CJS 30, 3 (2005) Feature Article

from 1970 to 2003.[3] Since the second problem, the misestimations of party support, is not unique to Quebec and has been the object of frequent analyses, although with different diagnostics, it will be examined first. The other problem, the satisfaction/performance gaps, which has not received any attention, will be examined subsequently. Finally solutions to the poll misestimations of party support in Quebec will be derived from the analysis.

1. Campaign Polls and the Misestimations of Party Support in Quebec

Let us start with an examination of the patterns of aggregate overestimation and underestimation of party support on the eve of the nine Quebec elections between 1970 and 2003, the first dependent variable. The relevant data are presented in Table 1. For each election and each party, the first and the second cells in the table present the average degrees of over- or underestimation of party support in campaign polls, the first cell presenting these misestimations by comparing the official actual vote with net vote intentions, that is, after proportional (prorata) reallocations of undeclared intentions, and the second cell, by comparing the official actual vote with raw intentions, that is, before any reallocation. [4]

The patterns exhibited by those data are quite interesting. In the four elections considered in section A in the Table, the Parti Québécois net support was largely overestimated, the overestimation reaching 4.3% or more and even about 7% in the first two cases, for an average of 5.8%. When however the estimations were based on comparisons with raw vote intentions, the Parti Québécois support was now under- rather than overestimated. But the important

3. The hypotheses regarding the misestimations of popular support were supported in a detailed examination of the 1998 election (Pinard, 2003a) as well as in an analysis of the three referendums of that period in Quebec (Pinard, 2000a).Back to text

4. Thus a plus sign indicates that, compared to the actual vote, the polls overestimated the support for a given party, and a minus sign indicates an underestimation. For instance, the first two cells for the PQ indicates that in 1970 its support was overestimated by 7.3%when the actual election result is compared to net vote intentions, after reallocations, but underestimated by 1.8% when compared to raw intentions, before reallocations. For the 1985 and subsequent elections, only the last few polls of the campaigns, presumably the most accurate, were taken into account; for the 1981 election and those before, all polls of a campaign had to be considered since there were only a few each time. The number of polls considered for each election is indicated in the last column of the Table. For a similar analysis, but one which considers only estimations after reallocations and ends in 1994, see Drouilly, 1997: 182-183. (My figures differ from his because he considers only the errors of the poll averages rather than the mean errors of the polls, that in many cases only the last campaign polls are here considered, and that his poll results present some errors, mainly for the 1989 election, his results then being at times adjusted rather than at the prorata. The patterns are however quite similar.) For the 1976 election, the IQOP poll is disregarded because of serious problems.Back to text

Table 1. Aggregate Overestimation and Underestimation of Support for Parties in Pre- Elections Polls, after and before a Prorata Reallocation of Undeclared Vote Intentions,Quebec, 1970-2003.

finding here is that in three of the four elections, the underestimations of raw intentions, before any reallocation, were much closer to, or equally distant from, the actual results than the corresponding overestimations after prorata reallocations. The average underestimation drops to 4.3% and even lower, to 2.2%, when the 1976 election is disregarded. This means that the best esti-

Political Ambivalence towards the Parti Québécois 285

mations of Parti Québécois support were the intentions originally expressed, without any reallocation of the undeclared intentions. Conversely, in those same elections, Liberal support was as badly estimated as the Parti Québécois after prorata reallocations, but underestimated rather than overestimated, for an average of 5.6%. Moreover, Liberal support was even more seriously underestimated before reallocations, this by as much as 13.6% on the average. This means that in section A, the best estimations of Parti Québécois support were the raw vote intentions, which even without reallocations were relatively high and at a level closer to their actual support. But for the Liberals, the raw intentions were very low and short of their support and even prorata reallocations, while better, were not enough of a correction. As for third parties, prorata reallocations did a good job, the intentions being on the average within 1.1% of the actual results. [5]

The situation is quite different in the case of the other five elections, as revealed in section B of the Table. In this case, it was the estimations of the net Parti Québécois vote that provided the best forecast, contrary to the findings in section A. Support for that party was then quite accurately assessed, within 1.3% on the average. The estimations before reallocations, which were low to start with, presented serious underestimations of the vote for that party, by an average of 7.0%. For the Liberal Party, however, the situation was not different than in A, although its raw vote was higher. Its net support was again underestimated, although slightly less, by an average of 3.0%. But this was much better than the underestimations before reallocations, which were as serious as in A, the average being at 13.0%. The reason for this was that, for the elections in section B, it was the pattern of support for third parties that reproduced the pattern of support for the Parti Québécois observed in section A.Third parties were overestimated after reallocations, by an average of 3.3%, but were more accurately estimated before reallocations, the average error being of only 2.7%. In these five elections, therefore, the Parti Québécois support, as revealed by the raw data, was too low and the electoral outcomes suggest that it succeeded, rather than third parties, in gathering in the end its proportional shares of the undeclared intentions. [6]

5. In two elections, the net third party vote was slightly overestimated, again at the expenses of the Liberal vote, and in the other two, it was slightly underestimated, reducing the underestimations of the Liberal vote. Back to text

6. The excellent aggregate estimations of PQ support in the second section of the Table compare well, for instance, with analogous estimations often observed in U. S. presidential elections. Thus between 1936 and 1972, Gallup's average misestimations were of only 2.4% (Schuman and Johnson, 1976: 168). The problem addressed in this paper is to explain why the misestimations were so often larger in Quebec, even though they remained always smaller than those observed in most attitudes and behaviour relationships, as noted in ibid and discussed in footnote 18. Back to text

286 Canadian Journal of Sociology

2. Factors Accounting for Poll Misestimations

While in Quebec much effort has been made to find ways of correcting such errors (see in particular Drouilly, 1997), not much attention has been paid to possible substantive explanations for all Quebec elections. Before presenting our own explanation, the literature on relevant cases in Quebec and elsewhere will be briefly examined. As will be seen, the main conclusion to be drawn from that research is that there are no general causes prevailing in all or even in most cases. Indeed, causes often vary greatly from one situation to another.

2.1 An Examination of Previous Studies

The methodological problems of polls, and in particular sampling problems, are among the causes most often hypothesized to be the sources of misestimations. Some studies reported that quota sampling produced more errors than random sampling, leading to an overestimation of Labour Party support in Britain and an underestimation for the Front national in France (Jowell et al., 1993; Curtice and Sparrow, 1997; Curtice, 1997; Boy and Chiche, 1999). Yet other analysts have concluded that the errors observed in other studies were not due to either quota sampling or other sampling problems (Crespi, 1988; Traugott and Price, 1992; Marsh, 1985). [7]

Durand and her colleagues (Durand, Blais and Vachon, 2001; 2002; also Durand and Blais, 1999) did however conclude that sampling problems were involved in the 1998 Quebec election. The underestimation of the Liberal vote was, according to them, the result of undersampling and survey nonresponse among subgroups disproportionately favourable to that party. They insisted in particular on the undersampling of persons living in collective households and persons whose phone numbers were unlisted, the size of both groups increasing rapidly in Quebec.

