CAIRN-INT.INFO : International Edition

1Election results do not necessarily reflect voters’ preferences. Condorcet was the first to argue this with his famous paradox, which is routinely tested in studies regarding voting patterns. [1] Here, however, we would like to examine a different discrepancy: that which exists between voter preferences for a particular candidate or party and the choice that they ultimately make at the ballot box. [2] Scholarly interest in this discrepancy between stated political preferences and voting choices has only recently been piqued, at a time when “cleavage votes” based on long-term social and religious affiliations have begun to carry less weight. The very meaning of electoral choice has changed, votes of support giving way to “negative” votes seeking to eliminate rather than select candidates.

2We shall demonstrate this phenomenon by taking the example of France’s 2012 presidential election. Our work was based on the answers provided regarding “voting probabilities” recorded in five pre-electoral surveys (from July 2011 to March 2012), conducted within the TriÉlec network’s research programme. [3] This data is particularly well suited to reconstituting the shifts in preferences and the crystallisation of choices over the course of this long election period.

3In the United States as in France, the seminal works on voting, in terms of electoral geography, [4] social and religious cleavages, [5] partisan identification [6] and political and institutional traditions [7] have painted voting choices as stemming from a permanent and exclusive adherence to a party or a camp that tends to intensify over a voter’s lifetime. Over the course of the past 30 years or so, however, numerous studies have highlighted the shortcomings of such interpretative frameworks. In this respect, France is no different than other Western democratic societies. A growing number of voters hesitate until the final moment, not knowing for whom to vote – even in the case of presidential elections, the major voting occasion of French politics. For example, in 1988 only 11% of voters said that they hesitated “until the last moment”, but in 2007, 22% admitted such ambivalence and in 2012, according to the TNS-Sofres TriÉlec exit poll, 19% of voters had made a decision during the last week of campaigning, and half of those had made it in the two days prior to the election. These levels of hesitation also mean greater electoral mobility. Voter participation has become irregular, fluctuating depending on the importance that voters bestow upon specific elections. [8] As a consequence of this electoral instability, some parties obtained double-digit percentages during one election and then found themselves below 5% in another. Faced with new empirical data such as this, voting models and academic interpretations of voters have been forced to change. A shift has occurred from cleavage voting to “issue-based voting”. [9] Many have declared that parties are now “without partisans”, [10] since they struggle to conserve their identity “in the eyes of the electorate”. Voters have begun to be perceived as reasoning actors, [11] updating their preferences during each election and selecting among different options: this is a significant departure from earlier perspectives, which depicted voters as faithful to their beliefs and political tendencies. For example, partisan identification in the United States is now deemed to be a “running tally”. [12] With this change in outlook, political campaigns – which generally were seen to have a minimal effect on voters [13] – have become a topic of study again, even if only through the effects of agendas, priming and framing. [14] Have we truly entered the age of the “circumstantial” voter, who is increasingly sensitive to each election’s context and inner logic?

4In order to understand how today’s voters make decisions, we must more than ever pay attention to their stated electoral preferences. Do some or all voters conform to the traditional model which emphasises positive preferences for a single party or candidate? Or do voters increasingly express multiple preferences? And in that case, how do they decide among their preferences? What kind of logic underlies their choice? This article will explore these questions with regard to 2012’s presidential election, following four different avenues of research that each shine a new light on this election and the voting preferences expressed therein. The first line of research investigates the traditional concept of the electorate, while demonstrating the value of an analysis in terms of candidates’ electoral potential. This potential is naturally constituted by long-time supporters, but also by more erratic voters whose political preferences are less marked (or at least not exclusive) and who make up their minds as the campaign unfolds. A longitudinal analysis allows us to see, for each candidate, how his or her potential evolves and finally, how s/he manages (or not) to convert this potential into actual votes. Our second line of research explores voter preferences. It addresses the electoral “space of possibilities”, [15] that is to say the entire group of candidates for which individuals could see themselves voting, thus addressing both issues of negative preference (demonstrated by Helena Catt [16]) and multi-preferences. Our third research avenue looks at what voting probabilities can teach us about the ideological reconfigurations currently taking place in France. Voters’ multi-preferences for candidates are highly structured, which has allowed us to establish a typology. The latter is very informative, revealing the tensions, complementarities and rivalries between various political oppositions, be they between Nicolas Sarkozy and Marine Le Pen or between Jean-Luc Mélenchon and François Hollande. A new vision of the political sphere has thus emerged: whereas votes (and voting intentions) divide and highlight cleavages, voting probabilities demonstrate how much various candidates are liked and disliked, and consequently the interactive dynamics of political forces. Finally, our fourth line of investigation cuts across all the aforementioned issues by questioning the sociological and political basis of these multi-preferences. Are they due to the political incompetence of ordinary citizens? In the past, studies have shown that a strong interest in politics correlates to a strong attachment to a particular political party. [17] As such, if some voters are more volatile, have less exclusive preferences and make choices by default, could this not be the result of their ignorance of the differences between parties and candidates? In this perspective, hesitant choices would be symptomatic of a weak relationship to politics, engendered by a lack of interest and knowledge. On the other hand, these new “political affinities” could be explained by a change in the relationship between citizens and the political sphere, provoked by cognitive mobilisation. [18] With a rise in education levels, the development of mass media and generational renewal, today’s voters apparently need parties less and less in order to make up their minds. [19] The preferences observed would thus be the manifestation of a more autonomous electorate, no longer dominated by politics. [20]

5This article presents a preliminary overview of the research conducted on these important questions using the data provided by the TNS-Sofres TriÉlec barometer. We shall start by describing the history of the voting probability tool, as well as its guidelines and research value. Then we shall analyse the electoral potential of each candidate and what its evolution tells us about the political dynamics occurring between July 2011 and March 2012. Next, voters’ multi-preferences will be analysed; that is to say, all the candidates for whom a given voter would consider voting. Here the goal is to identify and examine the logic underpinning the “space of electoral possibilities”. Finally, the fourth part of this article will isolate six main combinations of voting preferences; analysing them will reveal the political dynamics latent in the ongoing reconfiguration of French political life.

Voting probability: the history and guidelines of an indicator

How to measure political preferences: an overview

6Questions evaluating voting probabilities are relatively recent in the history of electoral polls. For a long time, questionnaires measured voter preferences by asking respondents to place themselves on an ideological spectrum (liberal/conservative, left/right), to say who they voted for and/or with which party they maintained ties (whether through identification or mere proximity). These questions overlook essential pieces of information: they do not allow us to determine voter hesitation between several components of the political offer, the degree of their preferences or the rationale behind their choices (positive, or by adherence – the best candidate; or negative, by elimination – the least bad candidate).

7Starting in the 1950s, American electoral polls included several open-ended questions asking voters what they liked or disliked about each party and candidate. This survey tool was costly both in terms of time and processing, and it also had much less impact on research than partisan identification. [21] More importantly, it quickly became apparent that this tool could not really be exported to countries with multi-party systems. In 1968, still in the US, the first political thermometers appeared measuring sympathy for candidates and parties; then in 1980, voters were asked to select the emotional response each candidate provoked in them (angry, hopeful, proud, afraid). [22] That time around, both positive and negative choices were clearly taken into account. Such measures were gradually included in most national voting polls. But here we must ask, what is effectively being measured? Do voters react to a candidate’s personality, or to the ideas that s/he embodies?

8In this kind of landscape, questions of voting probability seem like a particularly heuristic compromise to allow for an analysis of voters’ political preferences. The interviewees are asked to measure the probability of their voting for various parties and/or candidates. Such questions appeared in the Netherlands in 1982 and were repeated in all Dutch electoral surveys thereafter. From 1989, when they were incorporated into the European Election Studies (EES), [23] they also gained a comparative dimension, and were then included in national surveys in Great Britain, Ireland, Spain and Germany. The inclusion of these kinds of questions in French polls was much more recent: they first appeared in the RAPFI surveys in 2005. [24] They were also used in the Formation du jugement politique survey conducted by PACTE in Isère in February and April 2007, [25] as well as the Baromètre politique français in 2006-2007 established by Cevipof and the Ministry of the Interior. [26]

9The questions regarding voting probability contained in 2011-2012’s TriÉlec poll were worded as follows and adapted from the EES questions: “What are the chances that you might vote for the following candidates in 2012’s presidential election? 0 means that there is no chance of you voting for him/her, 10 means there is a very strong chance of you voting for him/her; you can nuance your response using the numbers in between”. Candidates were listed in random order. In July 2011, the list contained 21 individuals: all the socialist candidates in the primaries, Eva Joly and Nicolas Hulot, all the “major” candidates who had declared their campaigns or who were expected to run, as well as the different potential centre-right and right-wing candidates. This list became shorter from the second wave of polling in late October (after the socialist primaries), and subsequently as candidates dropped out. The fifth wave in March 2012, only contained candidates who were actually running. In addition to the standard “I prefer not to answer” and “I don’t know” options, a specific type of answer was explicitly available if the interviewee was not familiar with a candidate.

Initial qualitative exploration

10When they are used in a qualitative context, voting probability questions quickly reveal their superiority over traditional questions concerning vote intentions. All too often polls lack a systematic understanding of what different modes of questioning “do” to voters, despite the fact that this kind of research can prove to be extremely illuminating. [27] It was in this perspective that one of PACTE’s teams asked these questions to the members of a qualitative panel online right before the first round of the presidential election. [28]

11In Nathalie or Alain’s case, for example, the choice was simple. The former expressed only a 10 (highest probability of voting) for Marine Le Pen, while the latter marked the same in favour of Nicolas Sarkozy.

12Nathalie, 42 years old, executive secretary: “Yes I will vote. For Le Pen, I’m certain. Because her ideas are the closest to mine. She seems sincere to me and like she could make good decisions. My mind has been made up for a long time. In fact, I’ve been voting for the Front National for over 10 years now”.

13Alain, 58 years old, self-employed: “I will go vote, that’s for sure, I never heard any arguments in this campaign that would made me change my mind. My choice is because now I live better than I did five years ago, thanks to the RSA and the self-employed status (auto-entrepreneur). Thanks SARKO”.

