometers are measuring something different than vote choice. They are measuring the psychological appeal of the candidate, largely independent of whether one might vote for that candidate. 2 The question of whether to include party identification and ideological identification together, as independent variables in the models under study, is moot. While the latter is consistently measured in each of the four surveys, party identification is either not measured, or not consistently measured. Consequently, estimates, when they are available at all, vary wildly. For example, in 1988 the party identification measure (open-ended) estimates 71 percent identifiers; in 1995 an open-ended measures estimates 47 percent, a closed-ended measure 87 percent; in 2002 and 2007 the party identification measure (only closed-ended) estimates 94 and 86 percent, respectively. The issue of how best to measure party identification for the French case has a long and controversial history. [See especially Converse and Pierce, 1986; Evans, 2002; Pierce, 1995; Fleury and Lewis-Beck, 1993; Lewis-Beck, 1996; Lewis-Beck and Chlarson, 2002]. In the 1995 French National Election Study, party identification, uniquely, was measured in multiple ways, along with ideological identification and candidate feeling thermometers. In an analysis of these data, including party identification, ideological identification, and the feeling thermometers all together in an explanation of presidential vote choice, Lewis-Beck (1997) concludes that collinearity levels are severe enough to prevent the reliable disentangling of their different effects. 3 Multinomial logistic regression was applied to the first-round vote choice, in order to estimate more strictly the equation specification of Table 2A. The dependent variable was categorical, with the categories one of the four leading candidates, plus extreme left or extreme right, against a baseline choice of the incumbent candidate. The models generally fit the data well. The pseudo-r-squared are as follows: 2007 =.54, 2002 =.42, 1995 =.45, 1988 =.55. The baseline category is as follows: Chirac (1988, 1995, 2002) and Sarkozy (2007). To organize the presentation of results, we focus only on the significant thermometer coefficients. In 2007, Sarkozy stands as the baseline choice. Predicting that choice against a choice for Royal (controlling on the other choices, and the other independent variables, as multinominal logit does) yields absolute thermometer coefficients with essentially equivalent absolute values: for Sarkozy = -9.4, for Royal = 9.9. This is much the same result, in terms of equivalence, that the binomial logistic regression yields on the second-round (see Table 2B). This is not surprising since, statistically speaking, the multinomial logit set-up on the first-round choices makes it functionally equivalent to a second-round ballot choice, i.e., the vote impact of the feeling thermometers for the two final candidates (holding choice on the other candidates constant). The two absolute thermometer coefficients for the two leading candidates, running comparable multinomial logits for the 2002, 1995, and 1988 contests are likewise not too far apart in value: 2002 (Chirac = -8.2, Jospin = 9.7); 1995 (Chirac = -7.8, Jospin 7.1); 1998 (Chirac = - 9.7, Mitterrand = 10.2). The multinomial results also admit of other useful comparisons. In 2007, when the choice was Sarkozy v. Bayrou, the latter lost votes to those sympathetic with Sarkozy (b = -9.0) and, to a much lesser extent, with Royal (-1.0). With the choice Sarkozy v. Le Pen, the latter lost votes from Sarkozy sympathizers, but no one else. In 2002, with the choice Chirac v. Bayrou, the pattern was dissimilar to Bayrou lost votes to Chirac sympathizers (b = -6.9) but gained votes from Jospin supporters (b = 18

20 1.6). With the choice Chirac v. Le Pen, the pattern repeats 2007, Le Pen losing support from Chirac sympathizers (b = -7.7). In 1995, with the choice Chirac v. Balladur, the latter loses heavily from Chirac sympathizers (b = -10.5). The Le Pen pattern again repeats, with losses from Chirac sympathizers (-7.8). In 1988, with the choice Chirac v. Barre, the latter loses many votes to Chirac supporters (-7.9). The Le Pen pattern here is consistent, with him losing heavily from Chirac supporters (-9.7). These multinomial logit results are available upon request. 4 The standard deviation of these thermometer scores, across the four candidates and four races, falls within a narrow range from.22 (Barre in 1988, Bayrou in 2002) to.31 (Sarkozy in 2007). The average standard deviation, for the sixteen observations, is.26. Thus, setting the change in X at of a one-half a standard deviation seems reasonable. 5 Multinomial logistic regression was also applied to the first-round legislative elections for 2002 and 2007 (data not available for 1988 and 1995; UMP = baseline; full specification). The models fit the data relatively well (pseudo-r-squared =.38 for 2002 and.46 for 2007). The results reported below for the thermometer scores (significant coefficients only) suggest that the impact of leaders is somewhat smaller for legislative elections than presidential contests, at least for moderate parties (see note 2) PS UDF(MD) FN Th. Sarkozy Th. Royal 5.5 Th. Bayrou 6.9 Th. Le Pen PS UDF FN Th. Chirac Th. Jospin 7.5 Th. Bayrou 6.0 Th. Le Pen