Their conclusions cannot be accepted for many reasons, but only the main one will be mentioned here. [8] Their conclusions rest on the analysis of a single election. If valid, they would imply a steady growth of the size of the errors of polls since 1970, given the long-term regular growth of the groups not reached by polls. [9] But since 1970 the size of the errors has been decreasing, and in

7. Other methodological issues studied, such as sample size, weighting, the number of callbacks, the length of the field work and the days of the week during which it is carried out are issues which appear to present no problems in Quebec polling, given the high standards Quebec firms follow (on these issues in general, see Crespi, 1988; Traugott, 1987). On other related problems, see Jowell et al., 1993; Lau, 1994; Bolstein, 1991; Fox, 1999. Back to text

8. For a detailed critique of their arguments, see Pinard, 2003a. Back to text

9. My argument of course presumes that the propensity of the undersampled groups to disproportionately support the Liberals has been relatively constant during the whole period studied, something which is not unreasonable (for instance older people, in collective households or not, being more prone to vote Liberal). Back to text

Political Ambivalence towards the Parti Québécois 287

addition, as can be seen in Table 1, that decrease has not been steady. In particular that Table shows that while the errors were among the largest in 1998, they had been much smaller in the three preceding elections as well as in the subsequent one. [10] So if methodological problems were involved, they were not those identified in that study, but were of a different nature, involving the construction of questionnaires and special reallocations of the undeclared intentions, as discussed later.

Rather than attributing the sources of the errors to sampling problems, some researchers have claimed that the poll misestimations they examined stemmed from the fact that late shifts in voters' preferences had occurred between the last polls of a campaign and voting day. It is often difficult to verify that hypothesis, but since according to some studies the polls carried out closer to election day do better, it is possible to infer that preferences can change during a campaign and therefore also after the last polls (Crespi, 1988: chap. 9; Lau, 1994; Fox, 1999). Pollsters in Quebec have also suggested this hypothesis to explain the large errors of 1998, but they provided no evidence for it (see Giguère, 1998, from CROP, and Therrien, 1999, from SOM). Conversely, this was rejected by Durand and her colleagues (2001; 2002) in their analysis of the 1998 Quebec election, and a British study did not consider late changes to have been important (Jowell et al., 1993). But again there is no consensus in this regard and some have accepted it as plausible (Noelle- Neumann, 1984; Marsh, 1985). We also accept it, but we consider this as a secondary cause of errors in Quebec, for reasons to be mentioned below.

Some studies interpret the errors of polls in more theoretical terms. Thus Noelle-Neumann's (1984) analyses of German voting behaviour rely on the notion of a spiral of silence. The theory states that individuals supporting parties (or ideas) they believe to be unpopular and controversial will be more likely not to reveal their preference for fear of rejection and isolation. As a result, in a process taking the form of a spiral, the popular parties, for instance, will appear more and more popular and voters will increasingly come to expect them to win elections. At the very end of a campaign, a proportion of the supporters of unpopular parties will finally desert them, in a bandwagon effect, to support the popular ones. Thus in Germany, in the 1965 and 1972 elections, the support of the popular winning parties, the Christian Democrats and the Social Democrats, respectively, were even underestimated in the last polls of the campaigns.

10. The decrease in the size of the errors is as follows: for the first four elections (1970-1981), the average prorata misestimation of Liberal support was of 4.8%, but it dropped to 3.6% for the next five elections (1985-2003).The corresponding misestimations for the PQ were of 4.8% and 2.1% (all as absolute means). Back to text

288 Canadian Journal of Sociology

This theory was put forward to explain the underestimated support of federalist options in referendums and of the Liberal party in Quebec at the federal and provincial level (Drouilly, 1997: 193-195; Durand, 2002). It has also been advanced to account for the overestimated support of various major parties during the 1990s, such as the French and Spanish right parties, the British Labour Party, and the American Democrats, but also the underestimated support of radical French parties (the Communists in the past and the Front national recently) (Boy and Chiche, 1999: 243-245; Curtice and Sparrow, 1997).

But while in Noelle-Neumann's theory, only popular and winning parties are underestimated, in Quebec, since 1970, support for the federalist options and the provincial Liberal Party has been systematically underestimated, whether they were the least or the most popular at the time, and particularly so when the Liberals were the least popular. In addition, in most instances mentioned in Boy and Chiche, and in Curtice and Sparrow, the popular, winning parties are the ones which are overestimated, not the unpopular ones, as in Noelle-Neumann. [11] Hence the spiral of silence can neither be relied upon to explain poll misestimations in Quebec nor most of those in the other instances just cited.

According to a somewhat different hypothesis, misestimations in polls would result from the unwillingness of some respondents to reveal their true preference because they feel social pressures to give false, but socially acceptable or desirable, responses. Thus one could choose for that reason not to reveal a vote against a Black candidate in the United States or even a vote for the British Conservatives in some cases, or again, as was seen, for the Communist Party or the Front national in France or for Sinn Féin in Northern Ireland (Traugott and Price, 1992: 249-252; Howell and Sims, 1994; Jowell et al., 1993: 249-250; Breen, 2000). [12] This argument, rather than the spiral of silence, could be used to explain the cases of misestimations of the preceding two paragraphs, particularly when strong emotional feelings are involved. While at first sight, it could also be suggested as an explanation of the 1998 Quebec situation, with some francophone voters feeling social pressures to claim to be supporting the Parti Québécois, the theory presented here is more consistent with the available evidence, such as that revealing that the sovereignty issue became increasingly salient during the 1998 campaign (Pinard, 2003a) and the fact to be established below that overestimations of Parti

11. In these cases, Noelle-Neumann's theory was possibly misinterpreted on the basis of references to a sketchy presentation of her views (Noelle-Neumann, 1991). Back to text

12. Note that this argument has been often used to explain the frequently observed tendency of some voters to claim to have voted when in fact they did not (Presser and Traugott 1992; Katosh and Traugott 1981). But see also Belliet al., 1999, who suggest that memory failures rather than deliberate lies could be involved. Back to text

Political Ambivalence towards the Parti Québécois 289

Québécois support vary from election to election, according to prevailing degrees of ambivalence.

2.2 Nondisclosers and their Impact on Poll Misestimations in Quebec

A final hypothesis suggested to account for poll errors focuses on those who do not declare a vote intention. A pro rata reallocation of these nondisclosers [13] assumes of course that they will actually vote like those declaring their intention. Some analysts wondered whether the errors observed did not stem from the fact that they do not vote like the others. A few of them concluded that the nondisclosers' behaviour had no effect or only little effect on poll estimations (Crespi, 1988: chap. 9; Jowell et al., 1993; Lau, 1994; Durand, Blais and Vachon, 2001; 2002). Others have claimed that their voting behaviour exerts important effects, but in their view the nondisclosers are different because they disproportionately support unpopular parties or candidates (Curtice and Sparrow, 1997; Boy and Chiche, 1999; Noelle-Neumann, 1984). But Fox (1999) has accepted as conceivable the hypothesis that the underestimation of the NO vote in the 1995 Quebec referendum could have been due to the undecided who were more likely to have supported that option.