14In these two examples we see “traditional” voters. Both stated political preferences that were well established before the start of the campaign. The electoral season served only to strengthen convictions that were already firm, in line with the theory of campaign minimal effects. It is likely that all the political information with which these two individuals were confronted were subject to well-known practices of “selective perception”. For these two voters, there was no need to test or imagine another political option, and thus one question regarding vote intentions would be sufficient: they would no doubt have expressed a firm and positive voting probability and stated that they had made their choice a long time ago.

15But a single question regarding voting intentions would not have been so helpful in the cases of Sylvie and Élodie. The former gave Mélenchon a probability of 9, Joly an 8 and Hollande a 7. The latter was even more undecided, giving only a 3 to Mélenchon and a 5 to Hollande.

16Sylvie, 50 years old, social worker: “After having hesitated for a long time between Mélenchon and Joly (whose ideas feel closer to mine), I finally opted for Mélenchon. Like you suggested, I reread their literature. There were a lot of points in common, but Mélenchon’s language was harder-hitting. What finally decided things for me was wanting to have a greater impact, when I found out that my family (husband and children) were voting for Mélenchon, I decided to follow suit…”.

17Élodie, 30 years old, special education teacher: “Two days before the election, the only thing that I’m sure of is that I will go vote (even in this terrible weather), but on the other hand, I still haven’t made up my mind, I know who I absolutely won’t vote for, but who will ‘get my vote’, I still don’t know. No one stands out… I’m sorry, but I think my decision will probably be made on the day”.

18Both of these voters were racked with indecision, although in two very different situations. In Sylvie’s case, her hesitation could be explained by sharing ideas with multiple candidates, in a positive way. Within this space of possibilities, she made her choice by yielding to family pressure. Élodie’s indecision stemmed from negative reasoning, on the other hand. No candidate really won her support, without that meaning that she acted incoherently (the only two candidates to emerge were left-wing ones), or that she was unable to rank candidates for president. On the contrary, she was much more aware of who she would not vote for than who she would support. In any case, neither Élodie nor Sylvie’s hesitation was the product of a lack of political knowledge or interest in public affairs. But what could a traditional question about voting intentions have revealed in these cases? It would likely have attributed a vote for Hollande to the former and one for Mélenchon to the latter, while completely obscuring the complexity of a decision-making process that had nothing in common with Alain and Nathalie’s “traditional” voting patterns. Only voting probability can allow us to grasp the true complexity of electoral choices.

Accessible questions

19The questions asked over the telephone were easily understood and accepted: 93% of the 5,033 people called during all five waves of polling agreed to respond for all the candidates listed. But did they play the game and use the entire probability scale for the various candidates? One could, for example, pretend to answer and give every candidate a 5, as the central value of any scale often functions as a safe answer to hide behind. If those polled truly played the game and agreed to rank their preferences, the same level of probability should not (or should rarely) have been used for multiple candidates. Specifically, a 10 – which represents the firmest conviction – should usually only be used once. And likewise, 0 should prove to be the most common answer, as the polarising nature of voting orientations means that in expressing a preference for one or several candidates, others are necessarily eliminated.

20And in effect, 0 was far and away the most common probability expressed over the course of the five surveys: a little more than a third of those polled used 0 for at least three candidates, but only 2% of voters used it for all eight candidates studied. Conversely, just over one out of ten individuals did not eliminate any candidates from their voting possibilities. These two extreme responses were rare and surprisingly stable over time. On the other hand, the proportion of voters who eliminated between three and seven candidates increased from 59% in July to 67% in October and 69% finally in March. The median position was used often, by 54% of respondents. But 27% of those who used 5s gave that score to only one candidate, while 12% gave it to three or more, both proportions remaining very consistent over time. About 16% of those polled expressed a 10 in July 2011, compared to 34% in March 2012, which indicates a polarisation of choices over the course of the campaign. Nine times out of ten, a single candidate received this maximum score. The other probabilities were cited much more infrequently: 70% to 91% of those answering never used them, and when they did, it was very often for one candidate only.

21In short, the respondents behaved much as expected. They eliminated a significant number of candidates. They did not abuse the safe answer. They used all the gradients of the probability scale to express their preferences regarding each of the candidates they had not eliminated. Those polled likewise sparingly awarded the higher probabilities to candidates. Finally, the use of this scale changed over the course of the campaign, as electoral choices became firmer. This analysis of the data allows us to consider voting probabilities as valid indicators of the electoral potential of each candidate.

The dynamics of electoral potential

22The questions were asked five times: in July, October and December 2011, and in February and late March 2012. The TriÉlec polls thus allowed us to follow the evolution of the electoral potential of a number of different figures, several of whom ultimately became official presidential candidates. These observation points were chosen specifically to better follow campaign developments. July was thus an ideal starting point before the real onset of the campaign. October marked Hollande’s victory in the socialist primary, while February was the start of the incumbent president’s campaign and March saw the fallout from the Toulouse shootings (“l’affaire Merah”). This distribution over time thus facilitated a better understanding of some of the campaign’s effects via voting probabilities. We were hence able to follow the dynamics propelling each of the major candidates.

23We can also analyse the evolution over time of electoral potential in a variety of ways. Either we consider the answers given as indicators of support for a candidate: they are then treated as continuous variables. Or we consider them to be categorical variables, based on the principle that the scores given also express thresholds and various attitudes towards the candidate in question: rejection, neutrality or support, with variations of intensity in the positive or negative sentiment (absolute rejection, 0; excluded as a voting option, 1 to 4; neutrality, 5; possible voting choice, 6–7; and support, 8 or higher). [29]

Figure 1

The evolution of voting probability averages for eight candidates over the whole period studied

Figure 1

The evolution of voting probability averages for eight candidates over the whole period studied

How to read this graph: The average of the voting probabilities attributed to the different candidates between July 2011 and March 2012.
Sources: TriÉlec TNS-Sofres polls.

24All the candidates obtained relatively weak averages, ranging over the course of the five surveys from 0.90 for Nathalie Arthaud in July 2011 to 5.01 for Hollande just after he won the socialist primary. These averages correspond to very different types of answer profiles – some highly indicative of cleavages (lots of zeros and strong probabilities), others more consensual with few zeros, but also lots of average probabilities. In a highly polarised competition, these overall weak scores of support are not surprising, as even a candidate who was supported by half of all voters would see his or her average decrease as a result of being rejected by his adversaries (who would score him or her as a 0 probability). In order to better evaluate the degree of support garnered by each candidate, the averages must be recalculated; for example, in terms of the hope of seeing that person win the election. In the February and March polls, [30] the highest average went to Le Pen among voters hoping for her victory (8.78) and Mélenchon among his supporters (8.32). This average went down to 7.96 for Hollande, 7.82 for Sarkozy and 7.29 for Bayrou, in terms of their respective supporters. So the first significant result that emerges is that a minority of voters selected small-party, protest candidates, but their choices were the most positive and the most strongly expressed. Those supporting one of the two main presidential candidates averaged below an 8, which indicates a more reserved kind of support and a certain degree of scepticism even with regard to the candidate they hoped to see win.

25Even when calculated with the whole data set, these averages remain indicators of the overall electoral potential of candidates, thus allowing us to rank them. The ranking obtained is very different from that produced by the uninominal ballot of 22 April. [31] Hollande had the greatest potential (4.23 in March), ahead of Sarkozy (3.72). But Bayrou (3.19) stole the third-place spot away from Le Pen (2.15), who was likewise surpassed by Mélenchon (3.13). Bayrou’s third-place finish according to this metric can be explained by the weak distribution of probabilities concerning him: high probabilities were rare, but zeros did not surpass 39% over all five polls. [32] These answers reflect the relatively consensus-based nature of Bayrou’s MoDem party, as opposed to Le Pen, who was rejected by two-thirds of all respondents in every poll. Shifts in candidate support or electoral potential are the first indication of political dynamics and they naturally vary significantly among candidates: a consistent decrease for Joly (her average probability went from 2.67 to 1.30), a slight downward trend for Poutou, stability followed by a very slight bump for Pen at the end of the campaign (from 2.09 to 2.15 in March), an upwards trend from July (2.65) to February (3.43) for Bayrou, then a dip in March (3.19), sharp ups and downs for Hollande, who in March barely matched his score from July (4.23 compared to 4.24); finally, a consistent upward trend for Sarkozy, who went from 2.81 in July to 3.32 in February and ultimately 3.72 in March.

26A “categorical” analysis completes and nuances our analysis of averages by allowing the electoral “hard cores” of various candidates to be identified; we could likewise describe the nature of the rejections that occurred and account for the effects of celebrity or notoriety.

Figure 2a

Changes in voting probabilities for left-wing candidates (in %)

Figure 2a

Changes in voting probabilities for left-wing candidates (in %)

Figure 2b

Changes in voting probabilities for right-wing and centrist candidates (in %)

Figure 2b

Changes in voting probabilities for right-wing and centrist candidates (in %)

Sources: TriÉlec TNS-Sofres polls.

27The main result from this analysis is that the campaign helped to disseminate information, in particular about “minority” candidates. Questions about voting intentions assume that citizens are familiar with all their choices – and yet this was not the case here, especially at the beginning of the campaign period. Hence, in July 2011, 40% of voters did not know Poutou, 25% did not know Dupont-Aignan, 21% Arthaud, 15% Morin and 10% Mélenchon. The difference in recognition between the various candidates is connected to the issue of political competence. The more educated a voter, the greater his/her chances of knowing all the different candidates (or at least of giving the impression of knowing them), regardless of when the poll was carried out. 61% of those with no education or only a primary education did not know Poutou in July and 32% still did not know him in October. For these same dates, the other candidates’ percentages were as follows: 54% and 29% for Dupont-Aignan, 28% and 16% for Arthaud. On the other hand, among university graduates these percentages were much lower: 31% and 15% for Poutou, 25% and 18% for Dupont-Aignan, and 21% and 11% for Arthaud.