21 REFERENCES Bean, Clive, and Anthony Mughan Leadership Effects in Parliamentary Elections. American Political Science Review 83: Bélanger, Éric and Michael S. Lewis-Beck «French National Elections: Democratic Disequilibrium and the 2007 Forecasts». Paper presented at the American Political Science Association meetings, Chicago, September Bélanger, Eric, Michael S. Lewis-Beck, Jean Chiche, and Vincent Tiberj Party, Ideology, and Vote Intentions: Dynmaics from the 2002 French Electoral Panel. Political Research Quarterly 59: Boy, Daniel and Nonna Mayer «The Changing French Voter.» in The French Voter Decides, Daniel Boy and Nonna Mayer (eds.), Ann Arbor: University of Michigan Press: Boy, Daniel and Nonna Mayer «Que restent-ils des variables lourdes?» In L électeur a ses raisons, Daniel Boy and Nonna Mayer (eds.), Paris: Presses de Science Po : Campbell, Angus, Philp E. Converse, Warren E. Miller, and Donald E. Stokes The American Voter. New York: Wiley and sons. Cautrès, Bruno «Old Wine in New Bottles? New Wine in Old Bottles? Class, Religion and Vote in the French Electorate. The 2002 Elections in Time Perspective.» In The French Voter: Before and After the 2002 Elections, Michael S. Lewis-Beck (ed.), Hampshire, England: Palgrave Macmillan: Cautrès, Bruno and Anthony Heath «Déclin du vote de classe? Une analyse comparative franco-britannique.» Revue internationale de politique comparée 3(3) : Cautrès, Bruno, and Nonna Mayer, eds Le Nouveau Désordre Electoral : Les Leçons du 21 Avril Presses de Sciences Po : Paris. Clarke, Harold D., David Sanders, Marianne C. Stewart and Paul Whitely Political Choice in Britain. Oxford: Oxford University Press. Converse, Philip E., and George Dupeux «De Gaulle and Eisenhower : The Public Image of the Victorious General.» In Angus Campbell et al., Elections and the Political Order, chapter 15, New York: Wiley and Sons. Converse, Philip E. and Roy C. Pierce Political Representation in France. Cambridge: Harvard University Press. 20

22 Crewe, Ivor, and Anthony King Did Major Win? Did Kinnock Lose? Leadership Effects in the 1992 Election. In Labour s Last Chance? The 1992 Election and Beyond, Anthony Heath et al., eds., Adershot: Dartmouth, England. Curtice, John, and Soren Holmberg Leadership. In The European Voter, ed., J. Thomassen. Oxford: Oxford University Press. Evans, Jocelyn A. J «Ideology and Party Identification: A Normalisation of French Voting Anchors?», In The French Voter: Before and After the 2002 Elections, Michael S. Lewis-Beck (ed.), Hampshire, England: Palgrave Macmillan: Fiorina, Morris Retrospective Voting in American National Elections. New Haven: Yale University Press. Fleury Christopher J. and Michael Lewis-Beck «Anchoring the French Voter: Ideology versus Party.» The Journal of Politics 55(4): Haegel, Florence «Partisan Ties.» In The French Voter Decides, eds. Daniel Boy and Nonna Mayer. Ann Arbor: University of Michigan Press: Kessel, John H Views of the Voters. In Models of Voting in Presidential Elections, eds., Herbert F. Weisberg and Clyde Wilcox. Stanford, CA: Stanford University Press, Kinder, Donald Presidential Character Revisited. In Political Cognition, eds. Richard R. Lau and David O. Sears. Hillside, NJ: Earlbaum, King, Anthony(ed) Leaders Personalities and the Outcomes of Democratic Elections. Oxford: Oxford University Press. Lewis-Beck, Michael S «Economics and the French Voter: A Microanalysis.» Public Opinion Quarterly 47(4): Lewis-Beck, Michael S Economics and Elections. Ann Arbor: Michigan University Press. Lewis-Beck, Michael S «The French Voter: Steadfast or Changing?» In The French Voter Decides, eds. Daniel Boy and Nonna Mayer. Ann Arbor: University of Michigan Press: Lewis-Beck, Michael S «Cross-National Election Surveys: A French Pre-Test.» Electoral Studies 15(4): Lewis-Beck, Michael S Le vote du porte-monnaie en question. In L Électeur a ses raisons, eds. Daniel Boy and Nonna Mayer, Presses de la Fondation nationale de science politiques,