The position adopted here, one which has for long been well accepted in Quebec (see, for instance, Pinard and Hamilton, 1980; Drouilly, 1997, chap. 9 [14]), is that nondisclosers exert important effects, but simply because they exhibit sociodemographic characteristics and attitudes which make them more likely than others to support the Liberal Party and federalist options. To give but one example, the last CROP poll of the 1998 campaign revealed that nondisclosers were more likely to be older, to have a low level of education and to be inactive (not in the labour force). They were also more likely to be dissatisfied with the Parti Québécois government and to be NO voters in a potential referendum on sovereignty-partnership. In particular, while the consistent [15] nondisclosers who were both satisfied and YES supporters comprised only 2% of the nondisclosers, the consistent ones who were both dissatisfied and NO supporters comprised 30% of them (Pinard, 2003a, Table

13. The term nondisclosers is used to refer to all those who do not declare a vote intention, that is, those who say they do not know how they will vote, refuse to answer or claim they intend not to vote or to void their ballot. Back to text

14. Drouilly (1997: 184-185) established that most of the times Liberal or federalist vote intentions vary inversely with the proportion of nondisclosers, thus creating what he labelled a "mirror effect." Back to text

15. Consistent, as opposed to ambivalent, nondisclosers are those who are both satisfied with a Parti Québécois government (or dissatisfied with that of other parties) and YES voters, or who are both dissatisfied with it (or satisfied with that of others) and NO voters. Back to text

290 Canadian Journal of Sociology

2). These data clearly suggest that these voters were much more likely than others to support the Liberals. But these nondisclosers were consistent, not ambivalent, and the reason for not revealing their vote intentions may have been because they believed that "one's vote must remain secret" or because they feared negative consequences if their vote became known or simply because of their lack of political interest.

At any rate, this is accepted as a first important factor of the misestimations of vote intentions in Quebec and that claim is supported by the data of Table 1. They show that Liberal support has been systematically underestimated, at both the prorata level and before any reallocation of the nondisclosers, in all nine elections considered. They show above all that before reallocations Liberal support is much more underestimated (-13.2%) than support for the Parti Québécois (-5.8%) or for other parties (even at times overestimated before reallocations). It is therefore reasonable to infer that consistent nondisclosers, disproportionately Liberals, account for these larger misestimations. This is why researchers have for long devised various nonprorata forms of reallocation of nondisclosers in favour of the Liberals (see Pinard, 2003a: 274) and why Drouilly (1998), for instance, decided in 1998 to reallocate as many as 60% of the nondisclosers to the Liberals, but only 30% to the Parti Québécois, with the remaining 10% reallocated to the Action Démocratique du Québec.

2.3 Political Ambivalence: A More Critical Factor

But the proportions of consistent nondisclosers are not likely to vary much from one election to the other and therefore cannot account for the fact that the misestimations at the prorata level are much larger for the two main parties in section A of Table 1, and particularly so for the Parti Québécois. For this we have to turn to hypotheses about political ambivalence and to ambivalent nondisclosers. In 1998, for instance, some of the nondisclosers were clearly in an ambivalent position. In particular, at least 13% were satisfied with the Parti Québécois government, but supporters of the NO side on sovereignty. Many more may have been in the same position, but we do not know since half of the nondisclosers on their vote intention were also nondisclosers on their degree of satisfaction and/or their stand on sovereignty (Pinard, 2003a, Table 2). It is precisely the ambivalence of these nondisclosers which is most important to further explain the poll errors, not to mention an important motivational dimension of voting behaviour in Quebec. [16]

16. In Quebec, it is only towards the PQ that this ambivalence is very strong, although it could prevail in other places towards parties which, while otherwise very attractive, propose projects perceived to be too radical. The French Front national clearly lacks the strong attraction to fit that description and it is indeed underestimated. Back to text

Political Ambivalence towards the Parti Québécois 291

There are a few reasons why, in order to account for poll misestimations, it is assumed that nondisclosers are more important than late shifters. First the consistent nondisclosers are clearly an important part of the story, as just argued. Second it makes sense to think of ambivalence as more likely to lead first to indecision than to sudden shifts from one party to another. Third when the nondisclosers are reallocated according to their referendum vote intention, as suggested later, these corrections lead to good estimates of the actual vote, without assuming in addition the presence of late shifts.

The notion of ambivalence is not new, but it has never been used to explain the misestimations of preelection polls. It occupies a central place in Zaller`s theory of mass opinions, one of the most important recent contributions to that field. The very first deduction from his theory is that "most people`s attitudes on most issues" will exhibit "a tendency towards some degree of ambivalence". According to this political scientist, individuals tend to internalize contradictory considerations that incite them to both favour and oppose a given issue. As a result their attitudes often appear unstable, while in fact they stem each time from the considerations most salient to them at the moment they express them. The responses of these individuals to a pollster, for instance, may change. This however is not because of the lack of an attitude, but simply because of the lack of a single and stable attitude. Each divergent attitude expressed is but the result of favourable or unfavourable considerations which are "at the top of their head" when they express it (Zaller, 1992: esp. chaps. 3 to 5; citation, p. 59). [17]

The reader should notice that Zaller states explicitly that his views on ambivalence challenge both the claim of Converse's classic "nonattitudes" thesis and Achen's opposite claim that most people have "true attitudes" (Converse, 1964; Achen, 1975). Nonattitudes explain for Converse the frequent instability in the expressed views of many people, while Achen claims that most people have true attitudes that are stable. For Zaller, who writes that his position falls somewhere between these two, there is instability in the attitude statements of most people, not for the lack of attitudes, but because they respond to an issue on the basis of the considerations which are most salient to them when they make this response. In so doing, however, they express true

17. Note that the classic notion of cross-pressures, when considered at the attitudinal level rather than at the social level, is similar to the notion of ambivalence. Cross-pressures are then described as the presence of conflict between the many psychological forces affecting behaviour; among other things, they delay the time of a partisan choice during an election campaign (see Campbellet al., 1960: 77-88). We prefer to reserve the notion of cross-pressures to refer to conflicting social pressures, as in its classic formulation, and to use the concept of ambivalence at the attitudinal level, since it expresses more clearly the idea of conflicting attitudes. For a different approach, this one from a sociologist, which juxtaposes ambivalence and rationality, a perspective I do not share, see Smelser, 1998. Back to text

292 Canadian Journal of Sociology

feelings each time (Zaller, 1992: 30-32, 91-96). The position adopted here, in line with Zaller's, is that some voters, in making their political choices, are torn between opposite considerations the relative salience of which changes over time. Indeed it would be difficult to consider the attitudes of Quebecers towards sovereignty as nonattitudes.