28This traditional social rationale of political incompetence was nevertheless rapidly compensated for by the catching-up effects of the electoral campaign. The proliferation of political messages and information in the media allowed even those usually the most distant from political life – or least endowed with cultural capital – to catch up to their more “sophisticated” brethren. For this presidential election, this phenomenon occurred relatively early on, since as soon as the socialist primary was over, most of the gap between those with few and those with many educational qualifications had evaporated. In July, the ignorance gap between the least educated (primary school at most) and the most educated (higher education) was 37 points higher for Dupont-Aignan, 30 points higher for Poutou, 16 points higher for Morin, 15 points higher for Mélenchon and 12 points higher for Arthaud. In October, these differences were only 15, 16, 7 and 8 points, respectively. In March, they were all between 10 and 5 points. [33]

29A detailed analysis of voting probabilities clearly revealed the strengths and weaknesses of each of the candidates, as well as the shift in their respective audiences over the course of the campaign. For example, the rejection of Eva Joly was expressed a number of different ways. The total exclusion of a Green Party vote increased by 19 points between July and late March, which placed Joly at 59%, close to Le Pen or minority far-left candidates. During the same period of time, her potential electorate (voters who rated her above 5 in terms of voting probability) shrank by two-thirds. In July, before she was officially confirmed as a candidate, 10% of interviewees stated their probability of voting for her as a 6 or a 7; 4% as an 8 or higher. In October, the first group was only 7%, in December 5%, and ultimately 3% in March. The second group remained more or less consistent between July and February, but dropped to 2% in March.

30Voting patterns do not respect the “communicating vessels” principle. This is another valuable insight contributed by the voting probabilities method. A drop in one does not necessarily entail a rise in another, contrary to what is assumed by traditional voting intention questions. Hence, Joly’s losses did not benefit Mélenchon for example, who only saw significant changes during February and March, when the proportion of voters who ranked him higher than 7 went from 6% to 12% (for 6 or 7, the proportion went from 6% to 9%). In July, his potential voters were at 7%, shooting up to 21% at the end of March.

31For Le Pen, few changes could be observed over the course of the campaign, even in late March after the Toulouse shootings. This can be explained by the fact that the Front National’s president is extremely well known and preferences regarding her positions have long been crystallised, both for and against. Moreover, her potential electorate was very close to the results she obtained on 22 April: according to the polls, between 8% and 11% of those polled gave Le Pen an 8 or higher, while between 6% and 4% gave her a 6 or 7. At the end of March, that would have represented a potential electorate of about 14%. One out of every ten voters was thus a hard-line supporter of Le Pen, while a non-negligible number of others were sympathisers.

32For Bayrou, on the other hand, responses from the different polls showed a significant evolution. The beginning of the campaign was favourable to him, both in terms of a proportional drop in negative responses to him and an increase in the size of his potential electorate. In total, in July 2011, 71% of voters gave Bayrou a probability rating lower than 5 out of 10; in the spring of 2012, this figure had dropped to 51%. With regard to his potential electorate, it was primarily the “inner circle” (rankings above 7) that evolved during the campaign, going from 4% to 9%, while “sympathisers” (6s or 7s) represented about one-tenth of the sample population.

33Lastly, the two future finalists exhibited two very different profiles and evolutions. The incumbent president provoked a strong degree of rejection, if we measure the proportion of voters who completely excluded him as a voting possibility: 47% in July, 40% still in March; whereas the future president seemed to be the most widely approved candidate of all, even if he saw a growing degree of rejection over the course of the campaign: from 24% in July to 29% in March. With regard to “inner circle” voters – those ranking Hollande higher than 7 – they were consistently more numerous than those in Sarkozy’s base, until March when these two groups reached the same levels (around 22%).

34For a large number of voters, attitudes with regard to the major candidates seemed already to be quite consistent at the beginning of the campaign, whether positively or negatively. Many knew who they rejected and of whom they approved. For these voters, the campaign was in fact another way of strengthening their convictions. Following their voting probabilities also produced an unexpected result. Common sense would suggest that the campaign would have convinced voters to select a specific candidate, thus creating a positive relationship. In fact, this process of positive influence only occurred for Mélenchon on the left and Sarkozy on the right, between February and March. On the other hand, the campaign seriously penalised Eva Joly and other small-party candidates, who transitioned from anonymity into rejection, and it did not particularly help the socialist candidate. Therefore, it would be more apt to talk about campaigns in the plural, since the dynamics were significantly different from one candidate to another.

“The space of electoral possibilities”

35Until now, we have used voting probabilities separately. Taken all together, they allow us to understand the space of electoral possibilities experienced by voters: that is to say, all the candidates that a given voter might conceive of voting for. With this approach, we can identify not only “negative” voters, who, for lack of an attractive enough candidate, were forced to choose the lesser evil, but also voters with multiple preferences. The latter expressed true affinities for more than one candidate. Both groups of voters have generally been little studied, despite the fact that they would permit a better understanding of the logic behind a particular election (in this case, 2012’s), but also contemporary ideological and political reconfigurations.

Negativity on the rise and the persistence of multiple preferences

36In order to establish these two groups, we created an indicator that counted the number of times the interviewee gave a voting probability score higher than 5. Probability scores lower than 5 could be interpreted as a certain degree of distance from a candidate, at best, or rejection at worst. If we compare this scoring method with the school grading system, a 5 would thus be an average score, just passing. On the other hand, giving a score of 6 or higher expressed a preference for a certain candidate (or at least satisfaction), and possibly indicated that the individual was a sympathiser or even a true supporter. In order to establish our indicator, we only used the candidates that had officially declared their candidacy. However, Cheminade was only tested during the final poll. In order to consistently have the same number of candidates in our indicator, we thus included Morin in the first four polls. Consequently, the indicator always functioned on the basis of ten presidential candidates.

Table 1

The space of electoral possibilities: a comparison between 2012, 2004, 2007 and 2009 (in %)

Table 1
P1-July P2-Oct. P3-Dec. P4-Feb. P5-March Reminder: EES-2004 Reminder: EES-2009 Reminder: Out of these candidates, whom would you vote for? (Cevipof 2007) None 30 26.5 31 27.5 23.5 16.5 18.5 12 One candidate 38 36 35.5 36 41 28 32 34 or party Two 19.5 22 21.5 22.5 20.5 32 23.5 27.5 Three 8 9 7.5 9.5 10.5 16 14.5 16 Four or more 4.5 6.5 4.5 4.5 4.5 7.5 11.5 10.5

The space of electoral possibilities: a comparison between 2012, 2004, 2007 and 2009 (in %)

How to read this table: In July, 30% of those polled did not have definite electoral preferences; that is to say, probabilities higher than 5. In March 2012, this figure was 23.5%. For both these dates, 19.5% and 20.5% respectively expressed preferences for two candidates.
Sources: TriÉlec TNS-Sofres polls, EES 2004 and 2009, Cevipof pre-electoral survey 2007.

37Multi-preference voting is a growing phenomenon. 35.5% and 41% of voters envisioned only voting for a single candidate, whereas 20% of them could see voting for one of two candidates and 8% to 10% saw the possibility of voting for three different candidates. This phenomenon was not limited to this last presidential election, as we can see from the results of the last two European Electoral Studies that used the same measure. Voters who declared only one party as a possible choice represented 28% of the electorate in 2004 and 32% in 2009. The results obtained in 2012 were equally convergent with those obtained by 2007’s pre-electoral survey, where individuals were directly asked for which candidates they could see themselves voting: only 34% of them cited a single candidate, even though this election was touted as marking the re-bipolarisation of the French party system. [34] It is equally noteworthy that multi-preferences for three candidates or more remained relatively stable throughout this whole period, ranging between 12.5% and 15%.

38The 2012 campaign showed continuity with regard to the expression of multiple preferences but negativity, on the other hand, was on the rise. The percentage of voters who never gave a candidate a score higher than 5 reached its apex during the last presidential election. The level of negativity fluctuated between 26.5% and 31% during the first four TriÉlec polls in 2011 –2012, and although it dropped slightly in March it remained at 23.5%, or 5 points higher than in 2009 and 7 points higher than in 2004. [35] This indicator thus confirms what other studies have already shown: French citizens were invested in 2012’s presidential election (unlike 2002’s), but it took place in a context that was much less polarised than in 2007, where the two finalists both embodied a form of political renewal, each in their own way. In fact, this could also be observed in the most traditional indicators of politicisation: 73% of voters polled in March 2012 said that they were very or quite interested in the campaign, but only 40% of them felt that “the campaign addressed the true problems facing French citizens” and 44% that it “allowed for the different candidates’ positions to be known”. [36]

Incompetence or a shift in relationships to political choices?

39What logic can explain these phenomena? Is it related to levels of political competence? One could argue that more sophisticated voters expressed preferences, possibly even a single preference, because they were the most capable of understanding the different and contrasting positions held by the various candidates. Conversely, those voters who were the furthest removed from politics might be more likely to reject political alternatives. Multiple preferences might also be the result of voters’ inability to choose, and thus possibly again indicate low levels of political competence. Likewise, is there also a properly political logic underpinning the type and size of the space of possibilities? After all, left-wing voters had more candidates with whom they might identify than right-wing voters did; it thus seems logical that they would express multiple favourites. The analysis we conducted thus confirmed some of the results from 2007 and earlier. [37] We tested two multinomial logistic regression models: [38] one based solely on individuals’ social characteristics (age, gender, education level), [39] and the other supplemented these characteristics by including placement on a left/right spectrum and degree of interest in politics. Political competence only allowed us to distinguish between those individuals who expressed no preference and those who expressed at least one. In the first model, higher education graduates stood out from primary education graduates by being half as likely to express no preference for a candidate; men were 1.3 times less likely than women to express no preference. However, age did not have the expected effect. Generally, young voters are different from all other age groups due to the “political moratorium” effect. [40] During this period of their lives, young voters are still “trying to find themselves” politically and have not yet cemented their political beliefs. Nevertheless, here the dividing line was located between those younger and older than 50 years old. Those above 50 were two to three times less likely to have no preferred candidate when compared to 18 to 24-year-olds; this latter group, however, showed no difference when compared to 25 to 34-year-olds or 35 to 49-year-olds. In other words, this negative relationship to the political choices available was not a simple reflection of youthful uncertainty but in fact affected a much larger portion of the electorate, with older voters standing out as a result of their more positive approach to their voting options.