It should also be mentioned that the present concern with ambivalent political attitudes and their effect on voting behaviour can be seen as belonging to a larger and traditional area of inquiry, that of the relationships between attitudes and behaviour in general. [18]

The ambivalence towards the Parti Québécois is of a more fundamental form than that which prevails with regard to most attitudes considered by Zaller. Given the salience of the sovereignty issue, ambivalence towards that party does not affect, as in his theory, most people, but for the individuals affected, it tends to be more intense than in the case of most attitudes. The ambivalence towards the Parti Québécois stems from opposing forces prevailing among many voters located in the center, that is, on the one hand, a strong attraction to that party during certain elections, and on the other hand, serious reservations towards its fundamental goal, the sovereignty of Quebec. Contrary to these reservations, which are relatively stable, but become increasingly salient during a campaign, attraction varies much from one election to the other. Attraction can be limited to some of the party's most faithful supporters or else it can reach a large proportion of the unengaged or floating voters located in the center of the political spectrum. It is in the latter case that ambivalence becomes very widespread.

What does the attraction to the Parti Québécois consist of? This attraction derives from the forces which lead voters to consider supporting that party. It

18. In the study of the attitude-behaviour problem, a primary concern has been to account for the fact that "only" small to moderate relationships between attitudes and behaviour have generally been observed, as for instance in attitudes and behaviours towards an ethnic group. This has led to paying much attention to measurement problems for both dimensions and to characteristics of attitudes, such as their specificity, strength and clarity. But, as mentioned before, a strong exception to the relatively weak relationships generally observed was found for the act of voting, where "extremely high relationships" were uncovered (Schuman and Johnson, 1976). The possible presence of attitudinal ambivalence has also retained some attention in this research area, but again in terms of its measurement, not of its consequences, as in the present paper (e. g. Thompson, Zanna and Griffin, 1995). Notice that the authors in that tradition make a distinction between attitudes, behavioural intentions and behaviours, the latter being either reported or observed. In terms of that tradition, section 3.1 below bears mostly on the aggregate level relationships between behavioural (vote) intentions and observed behaviour as reported in official election results. In section 3.2, political attitudes (degrees of satisfaction with incumbent governments) are related to both behavioural vote intentions and to the observed voting behaviour, again all at the aggregate level. Back to text

Political Ambivalence towards the Parti Québécois 293

can involve its general orientations, its immediate platform, and/or the qualities of its leader and candidates; above all, if the Parti Québécois is the incumbent party, it involves the degree of satisfaction with its overall performance, and if it is the opposition party, the degree of dissatisfaction with the incumbents' performance. In short, everything which contributes in making the party more or less popular at any given time is involved.

Ambivalence towards the Parti Québécois takes at least two forms. There is first the ambivalence just described. It prevails at the end of an electoral campaign when the party is very popular, but it may remain unresolved among undecided voters until election day. It is the one that produces part of the errors of polls in those campaigns and the one that we are now considering. But there is also a more permanent form of ambivalence, which is resolved long before the beginning of an electoral campaign and leads many voters otherwise attracted to that party to never or rarely support it because of its option. This form will be examined in a subsequent section of the paper.

With regard to the first form, the main hypothesis proposed is as follows. When the Parti Québécois is very attractive, a high level of unresolved ambivalence prevails until the end of an electoral campaign and leads, at the prorata, to an overestimation of its support and an underestimation of Liberal support. Conversely when the Parti Québécois is not very attractive, much less unresolved ambivalence towards it prevails and estimations of its support at the prorata are quite accurate. In that case support for the Liberal Party remains however underevaluated to the advantage of overevaluated third parties. To account for this, a secondary hypothesis, but one which unfortunately cannot be tested with the data available, is that a contrary form of ambivalence towards the Liberal Party then prevails, with those otherwise attracted to it considering it not nationalist enough.

The following rationale lies behind these hypotheses. When the Parti Québécois is very attractive, it will tend to get the support not only of its most faithful partisans, generally without reservations towards sovereignty, but it will also gain the support of other voters at the center. While less partisan, the latter could be strongly attracted to the party, but subject to only mild reservations. Beyond those voters it will however be difficult for the party to make further gains among nondisclosers with strong reservations towards sovereignty and therefore much ambivalence. A prorata redistribution of those nondisclosers will then tend to overestimate the support of the Parti Québécois, since a disproportionate number of those nondisclosers will finally opt for the Liberal Party, which will have been underestimated at the prorata. Let us add that a few of the less partisan, but ambivalent, voters having declared an intention to support the Parti Québécois may make a late shift to the Liberals on election day, which may be compensated by a few Liberals or nondisclosers turning to the first party. In short, if the attraction is strong, ambivalence will

294 Canadian Journal of Sociology

tend to produce at the prorata an overestimation of Parti Québécois support with a corresponding underevaluation of Liberal support.

Conversely if the election is one in which the Parti Québécois exerts little attraction, its initial support in polls will tend to be limited to its strong partisans, likely to be without reservations towards sovereignty. On election day, it may however rally nondisclosers without much reservations, or even favourable to its option, and who may have been previous supporters not very attracted to it in such an election . Therefore a prorata redistribution of the nondisclosers will provide a relatively accurate estimation of its support on election day. But even in that situation the Liberal support will tend to be underevaluated at the prorata, though less so than in the former situation, this time to the advantage of third parties. The reason is that ambivalence now takes a different twist, affecting the Liberal Party. Some previous supporters of the Parti Québécois may be tempted to turn to the Liberals, but may feel reticent to support a party not deemed nationalist enough. They might have remained undecided or they may have expressed a preference for third parties, to finally disproportionately rally the Liberal Party on election day, as a likely winner in such an election. But this means that a prorata reallocation will have underestimated the support for the Liberals to the advantage of third parties rather than to that of the Parti Québécois.

3. The Impact of Political Ambivalence in Quebec: The Evidence

In line with the main hypothesis, it will be shown that the four elections of section A in Table 1 were elections in which the Parti Québécois exerted great attraction. Given the usual strong reservations towards its option, a high degree of ambivalence was therefore generated. Conversely in the five elections of section B, the attraction was more limited and the ambivalence much weaker.

3.1 Late-Campaign Ambivalence and Election Outcomes

More concretely, a high degree of attraction and ambivalence towards the Parti Québécois prevails under the following conditions. First, when voters strongly reject the other major party after a single term in power, such as the Union nationale in 1970, or more often, after two terms, such as the Liberal Party in 1976, while the Parti Québécois appears a worthwhile alternative. Second, when voters remain strongly attracted to the Parti Québécois after a single term, as in 1981 and 1998, while the opposition party fails to be perceived as more attractive. It should be noted that in three of these four elections an attractive Parti Québécois turned out to be the winner (1976, 1981 and 1998). In these four elections the net vote for that party was highly overestimated, by 5.8% on the average, but generally well estimated before any reallocation of the nondisclosers.

Political Ambivalence towards the Parti Québécois 295

Conversely, the ambivalence towards that party tends to be low under the following conditions. First, when voters remain attracted by the other major party that has been in power for only one term, such as the Liberals in 1973 and 1989, while the Parti Québécois fails to appear more worthwhile. Second, when voters reject the Parti Québécois after a second term, preferring the Liberal opposition, as in 1985 and 2003. Third, a somewhat particular case, when voters abandon the Liberals at the end of a second term, but fail to consider the Parti Québécois as a preferable alternative, as in 1994. This last case is actually the only election with a low degree of unresolved ambivalence at the end of the campaign and in which that party was nevertheless elected. In the other four, the Liberals were the winners. In all the elections of this subgroup, the prorata support for the Parti Québécois provided an estimation which was quite accurate, within 1.1% on the average.