40Introducing degree of interest in politics and left/right placement on a political spectrum likewise confirmed these results. Unsurprisingly, the less politicised the individual, the greater the chances were that they rejected all the candidates. Compared to those who responded “not at all interested”, those responding “fairly interested” were three times less likely to reject all candidates; those answering “very interested” were five times less likely. Moreover, compared to voters who identified as centrists, those on the left or right of the spectrum were between two and four times less likely to include no candidate in the space of possibilities. Equally unsurprising, as soon as these two political variables were introduced, individuals’ social characteristics became less informative. [41] But the main surprise was the persistence of differences with regard to age. Those aged 50 and above were again 1.5 times more likely than all other voters to include at least one candidate in their realm of possibilities. The two models did not allow us to meaningfully distinguish between voters planning to vote for a single candidate and those who felt affinities for two or three. This was true with regard to degree of interest in politics, placement on the political spectrum and education level. The sole trait differentiating single-preference voters from multi-preference voters was still age: those 50 or over were one and a half times less likely to imagine voting for more than one candidate when compared to 18 to 24-year-olds, whereas 25 to 34-year-olds and 35 to 49-year-olds did not differ significantly from the youngest group. This result with regard to age was particularly important. It was controlled by social and political variables, and could not be explained by the youngest cohort’s lack of political competence. It had already been observed eight years earlier, in 2004. This phenomenon has thus been stable over time and is not simply the fruit of particular circumstances. Moreover, it affects a large number of voters, including professionally active adults, deemed to be the most in touch with political life. We are thus confronted with two contradictory hypotheses: 1) either we maintain the effect of life cycles but argue that the strengthening of preferences for a single party (or candidate) occurs later in voters’ lives, or 2) we argue that a “cohort” effect can be observed: the oldest cohorts would thus stand out because of their mono-preferential relationship to politics, while newer cohorts would display either more negative or more multi-preferential relationships (at least until now). [42] As the latter population is set to grow in size, these two phenomena could likewise have an increasingly significant effect on elections. However, the data that we have did not cover a large enough period of time for us to decide which of these hypotheses is correct.

41If we look at multi-preference voters again, we can see that, for example, for voters on the left considering voting for two candidates, their “space of possibilities” stayed entirely on the left side of the political spectrum in 62.5% of cases (in 32% of cases, for Hollande and Mélenchon, and 19.5% of cases for Hollande and Joly). However, 19.5% of voters on the left hesitated between a left-wing candidate and Bayrou, thus illustrating that boundaries with the centre remained rather porous, as they had been in 2007. When left-wing voters had a space of possibilities that included three candidates, 48% of the time they remained on their side of the spectrum and 19.5% of the time the pool included Bayrou. Behind this result, we can see the erosion of cleavage lines on the left but also with regard to the centre, which echoes the electoral results and particularly strong score variations for each of the parties concerned since 2007.

Voting probabilities and political reconfigurations in 2012

42The method used up until this point quickly reaches its limits in terms of accounting for the multiple ways in which voters express their preferences. A specific statistical method is well suited to solve this problem: principal component analysis (PCA). It reveals the structure of response profiles to questions about voting probabilities, [43] which it then organises around eight factorial axes. A mixed classification carried out in the space defined by the first four axes of the PCA allowed us to identify the partitions that optimise the grouping of respondents by preference: individuals from the same group must be as similar as possible and the groups must be as distinct as possible. The typology adopted defined six groups of voters. [44] It is an analysis tool based on the structure of electoral preferences that allows for a better understanding of the logic behind the shifting reconfigurations of electoral preferences, used in conjunction with traditional factors.

43Two groups (1 and 2) were formed by voters largely favourable to left-wing candidates but with different configurations of electoral preferences. Group 3 was composed of voters who split their preferences between Sarkozy and Bayrou. Group 4 was characterised by voters who were very firmly in support of Le Pen and, to a lesser degree, of Sarkozy. Voters from group 5 rejected all candidates except for Sarkozy, for whom they expressed a positive, though more reticent, preference (when compared with group 3). Finally, group 6 was comprised of voters whose preferences went to all candidates except Sarkozy. Figure 3 shows the profiles of the average probabilities obtained for all eight candidates represented in the six groups, over the course of the five polls. The star plot’s variations accurately illustrate the specificity of the structure of preferences for voters in each group.

44The transformation of a preference into an instance of voting behaviour seems fairly evident for the groups of voters that expressed a strong favourite – if not a sole preference, at least a strongly asserted one (group 2 for Hollande, group 4 for Le Pen). Nevertheless, this transition to action remains only probable and is subject to factors that could not be controlled for in this analysis. For the other groups, in particular the first, third and especially sixth group, the move from preference to ballot-casting is impossible to evaluate using our current data. In fact, this typology does not seek to predict voting outcomes, but simply to better understand the complex logic of choice – observed only with difficulty when using traditional tools – to which voters are subjected during the campaign and up until their final choice is made. Voters’ degree of indecision is an essential piece of information here. This classification allows us to follow this phenomenon by type of voters by taking as an indicator the number of probabilities between 7 and 10: one probability signifies a firm choice, two or more, strong hesitation, and none corresponds to an absence of preference. Figure 4 shows the frequency of these levels of indecision for each group at three different moments in the campaign (July 2011, fall 2011, winter 2012). [45] With the help of these analysis tools, let us now examine the six groups of voters.

Figure 3

Average of the voting probabilities for eight candidates, for the whole sample, and divided into six preference groups. In parentheses, the frequency of each group in July 2011 and March 2012

Figure 3

Average of the voting probabilities for eight candidates, for the whole sample, and divided into six preference groups. In parentheses, the frequency of each group in July 2011 and March 2012

Sources: TriÉlec TNS-Sofres surveys.
Figure 4

For each group, the frequency of probabilities between 7 and 10 awarded to each of the eight candidates at three different moments in the campaign

Figure 4

For each group, the frequency of probabilities between 7 and 10 awarded to each of the eight candidates at three different moments in the campaign

How to read this graph: In July 2011, 46% of voters from group 1 and 19% from group 6 expressed a single probability equal to or greater than 7.
Sources: TriÉlec TNS-Sofres surveys.

45On average, voters from group 1 represented 15% of the sample population, but with some changes over the course of the campaign: from 12% in July, they rose to 17% in the fall and dropped back down to 13% at the end of March. They mainly split their preferences between the three “institutional” left-wing candidates who were represented in parliament. Hollande had the strongest average probability (7.12% the whole time), but Joly and Mélenchon better characterised this group. [46] Voters from this group had trouble choosing: in July, 46% of them only expressed one probability higher than 6; in February-March, 36% of them were still in the same situation. The comparison with voters from group 2 is quite interesting: the latter was the largest group, representing 21% of the sample population, having dropped by three points after July. Hollande was this group’s favourite candidate – not only did he have his highest average among these voters (7.22), but all the other candidates, even on the left, had lower group averages than across the sample as a whole. Only Mélenchon was able to maintain his average in this group. The decisiveness of these choices was clearly strengthened over the course of the campaign: in July, 45% only had one favourite choice, but this figure rose to 64% in October and 68% in February-March. This trend was accompanied by a hike in preference scores for the socialist candidate, who went from 6.38 in July to 7.40 in the winter, via a high of 7.52 in the fall. Here we can observe the dynamics that described the socialist candidate’s campaign: a strong upward trend in his potential up until the primary, following by a retreat throughout the electoral campaign and eventually a point of stability that remained one of the highest for all eight candidates. These first two groups share the fact that about three-quarters of their members consider themselves to be on the left (even firmly on the left) of the political spectrum, and that almost two-thirds of them can be characterised as “politically engaged”, as they are interested in both politics in general and the election at hand specifically (having answered “fairly” or “very” both times). Multiple preferences thus do not preclude a high level of politicisation, nor are they necessarily a sign of political incompetence. Both groups had 80% of members who found the socialist candidate fairly or very convincing, although adherence was slightly higher in group 2. But there were also slight differences between these two groups, and at times even stark contrasts that echoed their respective preference systems. Consequently, Hollande’s image was positive in both groups but slightly better in group 2; 38% of voters from group 1 positively assessed Joly’s positions and 58% of them liked Mélenchon’s arguments, compared to 18% and 49% respectively for group 2. Likewise, the latter were more numerous in supporting Hollande: 70% in October through March, compared with 56% for group 1 (where only 16% hoped for their favourite candidate’s victory). All these figures attest to a strong level of conviction in group 2. Logically, at the end of March, they were the group most likely to cast their ballots. [47] It is reasonable to believe that the majority of group 2 voted in both rounds of the presidential election for the socialist candidate. But it would be misleading to think that all these voters were steadfast supporters of the Socialist Party: in July 2011, 6% of them expressed two strong preferences. In February-March, this group was twice as large, and even within this group, at least one voter out of five still did not have strong preferences (above a score of 6).

46Group 6 was the one to exhibit the most significant progression from July to March: it doubled in proportion from 7% to 14%, ultimately reaching the same level as group 1. Politically rather mixed, this group still belonged to the left’s sphere of influence; 46% of its members identified as left-wing or centre-left, and the most extreme left-wing candidates obtained their highest potential scores here: 4.5 for Poutou, 5.08 for Arthaud and 5.3 for Mélenchon. Joly likewise did much better than her average score. Even though Hollande once again displayed the highest electoral potential, his level (5.6) was much lower than in the first two groups, and barely higher than Mélenchon’s, which translates the relative distance between these voters and their future president. The majority of voters in this group identified as centre, centre-right or right-wing (21.5%, 17% and 11% respectively). Consequently, Le Pen and Bayrou scored better with this group than average. Only Sarkozy had an electoral potential that was on par with his global average. In this group of voters, we find individuals both hostile to the incumbent president and wary of the governmental left. Uncomfortable with the dominant electoral offer imposed by the bipartisan nature of the presidential election and the Fifth Republic, these voters were largely unable to choose a favourite candidate. Far from helping them make up their minds, the electoral campaign only served to increase their uncertainty: while only 19% of these voters had a favourite in July and 18% in February-March, more and more of them defined their electoral space of possibilities with two or more candidates (42% in the summer, 49% in the fall and 54% in the winter, when only 28% of these voters expressed a firm preference). One-third of this group expressed no particular desire for any candidate to win, while 28% cited Hollande and 15.5% another more far left candidate. A tiny majority found Hollande’s ideas convincing (50%), while 42% felt the same way about Mélenchon’s propositions. This group of voters, predominantly young and working-class, included the highest proportion (31%) of individuals who were indifferent to the presidential election.