While reservations towards the sovereignist option tend to be stable, the strategies adopted by the Parti Québécois may greatly activate these reservations, as when the election is a referendum one, which implies a decision on sovereignty, as in 1970 and 1973. But even when it is not, the commitment to hold a referendum or the possibility of one if elected does not dampen the impact of the option, as in 1976, 1989, 1994, 1998 and 2003. [19] Only in 1981 and 1985, when a decision on the option was explicitly shelved for the forthcoming term, was its impact much reduced. Incidentally, it is precisely in 1981 that the Parti Québécois obtained its best result ever, with 49% of the vote, while in 1985, its attraction after two terms was very low to start with.

To substantiate our hypothesis, the relationship between the levels of ambivalence towards the Parti Québécois and the size of the poll errors in the estimations of that party's support in the nine elections since 1970 will be examined. [20] Attraction to the Parti Québécois is measured by the degree of satisfaction towards that party, if in power,[21] or the degree of dissatisfaction

19. In 1998, the party declared that a referendum would be held only if winning conditions were met and in 2003, the party planned to reopen the debate on sovereignty if reelected and to hold a referendum if the mobilization was successful. Back to text

20. For detailed narratives on these dimensions for each election between 1970 and 1998, see Pinard, 2000b; for the 2003 election, see Pinard, 2003b. Back to text

21. With the exception of the 2003 election. During that election campaign, satisfaction with the PQ government, at 51% on the average, appeared to be moderate, standing at a lower level than when parties are reelected, but at a higher level than when they are defeated, but this had prevailed for only a few months before the election, and the level of dissatisfaction had been very high during the previous year. More importantly, a large majority of voters had expressed a desire for a change of government all through the two years preceding the election. During the campaign, only 35% on the average expressed the desire to continue with the PQ government, a clear indication of a very low degree of attraction towards the PQ (Pinard, 2003a). In Table 2 I therefore present this together with the satisfaction data for the 2003 election. Back to text

296 Canadian Journal of Sociology

towards the other party in power.[22] The level of support for sovereignty association/partnership measures the absence of reservations towards the option at the time of the election, except for the elections of 1989 and 1994, for which support for sovereignty alone is used.[23] Since the level of support for the option is always lower than the level of satisfaction/dissatisfaction, the difference between the two provides a measure of the net proportion of ambivalent voters in each election, that is, of those who are attracted to the Parti Québécois, but who do not support its option.[24] Table 2 presents the relevant results.

As argued, the first column shows a great deal of variation in the degree of attraction towards the Parti Québécois from one election to the other. It varies between 34% and 71%, and the mean is much higher in section A of the table than in section B (means of 64% versus 47%). The second column indicates that the absence of reservations (i.e. support for the option) is much more stable. The absence of reservations varies only between 32% and 41%, the means for the two sections being practically the same, 36% and 35%. As a result, the degree of ambivalence, the difference between the two, varies a lot, within section A and between sections. This leads to the essential point of the table, the existence of a strong relationship between ambivalence and the prorata overestimations of Parti Québécois support.[25] A much higher level of ambivalence prevailed on the average in the elections of section A(28% versus 12% in B) and the overestimations of that party support were then very large

22. A secondary measure, the popularity of party leaders, could be considered, especially by establishing whether the PQ leader is more or less popular than his party. But according to Nadeau and Bélanger (1999), the vote in Quebec since 1973 was influenced by satisfaction rather than by the leaders' popularity. In turn, they found that satisfaction was largely dependent upon the economic conditions, specifically fluctuations in the unemployment rate. It can be argued however that the attraction to a party may be influenced by the leaders' popularity, especially when the attraction and popularity measures run contrary to each other. Note also that the popularity of Bouchard and Dumont had some effect in 1998, as did that of Charest in 2003 (Pinard, 2003a; 2003b). Back to text

23. In those two elections, under Parizeau's leadership, support for sovereignty alone, without any mention of an association, was the PQ option. Back to text

24. Ideally my theory should be tested with individual level data,showing which nondisclosers were ambivalent and how they finally voted. But unfortunately such data cannot be obtained since a very large proportion of nondisclosers do not reveal either their degree of satisfaction/ dissatisfaction and/or their stand on the option (for instance, 50% in 1998), and data on their actual vote are unavailable for all of them. (When queried after the election on how they voted, which is rare, large numbers again do not reveal it.) The aggregate level evidence presented is therefore the only evidence possible. Back to text

25. It should be noted that in the elections of section A, when the PQ was very attractive, the overestimations of PQ support cannot be explained by an "exaggerated" attraction during the campaigns, because in similar circumstances, in the five elections of section B, an attractive Liberal Party is not overestimated. Indeed it is even then underestimated. Back to text

Political Ambivalence towards the Parti Québécois 297

Table 2. Ambivalence towards the Parti Québécois and Estimation of its Support in Preelection Polls, Quebec Provincial Elections, 1970-2003

(mean of +5.8). Even within that section, a strong relationship can be observed between the levels of ambivalence and the size of the overestimations of Parti Québécois support. Correspondingly, the errors for the Liberal party were also very large in section A (mean of -5.6), but in the opposite direction, that of underestimations.

Conversely, little ambivalence prevailed in the elections of section B,

298 Canadian Journal of Sociology

except for 1994,[26] and in these elections, the average prorata estimations of Parti Québécois support were relatively accurate, ranging only between -1.0 and +2.0.[27] Liberal support was still underestimated in all these elections, although on a smaller scale (mean of -3.0). This was to the advantage of third parties, which were then the ones to be overestimated (see the data in Table 1).[28] It was suggested above that that might have been due to a different kind of ambivalence, one prevailing towards the Liberals this time and involving disenchanted Parti Québécois faithfuls not finding the Liberals nationalist enough. On election day, third parties support tended, like that of the Parti Québécois under strong ambivalence, to be limited to the expressed intentions before any reallocation, gaining none of the undecided.[29] The Parti Québécois on the other hand would finally reconquer only the equivalent of the prorata among the nondisclosers, while the rest of them finally turn to the Liberals, who were presumably perceived as the likely winners and who actually won four of these elections, the exception being 1994. This way, the Liberals did better than even what the prorata would have given them.[30]

3.2 Permanent Ambivalence: Satisfaction/Dissatisfaction and Vote

26. The case of the 1994 election is a particular one. As indicated in Table 2, dissatisfaction with the Liberals, which were at the end of a second term, was very high and this alone would have been a source of much ambivalence. But the attraction to the PQ was seriously dampened by its more radical option during this election, that is, sovereignty without association, and the commitment to hold a referendum within a year. In addition, the option was defended by a more radical party leader, Jacques Parizeau. The election took the appearance of a quasi-referendum on pure sovereignty and the issue became very salient at the end of the campaign, while an increasing majority of voters expected the PQ to win. All this contributed to dampening the attraction of the PQ. In the end it was as if ambivalence had led to a decline of potential PQ support even before the campaign was over. Back to text