47Groups 1, 2 and 6 represent most of the voters leaning to the left. They allow us to better understand the changes in electoral scores within this segment of the political spectrum. Group 2 likely explains why the Socialist Party remained the dominant left-wing party, as the majority of its voters included only Hollande in their possibilities (even if Mélenchon was tempting to some of them as well). On the other hand, groups 1 and 6, whose combined weight was much greater than group 2’s, showed how important multiple preferences were for the left of the political spectrum taken as a whole (some, however, argue that the left and the extreme left should be viewed as two different segments). However, in addition to Hollande, many voters in group 6 included Mélenchon and even his Trotskyist rivals among their possibilities – and the Front de gauche candidate was not rejected in groups 1 and 2, either. Consequently, we can confirm in 2012 what had already been observed in 2002 and 2007: [48] from the voters’ point of view, there was not necessarily a major difference between establishment left-wing candidates and their radical contenders. This was also the case for Bayrou. In 2007, a group of “ségo-bayrouistes” [49] had emerged, which was composed of individuals planning to vote for Ségolène Royal and Bayrou, but also for Besancenot and Voynet. The fact that we found a second group very close to this one five years later confirms that the boundary between left and centre remains blurred. In short, the left has become a playing field with more and more competitors, which explains why winning an election is oftentimes a short-lived victory for parties such as the NPA or the EELV. Save for the Socialist Party – and only to a certain degree – none of the parties on the left have their “own” steadfast voters.

48Group 3 was comprised of centrist (27%), centre-right (32%) and right-wing voters (18%), whose preferences were split almost evenly between Bayrou and Sarkozy: both candidates obtained their best electoral scores in this group, 5.63 and 5.53 respectively. The size of this group increased from 17% of the electorate in July to 22% in February-March, ultimately weighing a little more heavily than group 2 (favourable to Hollande). But this upward trend did not allow the UMP candidate to take the lead from Bayrou’s MoDem: in February-March, Sarkozy’s score was 5.68, while Bayrou was ahead at 5.88. The MoDem candidate’s lead was reflected in the perception voters had of his platform: 60% of them found his claims convincing, while they rated Sarkozy four points lower. This group illustrates the important discrepancy that can exist between the expression of preference for a candidate and the ultimate decision made at the ballot box. In terms of preferences and opinions, competition seemed close between Sarkozy and Bayrou, whereas on voting day Sarkozy won almost 6.5 million more votes than Bayrou. This discrepancy can also be seen in our polls when, for example, in February-March only 16% of voters hoped for Bayrou’s victory, compared with 31% for Sarkozy. The bipartisan nature of the election, the belief (or not) that a candidate could win and the notion of a “useful” vote all clearly played their part in producing this discrepancy between intention and reality. Above all, the incumbent effect played an important role. [50] In addition, the sociology of this group was clearly susceptible to this effect. It included the largest number of practising – whether frequent or infrequent – Catholics (34% compared to 22% on average) and the lowest number of non-religious voters (23% compared to 37% on average). Likewise, this group had the lowest proportion of young voters between 18 and 24 years old (5%, compared with 13%) and older voters were proportionally overrepresented (31%, compared with 23%). And in fact, for elderly Catholic voters, political experience and the legitimacy of the incumbent right-wing candidate were important criteria informing their voting decision: between two contenders on the (centre-) right, they overwhelmingly voted for the incumbent. [51]

49Group 4 comprised right-wing voters (37%, 20% of whom considered themselves to be far-right) who picked Le Pen as their favourite candidate. While her electoral potential was on average 1.79 for the general sample population, Le Pen averaged 7.38 over the entire period in this group, exhibiting a small progression from July (7.36) to February-March (7.54). But although the intensity of preferences for this candidate seemed to grow, the size of this group, on the other hand, diminished between July (16%) and February-March (13%). Sarkozy obtained his third-best score with this group: 3.61 on average, 3.39 in July and 3.80 at the end of the campaign. Much like those in group 6, voters in this class were among the least political: in February-March, 45.5% of them expressed an interest in politics and the election (compared to 53% on average), while 31% of them remained indifferent (compared to 24% on average). This attitude can be linked to the professional sociology of this group, which, like group 6, primarily comprised working-class individuals (31%) and employees (28%), while executive and white-collar professions only represented 10% (compared to 22% to 25% in groups 1, 2 and 3). Le Pen’s proposals hit the bull’s-eye for these voters: 74% of them found her ideas convincing, even if only 34% of them hoped that she would win. Similarly, 44% were also convinced by Sarkozy’s ideas, and 23% hoped for his victory. All these elements illustrate the fact that the marked preference for the far right exhibited by these working-class voters was also compatible with a preference for Sarkozy. Hence, the permeability between moderate right-wing voters and part of the FN’s electorate continued well beyond 2007, just as other studies have already shown. [52]

50Finally, for voters in group 5, no candidate got their approval. All the candidates obtained weak probability scores, significantly lower than their general averages (with the exception of Sarkozy, who received his second-best score). Figure 4 illustrates the oddity of this group: between 26% of these voters in July and 36% in the winter expressed a preference for one candidate, but between 73% and 60% did not grant any candidate a score above a 6. The absence of favourites is the hallmark of this group. Sarkozy did better here than his global average, but he remained a choice made “for lack of anyone better”. His probability score remained weak, even if it regularly rose, going from 3.77 in July to 4.37 in the fall and 4.41 in February-March. Similarly, the size of this group decreased from 24% in July to 19% in the winter. These observations show us that even for this largely apolitical group (30% were indifferent to the election), with uncertain political orientations (12% did not position themselves on the political spectrum, compared with 4.5% globally), the campaign had a significant mobilising effect that benefited the incumbent. In a somewhat surprising turn of events, voters in this group came from all different socio-economic segments of the population, and their ages and occupations were astonishingly representative of the sample population as a whole. Executives and white-collar professionals made up 21% of this group, compared to an average of 19%; the only important discrepancy concerned those aged 65 or older, who were somewhat over-represented (29.5%) in this group compared to the sample as a whole (23%). These voters were the most distant from political life and electoral competition, without this attitude being explained by the effects of class-based domination.

51It may seem tempting to interpret our groups – or at least those that converged on a clear preference – as so many different electorates. This temptation becomes even stronger if we consider that their relative weight in the polls came very close to matching the results of the first round of the presidential election. For example: group 2 favouring Hollande accounted for 21% in the March poll and similarly 22% of registered voters, while group 4 which preferred Le Pen accounted for 13% in March and for 14% of registered voters. But even in seemingly favourable cases, the connection between preferences and actual votes has not been well established. It depends on a number of factors associated in particular with the campaign and political circumstances which, at the very moment of casting their ballot, cause voters to make a different decision: voting for Sarkozy instead of Le Pen or vice-versa.

52The goal of this analysis of voting probabilities is not to try to obtain what was ultimately the final result of the election, but instead to provide some insight into the political dynamics that produced this result. From this point of view, it is particularly interesting to observe that as he had not gained a strong following in his own right, Sarkozy appeared in no group with a high voting potential, which evidently presented an important obstacle to his re-election during the second round. Support for the UMP candidate appears to have been largely by default, according to our analysis: among centre-right legitimists who appreciated Bayrou, among a portion of apolitical voters who shunned the election and among some Front National supporters. On the other hand, Hollande was able to rely on an electoral base that had chosen him as their preferred candidate. But our analysis of voting preferences also showed that other candidates disputed his leadership of the left. For an important percentage of voters, especially those in groups 1 and 6, voting for the socialist candidate in the first or even the second round was also probably a vote by default. The analysis of voting probabilities and electoral preferences works better than traditional methods to allow us to understand the complex decisional logic voters experience over the course of an electoral campaign. Thanks to this logic, we can account for the role of votes in support of a candidate as against votes cast for lack of a better option, thus obtaining a more nuanced picture of the meaning of electoral choices.


54* *

55These analyses invite us to revisit the traditional notions used to understand voting choices. It is a mistake to see in a ballot cast merely a positive preference for a candidate or party. Helena Catt has already suggested this fallacy: here we were able to confirm it, with a situation in which almost one out of every four voters chose the least worst option. Negative votes are a reality, and they seem to have never been more prevalent than in 2012’s election. It is likewise a mistake to insist on exclusive preferences. Almost one out of every three voters polled described a space of possibilities that included two or three candidates. Most importantly, both negative voters and those with multiple preferences are likely to represent larger swathes of the electorate if the phenomena that we have identified are in fact due to generational renewal and not merely due to life cycle effects. Let us not forget that the “traditional” voters who had only one preferred candidate were the oldest among our sample population.

56Regardless of what is causing these phenomena to occur, voter participation is becoming increasingly dependent on context and the issues at stake in a particular election. Judging from our analysis, we can likewise speculate that the same will be true for parties’ electoral results. The elections that have taken place since the beginning of the 2000s can also be seen as a consequence of greater and greater numbers of multi-preference voters. This is particularly the case because political boundaries are shifting, both between the left and centre and between the right and the far right. Additionally, electoral campaigns are now exhibiting multiple effects on voting choices, at times strengthening or reawakening convictions, while at other times weakening existing preferences. Consequently, elections to come are likely to become much more unpredictable.

Methodological annex

57Here we aim to explain to our readers how we established our descriptive typology of the responses to questions regarding voting probabilities given by the 5,033 individuals polled between early July 2011 and late March 2012 in the five TNS-Sofres TriÉlec surveys. It was established in three distinct stages [53] using the SPAD software programme. As Ludovic Lebart, Marie Piron and Alain Morineau remind us, we must not “exaggerate the relevance and importance of the number of groups of a classification, as classification is never an end in and of itself”. [54] Although this classification was obtained based on structuring the answers, it is neither true nor false, but merely useful or not for future research. The processes of automatic group description and the statistical tests used to characterise them allow us to measure the value of the partitions obtained. But, naturally, other interesting groupings are possible.

58First of all, principal component analysis was conducted using the answers given to voting probability questions regarding the eight candidates present throughout all five polls as active variables. Dupont-Aignan and Cheminade were therefore not taken into account. Answers to probability questions were scored between 0 and 10, plus the option to answer “I don’t know him/her” or “Unsure” (which had specific codes). In order to work with all of the respondents who were potential voters and to take into account the possible absence of preference for a candidate (generally a small party candidate), blank answers were recoded as 0. This slightly changed the political significance of the zero score: in the very large majority of cases (see Figure A1 below), it corresponded to an explicit refusal to vote for a candidate, and only much more rarely did it signify a lack of familiarity the candidate, or a voter’s inability to gauge their preference for said candidate. Nevertheless, 1) regardless of the reasons, zero indicates that a voter did not consider a candidate to be part of his/her electoral possibilities and 2) if we had not made this change, this would have meant excluding voters who gave one blank answer out of eight. The voting probabilities thus documented can be treated as numeric variables for all 5,033 individuals polled.

Figure A1

The distribution, in numbers, of “0” answers (refusals to vote for a given candidate, “I don’t know him/her” or “Unsure”) to the question regarding voting probabilities for eight presidential candidates in 2012. Total of all five polls, n = 5,033, results are not weighted

Figure A1

The distribution, in numbers, of “0” answers (refusals to vote for a given candidate, “I don’t know him/her” or “Unsure”) to the question regarding voting probabilities for eight presidential candidates in 2012. Total of all five polls, n = 5,033, results are not weighted

59The eight axes of the principal component analysis bring together the many possible systems of electoral preferences. The bar graph with eigenvalues (see Figure A2) shows that the first axis accounts for 30% of the data spread’s inertia. The first four axes account for 70%: after that, we can see a dip followed by a consistent decrease in eigenvalues; the information summarised by these last axes becomes more and more difficult to interpret and produces an increasingly marginal explanation.

Figure A2

Bar graph showing the PCA’s eigenvalues (in explained inertia percentage)

Figure A2

Bar graph showing the PCA’s eigenvalues (in explained inertia percentage)

60Table A1 showing the factorial coordinates or the correlations between active variables and the first four factors allows us to see which information these axes summarise and thus interpret it accordingly. Axis 1 shows an opposition between preferences for the five candidates on the left, and to a lesser extent, for Bayrou (positive correlations) and in the negative direction, preferences for Sarkozy (- 0.31). However, these preferences are better accounted for by axis 2 (0.65), like those for Le Pen (0.67) who did not contribute to axis 1; Bayrou (0.40) is also present on this second axis. The plane shaped by these two axes essentially illustrates the opposition between Hollande (coordinates 0.55 and - 0.44) and Sarkozy (- 0.31 and 0.65). Among the left-wing candidates, axis 2 differentiates the two Trotskyist contenders from Hollande, while Mélenchon and Joly occupy a half-way position. The third axis only represents 12% of our data. It is strongly anchored at the negative pole by Bayrou (- 0.72), then less clearly but almost equally by Hollande (- 0.34) and Sarkozy (- 0.32). At the positive pole we find Le Pen (0.36) and the three far-left candidates: this axis thus pits the moderate candidates against the “extremes” of the far left or far right. Finally, axis 4 (9.9% of the inertia) is established by preferences for Le Pen (0.61), Hollande (0.35) and Bayrou (0.21), in fairly stark opposition to Sarkozy (- 0.35).

Table A1

Factorial coordinates of the first four axes

Table A1
Variable name Axis 1 Axis 2 Axis 3 Axis 4 N. Arthaud 0.71 0.30 0.19 - 0.09 J.-L. Mélenchon 0.70 - 0.09 0.17 0.04 P. Poutou 0.69 0.29 0.21 - 0.31 E. Joly 0.67 - 0.07 - 0.08 - 0.18 F. Hollande 0.55 - 0.44 - 0.34 0.35 F. Bayrou 0.36 0.40 - 0.72 0.21 M. Le Pen - 0.01 0.67 0.36 0.61 N. Sarkozy - 0.31 0.65 - 0.32 - 0.35

Factorial coordinates of the first four axes

61Examining these four dimensions reveals the complexity of voters’ preferences, which do not boil down to a simple opposition between left and right, even if this distinction remains structurally significant. For most of the candidates, positive or negative correlations were strong with the first four axes, which illustrates a complex web of preferences linking different contenders.

62As a second step, we established a classification based on these four factors in order to clarify how these preferences were interconnected. Leaving out the following increasingly difficult to understand factors amounted to smoothing out the data set, which in general allows for more homogenous groups to be composed. This classification sought to bring together the individuals polled in a fixed number of groups, according to their similarity and measured by their distance in the space defined by the first four factors of the PCA. This method attempts to establish a system of partitions where the individuals grouped together are as similar as possible, while the groups themselves are as distinct as possible. In other words, it looks to minimise intra-group variance (individuals in the same group are similar) while maximising inter-group variance (each group is quite different from the others).

63Several methods of classification are possible. Bottom-up hierarchical methods of classification offer a series of nested partitions, the difficulty thus consisting in choosing the most relevant cut-off point. Partitioning methods establish a single partition in a number of groups that was fixed a priori by repeating the steps allocating individuals to groups, but the final partition depends on the partition chosen during the initial step. The methodology that we chose, as suggested by the SPAD software, alternates between these two approaches.

64As a first step, two partitions of ten groups were established using the mobile cluster model based on two different initial partitions, chosen at random. These two final partitions were crossed in order to obtain stable groupings composed of individuals grouped together in the two partitions. We thus produced a 53-group partition composed of a small number of stable groupings accounting for a large number of the individuals, and numerous groupings of isolated individuals. This first construction showed that groupings could be structured independently from initialisation. The number of groups in the partition was naturally too large. We thus used a bottom-up hierarchical classification method in order to bring together the groups obtained in the previous step. Groupings were based on Ward’s variance criterion in accordance with the mobile cluster method. The dendogram shows that various partitions were possible, in particular with 5 or 8 groupings. Nevertheless, we chose to preserve 6 groupings, as cutting off the hierarchical tree at this level allowed us to work on a small number of groups of relatively the same size, each one representing between 12% and 22% of the sample population. The hierarchical node immediately above would bring together two close groups of significant size (17% and 13%): a five-group partition would thus lose some amount of precision. The partition immediately below in 8 groups would clearly have more unequal numbers. Finally, the loss of inter-group inertia caused by the formation of the 6-group node was significantly less than for the nodes above it.

Figure A3

Dendogram produced by the mixed classification of factors with a 6-group partition

Figure A3

Dendogram produced by the mixed classification of factors with a 6-group partition

65The third and final step in consolidating the groups was accomplished using a mobile cluster partitioning method based on the partition obtained previously. This step allowed us to reallocate individuals to groups in order to improve their internal homogeneity and increase inter-group distance. The results of this consolidation are presented in table A2. The consolidation caused the numbers of all the groups to shift. Group 5 was the most stable, while group 6 lost the greatest number of individuals and group 2 gained the most. These reallocations greatly increased the homogeneity of groups 3, 5 and 6 (for example, in group 6, inertia dropped from 0.73 to 0.466) and somewhat decreased the homogeneity of groups 1, 2 and 4. Globally, however, the groups became more homogenous and more distinct from each other, since the proportion of inertia between groups in the total variance went from 60% to 63% post-consolidation. The frequency of these consolidated groups in the sample is presented in percentages in the last column of the table.

Table A2

Inertia, numbers and frequencies of the six groups, before and after the consolidation of the reallocation of individuals in groups

Table A2
Inertia Numbers Frequency Before After Before After Before After Inter-group inertia/total inertia 0.605 0.634 Intra-group inertia Group 1 0.2200 0.312 632 744 13 15 Group 2 0.1608 0.2460 839 1,039 17 21 Group 3 0.5138 0.3812 1,124 981 22 19 Group 4 0.2989 0.3843 586 669 12 13 Group 5 0.3024 0.2736 1,080 1,015 21 20 Group 6 0.7305 0.4661 772 585 15 12 Total 5.6389 5.6389 5,033 5,033 100 100

Inertia, numbers and frequencies of the six groups, before and after the consolidation of the reallocation of individuals in groups

66The content of these groups remains to be studied, in particular the traits that distinguish them from the rest of the sample population. To this end, table A3 compares the average voting probability for each candidate by group and for the sample population as a whole.

Table A3

Characterising the six groups by average voting probabilities by group compared with the sample population as a whole. In parentheses, the number and frequency of each class in percentages

Table A3
Group 1 (n = 744; 15%) Group 2 (n = 1,039, 21%) Characteristic variables Group average Overall average Test value Characteristic variables Group average Overall average Test value E. Joly 4.591 1.991 28.58 F. Hollande 7.215 4.559 27.46 J.-L. Mélenchon 4.59 2.259 23.53 J.-L. Mélenchon 2.41 2.26 N.S. F. Hollande 7.118 4.559 21.61 E. Joly 1.488 1.991 - 6.5 N. Arthaud 2.403 1.295 14.85 P. Poutou 0.066 0.837 - 15.5 P. Poutou 1.285 0.837 7.36 N. Arthaud 0.269 1.295 - 16.84 F. Bayrou 3.418 2.769 6.9 F. Bayrou 1.449 2.769 - 17.17 M. Le Pen 0.484 1.791 - 12.98 M. Le Pen 0.22 1.791 - 19.1 N. Sarkozy 0.934 2.962 - 17.3 N. Sarkozy 0.31 2.962 - 27.71 Group 3 (n = 981; 19%) Group 4 (n = 669; 13%) Characteristic variables Group average Overall average Test value Characteristic variables Group average Overall average Test value F. Bayrou 5.617 2.769 35.75 M. Le Pen 7.384 1.791 52.20 N. Sarkozy 5.529 2.962 25.88 N. Sarkozy 3.608 2.962 5.19 F. Hollande 4.380 4.559 N.S. N. Arthaud 0.934 1.295 - 4.54 M. Le Pen 1.370 1.791 - 4.94 F. Bayrou 2.064 2.769 - 7.04 E. Joly 1.511 1.991 - 6.14 P. Poutou 0.271 0.837 - 8.74 N. Arthaud 0.662 1.295 - 10.03 J.-L. Mélenchon 1.280 2.259 - 9.30 P. Poutou 0.285 0.837 - 10.70 E. Joly 0.554 1.991 - 14.27 J.-L. Mélenchon 1.236 2.259 - 12.20 F. Hollande 2.357 4.559 - 17.47 Group 5 (n = 1015; 20% Group 6 (n = 585; 12%) Characteristic variables Group average Overall average Test value Characteristic variables Group average Overall average Test value N. Sarkozy 4.239 2.962 13.15 P. Poutou 4.501 0.837 52.42 P. Poutou 0.091 0.837 - 14.79 N. Arthaud 5.079 1.295 44.14 M. Le Pen 0.339 1.791 - 17.40 J.-L. Mélenchon 5.297 2.259 26.71 N. Arthaud 0.203 1.295 - 17.66 E. Joly 4.457 1.991 23.61 E. Joly 0.384 1.991 - 20.55 F. Bayrou 4.362 2.769 14.75 J.-L. Mélenchon 0.279 2.259 - 24.13 M. Le Pen 3.075 1.791 11.10 F. Bayrou 0.436 2.769 - 29.91 F. Hollande 5.603 4.559 7.68 F. Hollande 0.988 4.559 - 36.37 N. Sarkozy 2.990 2.962 N.S.