27. Put differently, if a plot of ambivalence by misestimations of PQ support is drawn, what can be observed is that almost all points from the data of section B are close to one extreme, the zero point on each axis, while the points from the data of section A are found towards the other end of each dimension, but that they are distributed over a greater range and show that the two variables vary with one another. Back to text

28. This pattern did not seem to vary with the different third parties involved, as all tended to be overestimated. Back to text

29. One could suggest that the overestimation of third party support is simply due to the fact that some respondents "hide" behind third parties in polls or that the absence of candidates for these parties in their district does not allow them to actually vote for them. But if this were true, these parties would always be overestimated, not only when there is little ambivalence. Back to text

30. The case of one third party, the ADQ, is interesting in this regard. In 1998, its support was relatively well estimated at the prorata (+1.0), as was the case for other third parties in elections with much ambivalence. In 1994, it was slightly more overestimated (+2.0), although less than other third parties in elections with little ambivalence; but 1994 is a special case. It must be mentioned however that this party support was underestimated in 2003, the only exception to the pattern of overestimations of third parties in section B. Back to text

Political Ambivalence towards the Parti Québécois 299

Intentions

The analysis of ambivalence effects turned out to be very revealing in another respect. The first hypothesis was that strong late-campaign ambivalence led to an overestimation of PQ support in prorata vote intentions. But an examination of the data unexpectedly led to a second hypothesis. The data suggested indeed that ambivalence starts exerting its effects much earlier.

The new hypothesis generated is that the ambivalence towards the Parti Québécois almost always makes it impossible for that party to succeed in gathering all the vote intentions which either the degrees of satisfaction towards it or the degrees of dissatisfaction towards its adversaries would lead one to expect. Conversely, the Liberals are then often able to gather more vote intentions than the degrees of satisfaction or of dissatisfaction would lead one to expect. The rationale behind this hypothesis is analogous to the one presented above, except that the processes now described are at work well before the last stages of electoral campaigns. Caught on the one hand between high levels of attraction to the Parti Québécois, as indicated by high levels of satisfaction towards its governments or of dissatisfaction towards its opponents' governments, and on the other hand strong reservations towards its sovereignty option, many voters decide long before any electoral campaign that they are unwilling to support it. Instead they turn to their opponents, even though they may not be strongly attracted to them in terms of the satisfaction/dissatisfaction component.

A test of this hypothesis with individual-level data from the 2003 election yields the following results. The last CROP poll of the campaign revealed that among YES voters, the proportion satisfied with the Parti Québécois government was very high (72%) and that the party was indeed able to obtain from YES voters a proportion of prorata vote intentions almost equal to that level of satisfaction (69%). That is, the YES voters, not subject to any ambivalence, gave that party the vote intentions expected from that high level of satisfaction. However among NO voters, the proportion of satisfied voters, although much lower to start with (38%), was still much higher than the proportion intending to vote for that party (only 12%). These satisfied voters were presumably ambivalent because of their NO option and a large proportion of them did not support the Parti Québécois. Indeed, among NO voters, while the proportion of dissatisfied voters stood at 59%, the opposition parties did much better, obtaining 88% of the vote intentions (67% for the Liberals, 17% for the Action Démocratique du Québec and 4% for other parties).[31] Very similar results (not presented) obtained in 1998.

A test of this hypothesis at the aggregate level for all elections is presented

31. Viewed differently, the data indicate that among those who were both satisfied and NO voters, only a very small minority (27%) intended to vote for the PQ (Pinard, 2003b). Back to text

300 Canadian Journal of Sociology

in Table 3. It summarizes this new analysis within the context of the preceding one.

Let us first consider the case of the Parti Québécois detailed in the top part of the Table. In sections 2 and 3, it can be seen that the raw and net vote intentions (the latter, as we have seen, often already overevaluated) obtained by this party were in much lower proportions than the proportions of satisfied or dissatisfied voters who should have been attracted to it. Thus in 1976, for instance, the high level of dissatisfaction towards the Liberal government (68%) brought to the Parti Québécois only 31% of the raw vote intentions and 48% of the net ones. Similarly in 1998 the high level of satisfaction towards the Parti Québécois government (54%) brought it vote intentions standing at only 43% before reallocation and 48% after.

Moreover a new phenomenon emerges. The pattern described did not prevail only when the Parti Québécois was highly attractive, but also when the attraction to it was low, as in the last five columns, the only exception being the election of 1985, when net vote intentions are equal to satisfaction with the party. In all other elections, the differences between vote intentions and satisfaction/dissatisfaction were negative. This indicates that even when a low level of attraction to the Parti Québécois prevailed, the reservations towards its option limited the gains which satisfaction or dissatisfaction should have generated for the party.[32]

It must be stressed however that for the Parti Québécois the differences between the degrees of satisfaction or dissatisfaction prevailing and the vote intentions obtained were on the average much larger when the attraction and ambivalence levels were high (e.g. average difference of -19 in section 3) than when they were low (average difference of -9), something which is consistent with the general hypothesis.

Finally it is quite interesting to note that when the degree of satisfaction is compared to net vote intentions, the Parti Québécois losses are always larger than when these intentions are compared to the actual (official) vote, considered previously (compare the differences in section 3 with the first differences mentioned in section 4). But when ambivalence towards the Parti Québécois is strong, losses at both levels (first in vote intentions and then in actual votes) add up to produce total losses always greater than 10%, as indicated in the total differences of section 4. In 1998, for instance, after a loss

32. The data of Table 3 also reveal that in 1989 and 1994, under the more radical option of sovereignty alone, PQ ambivalence had already exerted all its effects before the end of the campaigns. Indeed the differences between dissatisfaction with the Liberals and PQ net vote intentions were then among the largest since 1981. Back to text

Political Ambivalence towards the Parti Québécois 301

Table 3. Aggregate Satisfaction or Dissatisfaction with Incumbent Governments, Vote Intentions before and after Reallocation, and Actual Vote, Quebec Provincial Elections, 1970-2003

302 Canadian Journal of Sociology

of 6% in net vote intentions, the Parti Québécois lost another 5% in actual votes, for a total loss of 11%.[33]

The situation tended to be the obverse in the case of the Liberal Party (and the Union nationale in 1970). As can be seen in the bottom part of the Table, in elections with high ambivalence, the Liberals (and the Union nationale) gathered net vote intentions larger than what the degrees of satisfaction or of dissatisfaction would have entailed in three elections out of four.[34] Here again, instead of losses, we find that the gains in both vote intentions and actual vote add up to one another. Thus in 1976, while only 26% declared they were satisfied with the Liberal government, that party got 30% of the prorata vote intentions and 34% of the actual vote, for a total gain of 8%. Similarly in 1981, while only 34% were dissatisfied with the Parti Québécois, the Liberals nevertheless obtain 42% of the prorata vote intentions and even 46% of actual support on voting day, for total gains of 12% (see sections 7 and 8 in the Table). This pattern did not however always prevail when ambivalence was low (the last five columns of the Table). Indeed in three elections out of five the Liberal Party then performed slightly worse in net vote intentions than the degree of satisfaction or dissatisfaction would entail (section 7). It did however make gains in actual vote in all elections. The losses made in net intentions occurred to the advantage of third parties, although the Liberals tended to regain that support from them in actual vote, as we have seen previously (see also section 8). Note however that in the case of the Liberals, their gains were mostly registered in the comparison of net vote intentions to actual vote, not in that of satisfaction/dissatisfaction to net intentions.