Characterising the six groups by average voting probabilities by group compared with the sample population as a whole. In parentheses, the number and frequency of each class in percentages

67Consequently, in group 1, Joly has a voting probability of 4.59 compared to her overall average of 1.991. A statistical indicator (the test value, 28.58 in this case) measures the strength of the specificity of this group average in relation to the average of individuals as a whole; the candidates/variables are ranked according to the decreasing value of this indicator. Hence we can see that preferences for Joly are the primary trait of group 1, even if her score remains lower than Hollande’s. But in addition to the candidates’ averages in this group and their degree of specificity, it is important to observe the strong tendency towards multiple preferences demonstrated by this group of voters. On the other hand, voters in this group gave significantly lower probabilities than the average to Le Pen and Sarkozy. The reliability of these results is measured by a test which is not present in the table, as all the probabilities of the results caused by chance were less than 1 out of 1,000, except in three cases; the value of the test value is then marked as N.S., “not significant”.

68Every group thus represents systems of electoral preferences that are significantly different from each other.

69Group 1 (15%) is composed of voters who had several favourites on the left: Joly, Mélenchon and Hollande, and to a lesser extent the Trotskyist candidates and Bayrou.

70Voters in group 2 (21%) were also on the left, but expressed a different set of preferences: Hollande was the only candidate to characterise this group, with the average probability of voting for him at 7.21 compared to his 4.65 average and 7.12 in group 1. It is noticeable that the other left-wing candidates had much weaker scores in this group, with the exception of J.-L. Mélenchon.

71In group 3 (19%), Bayrou and Sarkozy obtained their best scores, 5.62 and 5.5 respectively. This group assembled moderate voters whose preferences were split between the centre-right candidate and the incumbent president.

72In group 4 (13%), Le Pen got her best score by far (7.39 compared to 1.79), but Sarkozy also did a little better than his average (3.61); all the other candidates scored lower than their averages. A portion of those voters expressing a strong preference for Le Pen also declared a relatively strong preference for the UMP incumbent.

73Group 5 (20%) was weakly characterised by Sarkozy’s score (an average of 4.24 and test value of 13.15); the seven remaining candidates obtained much lower scores than their overall averages (very negative test values). Voters in this group did not have a clearly expressed favourite but granted the incumbent president a slight margin of preference.

74Finally, in group 6 (12%), all the candidates had markedly better scores than their overall averages except for Sarkozy (non significant test value). Far left candidates obtained their best scores in this group, whereas Hollande barely surpassed his global average. Bayrou and Le Pen had their second-best scores among these voters. This group, the most politically diverse, brought together voters whose main shared trait was rejecting the incumbent president from their space of possibilities.


  • [1]
    Jean-François Laslier, Karine van der Straeten, “Vote par assentiment pendant la présidentielle de 2002: analyse d’une expérience”, available online at: <>; Antoinette Baujard, Herrade Igersheim, Expérimentation du vote par note et du vote par approbation lors de l’élection française du 22 avril 2007, final report from the Centre d’analyse stratégique, 2007, available online at: <>.
  • [2]
    In particular, see Mark Franklin’s work (with two of his colleagues), which shows the gap between party preferences and party-based choices during European elections: Wouter Van der Brug, Cees Van der Eijk, Mark Franklin, The Economy and the Vote. Economic Conditions and Elections in Fifteen Countries (New York: Cambridge University Press, 2007).Online
  • [3]
    TriÉlec is a research group that brings together three laboratories associated with the Fondation nationale des sciences politiques (FNSP – National Foundation for Political Science): the Émile Durkheim Centre in Bordeaux, the Centre d’études européennes in Paris, and PACTE in Grenoble. Its main goal is to contribute to the study of voting, by addressing new issues and methods, as well as participating in international scholarly debate. The project “Dynamiques politiques 2012”, coordinated by Sylvain Brouard, comprises five different surveys (between July 2011 and March 2012) conducted with national samples established according to the quota method of 1,000 French citizens aged 18 or over, registered to vote and surveyed over the phone by TNS-Sofres. The project was funded by the Ministry of Internal Affairs, Sciences Po Paris, Sciences Po Bordeaux and Sciences Po Grenoble.
  • [4]
    André Siegfried, Le tableau politique de la France de l’Ouest sous la Troisième République (Paris: Armand Colin, 1913).
  • [5]
    Paul Lazarsfeld, Bernard Berelson, Hazel Gaudet, The People’s Choice (New York: Columbia University Press, 1944); Guy Michelat, Michel Simon, Classe, religion et comportement politique (Paris: Presses de Sciences Po/ Éditions sociales, 1977).
  • [6]
    Angus Campbell, Philip Converse, Warren Miller, Donald Stokes, The American Voter (New York: Wiley, 1960).
  • [7]
    Martin Seymour Lipset, Stein Rokkan, Party Systems and Voter Alignments (New York: Free Press, 1967); Elmer Schattschneider, The Semisovereign People. A Realist’s View of Democracy in America (New York: Holt, Rinehart and Winston, 1960).
  • [8]
    François Héran, “Les intermittents du vote: un bilan de la participation électorale de 1995 à 1997”, Insee Première, 546, 1997, 1-4; Stéphane Jugnot, “La participation électorale en 2007: la mémoire de 2002”, Insee Première, 1169, 2007, 1-4; Stéphane Jugnot, Nicolas Frémeaux, “Les enfants des baby-boomers votent par intermittence, surtout quand ils sont peu diplômés”, in France. Portrait social. Édition 2010 (Paris: Insee, 2010), 121-31. See also Anne Muxel’s article in this issue.
  • [9]
    Mark Franklin, Tom Mackie, Henry Valen (eds), Electoral Change. Response to Evolving Social and Attitudinal Structures in Western Countries (Cambridge: Cambridge University Press, 1992).
  • [10]
    Russell Dalton, Martin Wattenberg (eds), Parties Without Partisans. Political Change in Advanced Industrial Democracies (Oxford: Oxford University Press, 2000).
  • [11]
    Samuel Popkin, The Reasoning Voter. Communication and Persuasion in Presidential Campaigns (Chicago: The University of Chicago Press, 1991).
  • [12]
    Robert Erikson, Michael MacKuen, James Stimson, The Macro-Polity (Cambridge: Cambridge University Press, 2002).
  • [13]
    Stephen Ansolabehere, “The paradox of minimal effects”, in Henry Brady, Richard Johnston (eds), Capturing Campaign Effects (Ann Arbor: The University of Michigan Press, 2006), 29 –44.
  • [14]
    Henry Brady, Richard Johnston (eds), Capturing Campaign Effects (Ann Arbor: The University of Michigan Press, 2006); Shanto Iyengar, Donald Kinder, News that Matters. Television and American Opinion (Chicago: The University of Chicago Press, 1987); Richard Lau, David Redlawsk, How Voters Decide. Information Processing During Electoral Campaigns (Cambridge: Cambridge University Press, 2006).Online
  • [15]
    Vincent Tiberj, “Le système partisan comme ‘espace des possible’”, in Florence Haegel (ed.) Partis politiques et système partisan en France (Paris: Presses de Sciences Po, 2007), 287-319; Vincent Tiberj, in collaboration with Bruno Cautrès, “L’espace des possibles électoraux”, in Bruno Cautrès, Anne Muxel (eds), Comment les électeurs font-ils leur choix? Le Panel électoral français 2007 (Paris: Presses de Sciences Po, 2009), 77-99.
  • [16]
    Helena Catt, Voting Behavior. A Radical Critique (Leicester: Leicester University Press, 1996).
  • [17]
    Guy Michelat, Michel Simon, “Les ‘sans-réponse’ aux questions politiques: rôles imposés et compensation des handicaps”, L’Année sociologique, 32, 1982, 81-114; Gérard Grunberg, “Sondages et participation politique”, in Bertrand Badie, Pascal Perrineau (eds), Le citoyen (Paris: Presses de Sciences Po, 2000), 165-82.
  • [18]
    Ronald Inglehart, The Silent Revolution. Changing Values and Political Styles among Western Publics (Princeton: Princeton University Press, 1977); Russell Dalton, Citizen Politics. Public Opinion and Political Parties in Advanced Industrial Democracies (Chatham: Chatham House, 1988).
  • [19]
    This is an idea that can be found both in French sociological criticism and in American political science. Pierre Bourdieu, La distinction. Critique sociale du jugement (Paris: Minuit, 1978); Martin Wattenberg, The Decline of American Political Parties. 1952–1988 (Cambridge: Harvard University Press, 1990).
  • [20]
    Daniel Gaxie, Le cens caché. Inégalités culturelles et ségrégation politique (Paris: Seuil, 1978).
  • [21]
    Michael Lewis-Beck, William Jacoby, Helmut Norpoth, Herbert Weisberg, The American Voter Revisited (Ann Arbor: The University of Michigan Press, 2008).
  • [22]
    An original use of these questions is illustrated in George Marcus, Russell Neuman, Michael MacKuen, Affective Intelligence and Political Judgement (Chicago: University of Chicago Press, 2000).
  • [23]
    Available online at: <>.
  • [24]
    1,000 French voters of Maghrebi, African and Turkish origins, and 1,000 individuals representative of the French electorate as a whole. See Sylvain Brouard, Vincent Tiberj, Français comme les autres? (Paris: Presses de Sciences Po, 2005).
  • [25]
    In particular, see Bernard Denni, Philippe Caillot “Les probabilités de vote”, in Philippe Guilbert, David Haziza, Anne Ruiz-Gazen, Yves Tillé (eds), Méthodes d’enquêtes. Applications aux enquêtes longitudinales, à la santé et aux enquêtes électorales (Paris: Dunod, 2008), 164-9; and “Que nous apprend la méthode du vote probabiliste appliquée à la pré-campagne de l’élection présidentielle?”, in the issue “2007: élections du changement?”, Revue politique et parlementaire, 1044, September 2007, 230-5.
  • [26]
  • [27]
    Here we are referring to the stop and think experiment put in place by Stanley Feldman and John Zaller, which allowed for a better understanding of what Americans had in mind when they answered questions about state services or helping the African-American community (John Zaller, Stanley Feldman “A simple theory of the survey response: answering questions versus revealing preferences”, American Journal of Political Science, 36(3), 1992, 579-616).Online
  • [28]
    The quotations from voters used in this article all come from an online qualitative forum or an On Line Bulletin Board (OLBB). This polling tool, although heavily used in marketing studies, was first implemented in electoral research (to our knowledge) by the PACTE laboratory, with the financial support of EDF’s GRETS (Groupe de recherche énergie, technologie et société) research group, and the methodological support of a consultant, Édith Vassaux. The reactions of eighteen different voters to the campaign were also studied from mid-November 2011 to 24 April 2012.
  • [29]
    Other distributions are possible; one could, for example, make 10 a separate category corresponding to the maximum support possible for a candidate and one of the markers of his or her core electorate.
  • [30]
    Grouping these last two polls together allows us to work with significant numbers, the weakest being for Jean-Luc Mélenchon (53 individuals) and François Bayrou (96).
  • [31]
    See, for example, the results of the different types of voting proposed in 2012 by the plurinominal vote team, available online at: <>.
  • [32]
    Moreover, his standard deviation was systematically lower than that of Marine Le Pen, Nicolas Sarkozy or François Hollande.
  • [33]
    The levelling-out of minority candidates in terms of media presence was first visible in the zero response. Knowing candidates better does not necessarily mean approving of them.
  • [34]
    Gérard Grunberg, Florence Haegel, La France vers le bipartisme? La présidentialisation du PS et de l’UMP (Paris: Presses de Sciences Po, 2007).
  • [35]
    With a number of parties included that was smaller than the number of candidates in the 2011 –2013 surveys.
  • [36]
    Bernard Denni, “Des électeurs toujours attentifs mais perplexes et inégalement mobilisés”, March 2012, available online at: <>.
  • [37]
    Vincent Tiberj, “Le système partisan comme ‘espace des possibles’”.
  • [38]
    More details about these analyses are available from the authors.
  • [39]
    Generally, these variables are used to approximate political competence, and especially when one does not have access to questions about political knowledge, for example. See Vincent Tiberj, “Compétence et repérages politiques en France et aux États-Unis: une contribution au modèle de ‘l’électeur raisonnant’”, Revue française de science politique, 54(2), 2004, 261-87.
  • [40]
    Anne Muxel, L’expérience politique des jeunes (Paris: Presses de Sciences Po, 2000).
  • [41]
    For example, education level no longer showed a significant influence. This can largely be explained by the fact that the more educated an individual is, the more likely s/he is to be interested in politics. No “double penalty” exists in terms of academic qualifications: a secondary education graduate is less likely to be very interested in politics than a higher education graduate, but if both are, they are just as likely to not declare any candidate in their space of possibilities.
  • [42]
    With this kind of phenomenon, it behoves us to remain cautious. These cohorts could become less “multipreferential” in the decades to come. In that case, we would fall back upon the life cycle argument. On the other hand, if these multi-preferences are conserved, we will in effect be witnessing the effects of generational renewal.
  • [43]
    These analyses were conducted with the SPAD software programme. The method was similar to that used by Vincent Tiberj in collaboration with Bruno Cautrès in “L’espace des possibles électoraux”.
  • [44]
    The typology elaborated, based on the five successive polls used in our study, or on 5,033 individuals, only addressed the eight candidates that were present during all five polls. The method behind this classification by factors is described in detail in the annex. These analyses were not weighted.
  • [45]
    For reasons of statistical reliability, we have grouped all five polls together here.
  • [46]
    In the methodological annex, readers will find a detailed table of the statistical discrimination for each group, with averages and test values.
  • [47]
    Voting probabilities can also be used to measure electoral participation. A score of 10 is accordingly a good measure to gauge likelihood of voting: in March 2012, 79% of those polled gave this response.
  • [48]
    Gérard Grunberg, Étienne Schwesiguth, “La tripartition de l’espace politique”, in Pascal Perrineau, Colette Ysmal (eds) Le vote de tous les refus (Paris: Presses de Sciences Po, 2003), 341-62; Vincent Tiberj in collaboration with Bruno Cautrès, “L’espace des possibles électoraux”.
  • [49]
    Vincent Tiberj, in collaboration with Bruno Cautrès, “L’espace des possibles électoraux”.
  • [50]
    This is what we learn when analysing the perception of both candidates in this group: voters were less worried about François Bayrou than the outgoing president; Bayrou was seen as being much closer to the people and their issues; his desires for reform were much better understood. But Nicolas Sarkozy continued to better embody the presidential role: his 23-point lead in the fall remained an 18-point lead in February-March. In other words, the majority of voters from this group liked François Bayrou, perhaps even better than they liked Sarkozy, but the latter had proven himself in his role as president, and this became a decisive argument when it was time to vote.
  • [51]
    Bernard Denni, “Le conservatisme des seniors, une affaire d’âge?”, in Anne Muxel (ed.), La politique au fil de l’âge (Paris: Presses de Sciences Po, 2011), 123-6.
  • [52]
    Nonna Mayer, “Comment Nicolas Sarkozy a rétréci l’électorat Le Pen”, Revue française de science politique, 57(3-4), 2007, 429-45.
  • [53]
    This methodological annex was drafted by Bernard Denni with the help of Catherine d’Aubigny, a lecturer in statistics at the Université Pierre-Mendès-France, regarding the finalising of this typology and the organisation of this methodological annex (whose content remains the authors’ responsibility).
  • [54]
    For a general overview of this method of data analysis, see Ludovic Lebart, Marie Piron, Alain Morineau, Statistique exploratoire multidimensionnelle. Visualisation et inférences en fouille des données (Paris: Dunod, 4th ed., 2006), and in particular chapter 6 “Méthodes de classification”.

Voting and voters have changed. The growing level of hesitation, the ups and downs of turnout and of partisan support are clear evidence of this change. In this article, we attempt to understand this “new deal” via the presentation and use of a particular tool: voting probability questions. These questions present a very different portrait of voters than the one drawn by the classical voting intentions questions. A negative logic of voting is now visible, as is the space of voting possibilities of individuals. Furthermore, these questions shed new light on the dynamics of the 2012 presidential campaign and on the ongoing ideological reconfiguration both on the left and on the right. These analyses are based on the TriÉlec surveys conducted between July 2011 and March 2012.

Vincent Tiberj
A FNSP research fellow at Sciences Po Paris, in the Centre d’études européennes (CEE), Vincent Tiberj also coordinates the TriÉlec network which brings together researchers from Sciences Po Bordeaux, Sciences Po Grenoble and Sciences Po Paris. His main publications are: Des votes et des voix. La France des urnes de Mitterrand à Hollande (Paris: Champ Social Éditions, forthcoming); La crispation hexagonale: France fermée contre France plurielle, 2001–2007 (Paris: Plon/FJJ, 2008); (with Sylvain Brouard) Français comme les autres? Enquête sur les citoyens d’origine maghrébine, africaine et turque (Paris: Presses de Sciences Po, 2005). He specializes in the analysis of political and voting behaviour as well as political psychology in France, Europe and the United States. He works on the reasoning methods of “ordinary” citizens, the political sociology of social and ethnic inequality, as well as xenophobic prejudice and value systems (CEE, 28 rue des Saints-Pères, 75007 Paris).
Bernard Denni
Bernard Denni is a professor of political science at Sciences Po Grenoble and a CNRS researcher with the PACTE group (UMR 5194, Sciences Po Grenoble/Université Pierre-Mendès-France/Université Joseph-Fourier). In 2012, he led TriÉlec’s Grenoble-based team. He has recently published: “Comportement politique et préférences électorales des seniors en 2012”, in “Le pouvoir gris. Du lobbying au pouvoir sur soi”, Gérontologie et société, 143, December 2012, 39-50; (with Pierre Bréchon) “L’élection dans son contexte: candidats et forces en présence”, in Pierre Bréchon (ed.) Les élections présidentielles en France: cinquante ans d’histoire politique (Paris: La Documentation française, 2013), 19-76; (with Annie-Claude Salomon) “Présidentielle: la campagne électorale laisse les électeurs sur leur faim”, Revue politique et parlementaire, 1063-4, April–September 2012, 271-80; “Le conservatisme des seniors. Une affaire d’âge?”, in Anne Muxel (ed.) La politique au fil de l’âge (Paris: Presses de Sciences Po, 2011), 113-60. He specialises in political opinions, attitudes and behaviours, in particular studying the voting patterns of elderly voters (PACTE, Sciences Po Grenoble).
Nonna Mayer
Nonna Mayer is a director of research with the CNRS, affiliated with the Centre d’études européennes de Sciences Po Paris (CEE). She is also the president of the Association française de science politique since 2005 and is in charge of the “Contester” collection published by Presses de Sciences Po. She has recently published: Sociologie des comportements politiques (Paris: Armand Colin, 2010); “Comment aborder les métamorphoses du vote ouvrier en Europe?”, in Jean-Michel de Waele, Mathieu Vieira (eds), Une droitisation de la classe ouvrière en Europe? (Paris: Economica, 2012), 27 –40; “From Jean-Marie to Marine Le Pen: electoral change on the far right”, Parliamentary Affairs, 66, 2013, 160-78. Her research interests include the sociology of political behaviour, racism, anti-Semitism, right-wing extremism and the relationship between social insecurity and voting (CEE, 28 rue des Saints-Pères, 75007 Paris).
Translated from French by 
Sarah-Louise Raillard
This is the latest publication of the author on cairn.
Latest publication on cairn or another partner portal
Uploaded on on 05/03/2014
Distribution électronique pour Presses de Sciences Po © Presses de Sciences Po. Tous droits réservés pour tous pays. Il est interdit, sauf accord préalable et écrit de l’éditeur, de reproduire (notamment par photocopie) partiellement ou totalement le présent article, de le stocker dans une banque de données ou de le communiquer au public sous quelque forme et de quelque manière que ce soit.
Loading... Please wait