In short, this evidence supports the hypotheses on the multiple effects of ambivalence towards the Parti Québécois. This party almost always performs less well in vote intentions, even after reallocation, than what the degree of satisfaction or dissatisfaction should entail, while the Liberal Party then tends to do better, and almost always does when attraction to the Parti Québécois is strong. More generally, for the Parti Québécois, the degrees of satisfaction or dissatisfaction almost never translate into vote intentions, and even more so when ambivalence is high, and the prorata vote intentions do not translate in actual support when the party enjoys a high level of attraction. Conversely, in the case of the Liberals, they make gains in prorata vote intentions, but mainly

33. It could be objected that those results are due to the fact that the PQ was not alone in gaining from the dissatisfaction with the Liberals, since that dissatisfaction profited also to third parties. To be sure, this explains why the differences in section 2 are much larger before 1981. But the phenomenon described has continued to prevail ever since, even though the PQ had become one of the two major parties, with very weak third parties. Back to text

34. The 1998 election is the exception, the Action Démocratique du Québec then gaining from that situation rather than the Liberals. Back to text

Political Ambivalence towards the Parti Québécois 303

under high ambivalence, and compared to the degrees of satisfaction/ dissatisfaction, they make gains in actual support in almost all elections. Moreover, the Parti Québécois practically always loses more in the comparison of satisfaction/dissatisfaction with net vote intentions than in the comparison of these intentions to the actual vote. Ambivalence therefore exerts more of a negative effect in the first than in the second set of comparisons. The obverse often prevails for the Liberals.

4. Political Ambivalence and Possible Solutions to Poll Misestimations

What can be done about polls misestimations? It is of course obvious that when the errors are the results of sampling or other methodological problems, the solutions are simply the correction of such problems (see Crespi, 1988: Chap. 3). But the solutions are not that simple when the errors result from substantive problems, such as the spiral of silence or political ambivalence. These are the types of problems one must be concerned with, in particular the fact that nondisclosers often end up voting differently from disclosers.

Many solutions have been considered in the literature. Some have suggested the use of the ballot box technique, the weighting of the results on the basis of reported past vote or party identification and latent class analysis, but the difficulties involved in using these methods have generally been recognized (Jowell et al., 1993; Crespi, 1988; Curtice and Sparrow, 1997; Boy and Chiche, 1999; Breen, 2000). In dealing with the problems of errors in Quebec, early efforts were directed towards the presentation of different hypothetical outcomes on the basis of internal analyses of the polls (Lemieux, 1970; Pinard and Hamilton, 1976; 1980). Others have suggested adjustments to poll results when prorata allocations of the nondisclosers lead to systematic misestimations of some parties (Curtice and Sparrow, 1997; Boy and Chiche, 1999; Drouilly, 1997). In Quebec, the adjustments suggested by Drouilly[35] have been widely used, but they are opened to serious criticisms, the most fundamental being that there is much variation in the size of the errors, while his method assumes that the errors are basically invariant (for a detailed criticism, see Pinard, 2003a).

A few authors have suggested that the allocation of nondisclosers be made on the basis of their political attitudes or positions on issues, their ratings of the candidates and the like, and in particular whether they intend to vote or not (Crespi, 1988: Chap. 5; Jowell et al., 1993; Howell and Sims, 1994; Boy and Chiche, 1999). This solution is part of the proposals made below. Finally some

35. Such as his solution of attributing 60% of the nondisclosers to the PLQ, 30% to the PQ and 10% to other parties, as mentioned before. Back to text

304 Canadian Journal of Sociology

have suggested that vote intention questions should follow rather than precede other political questions. As indicated below, I strongly agree with that suggestion, even though many disagree with it (for a discussion of the pros and cons, see Crespi, 1988: Chap. 5; Lemieux, 1988: 47-48).

To avoid poll errors in Quebec, I propose three solutions which ideally should all be attempted at the same time. Their justifications are largely rooted in the fact that political ambivalence must be taken into consideration. It must be stressed that it is much more important to adopt these solutions when the levels of ambivalence are high, given that the errors to be corrected are then larger.

The first solution seems to be the most important one in the Quebec context. Since the degree of ambivalence is often high among many respondents, it is essential at the time of the interviews to make salient, before any vote intention question, the psychological context which ultimately will prevail among them on election day. To do this, before asking ambivalent respondents how they will vote, it is important to bring clearly to their mind, in a balanced way, both the attractive features of the parties and candidates, and the reservations that they may feel towards them. To be sure, many object to not placing the vote intention questions at the beginning of the questionnaire on the basis that it may bias the respondents' political preference. I claim that this is not a problem, as long as both positive and negative aspects are raised in a balanced way. In line with Zaller (1992), it is important to bring to the top of the mind of respondents considerations which may otherwise not be salient to them at the time of the interview, but which are very likely to be so on election day. It is important that the least attentive voters be immersed into the concrete context of the main considerations which will ultimately influence their final choice. And indeed some analysts have found such a procedure to reduce the estimation errors (Crespi, 1988: 104-107). As a secondary benefit, this procedure will also tend to raise the respondents' confidence in pollsters and reduce the proportion of them who would refuse to reveal their preference (ibid.).

A second solution is to carry out an internal analysis of the interviews with yet undeclared vote intentions. Examining the answers of these respondents to one or many other questions, some hypotheses can be developed on the expected voting behaviour of these "silent" voters and one can then proceed to a reallocation of these intentions to parties (or options) on the basis of these hypotheses. Alternatively combining their answers to many other questions, a single hypothesis can be advanced. If the previous arguments presented are sound, in so doing, the degree of ambivalence towards the Parti Québécois or that between the Liberals and third parties will have to be taken into account. In each case it is of course essential to provide an account of the procedures followed. Let us note immediately that the results, contrary to that

Political Ambivalence towards the Parti Québécois 305

of Drouilly, are very likely to vary from poll to poll and from election to election.

A third solution consists in reducing as much as possible the proportion of silent respondents. Thus in 1998 the Reid polls had lower rates of undeclared vote intentions than others and their last poll turned out to be the most accurate in prorata vote intentions. That poll reported only 5% of undeclared intentions, compared to about 12% in the last polls of other firms, and at the same time reported a larger proportion of prorata Liberal intentions and a lower proportion of PQ intentions, a result which turned out to be more accurate. The Hamilton-Pinard 1980 referendum poll is another example. Only 14% in that poll did not report a vote intention, compared to more than 20% for the other polls at the end of the campaign, and its results turned out to be much more accurate.[36]

To demonstrate the impact of these procedures, the vote intention results of two polls carried out before the 1998 election will be compared. They provide a good test since in that election ambivalence was high and the misestimations tended to be large. The two polls were similar in almost all respects, except in questionnaire construction and in particular in the order of the questions. Both were telephone polls conducted by the same firm, CROP, at practically the same time, mid-October 1998, and with the same type of random systematic sampling. The first was a CROP poll for media outlets, while the second was a poll for the Centre for Research and Information on Canada (CRIC).[37]

However with regard to the order of the questions, the differences were substantial. The CROP-media poll contained only five political questions

36. On this, see Pinard, 2003a. It is not clear why the Reid poll had such a low proportion of silent voters. This could have been the result of quota sampling, which would allow the pollster to reach a larger proportion of confident respondents, more likely to reveal their vote intention. As for the Hamilton-Pinard poll, the fact that many questions preceded the vote intention question reduced the proportion of nondisclosers. Let me add that between 1998 and 2003, CROP conducted separately during the same month six long polls for the Center for Research and Information on Canada (CRIC) and its usual omnibus polls. In four cases, the CRIC polls generated fewer nondisclosers and more Liberal intentions. For a generalization stating a negative relationship between the proportions of nondisclosers and the proportions of federalist vote intentions, see Drouilly, 1997: 184 ff. Back to text

37. The CROP-media poll, for La Presse and TVA, which started with political questions, was carried out between October 15 and 21. The CROP-CRIC poll was much longer and was carried out between October 14 and 27. Both polls were therefore done on the very eve of the 1998 election campaign, which started on October 28. In each case, the sample was drawn randomly from the published lists of all telephone subscribers in Quebec. Their regional sample distributions were slightly different, but the results were weighted by regions as well as by sex, age and language spoken at home. I am very grateful to CROP,for additional data, and to CRIC, for allowing me to use its data in this analysis. Back to text

306 Canadian Journal of Sociology

before that on vote intentions. They were about the degree of satisfaction with the Quebec government, the degree of satisfaction with each of the three main political leaders and the choice of the leader most apt to govern Quebec. In all respects, these questions evoked the high level of attraction towards the Parti Québécois in that election. The context created was therefore one that was biased in favour of that party. On the other hand, the much longer CROP-CRIC questionnaire contained no less than 51 items before the vote intention question. Some of these items bore on subjects which could evoke attraction towards the PQ, as for instance the questions on attachment to Quebec versus Canada and on self-identification as Quebecer or Canadian, questions yielding attitudes favourable to Quebec, as well as many questions on largely shared traditional grievances among Francophone Quebecers. But other questions were likely to activate many respondents' reservations towards that party, in particular those bearing on vote intentions in a possible sovereignty referendum and related aspects, as well as those bearing on attachment to Canada and its objectives. Finally some other questions were neutral in these respects. It turned out that these two polls yielded different results with regard to provincial vote intentions, as revealed by the data of Table 4.

As indicated on the first three lines of the Table, the CRIC poll yielded, before any reallocation, Liberal vote intentions which were 6% higher than those obtained in the media poll, while PQ vote intentions were 2% lower. As a consequence, while the media poll had the Liberals trailing the Parti Québécois by 5 points, the CRIC poll placed the Liberals, again before reallocation, 3 points ahead of that party. There were also in the CRIC poll 3% fewer undeclared intentions, a result consonant with the third solution proposed and which presumably partly explains the Liberal gains in that poll.[38] For the purposes of the quasi-experiment involved in these polls, it is regrettable that there were no new polls at the end of the campaign similar to the original CRIC poll. It is however possible to try to anticipate what such a poll would have produced, if carried out at the same time as the last CROP-media poll of the campaign, which was constructed like the October CROP-media poll just reviewed. The vote intentions in this November poll are presented on line 3 of the Table, again without reallocating the undeclared intentions. If these intentions are altered on the basis of the differences between the first two polls, the hypothetical results of a November CRIC poll are those of line 4. After a prorata reallocation of nondisclosers in both polls, the results of lines 5 and 6

38. With regard to third parties, the results are somewhat more equivocal. The ADQ obtained 3% fewer intentions in the CRIC poll, but the other third parties, 2% more. It should however be mentioned that the CRIC result for the ADQ was closer to the 5%average obtained in the other polls of the preelection campaign period. Back to text

Political Ambivalence towards the Parti Québécois 307

Table 4. A Comparison of Provincial Vote Intentions in Three CROP Polls of October and November 1998 and their Implications

are obtained. It can be seen that the prorata results of the hypothetical CROPCRIC poll would have been very close to the election results shown on line 9. The support of the three main parties would have all been evaluated within 2%. These results would have been much better than the prorata results of the last CROP-media poll, which underestimated the support of the Liberals by 8 points and overerestimated that of the PQ by 6 points (compare lines 5 and 9). The fact remains that, as just seen, the hypothetical CRIC poll would have nevertheless slightly underestimated the Liberal vote and slightly overestimated the Parti Québécois vote. But in line with the second solution, an internal analysis revealed that even those who remained silent in the original CRIC poll exhibited predispositions which were much more pro-Liberal than pro-PQ. Thus for instance 40% of them would have voted NO in a sovereignty

308 Canadian Journal of Sociology

referendum and only 8% would have voted YES.[39] On the basis of the second solution, if the undeclared intentions of the hypothetical CRIC poll are reallocated according to the referendum vote intentions of such respondents in the original CRIC poll,[40] the results of line 8 are obtained. They happen to correspond exactly to the actual vote for the two main parties, although the ADQ vote is underestimated by 2%. By contrast, a similar reallocation for the last CROP-media poll yields results still distant from the actual results (see line 7). The procedures proposed also gave excellent results in 2003.[41]

Other aspects of the data from the two October polls support the hypothesis that better estimations of the vote are obtained when attraction and above all reservations[42] are made salient. As seen before, both polls contained referendum vote intention question, but in the CRIC poll that question came after thirty other items, but before the provincial vote intention question, while that question immediately followed the vote intention question in the media poll. It turns out that both polls obtained practically identical results on the referendum question, that is, 37% YES in both cases, and 49% and 50% NO, for the media and CRIC polls, respectively, the rest remaining silent in each case. These identical results suggest that the differences in provincial vote intentions were not due to sampling errors.[43]

39. This suggests that both the second solution, internal analysis, and the first, construction of the questionnaire, should be used concurrently. Back to text

40. In this special reallocation, the ADQ was first given the equivalent of the prorata from the undeclared intentions. For the rest, those who would have voted YES were reallocated to the PQ and those who would have voted NO, to the Liberals. As for those who did not reveal a referendum vote intention either, a third were allocated to the PQ and two thirds to the Liberals, since they traditionally appear to be more federalist. Back to text

41. In the 2003 election, there was little ambivalence to start with, so that the misestimations of prorata party support were small on the average in the last three polls of the campaign (see Table 1). With regard to the last CROP poll, the prorata support for the PQ was overestimated by 2% and that for the Liberals was underestimated by the same proportion. In addition to the low level of ambivalence, these small errors may have been facilitated by a questionnaire containing twelve items before the vote intention question, some evoking pro-Liberal, others pro-PQ predispositions. In addition a simple reallocation of the nondisclosers (14%) according to their referendum vote intentions led to estimations corresponding exactly to the actual vote for the Liberals (46%) and for the PQ (33%), the ADQ being underestimated by 1% and the other parties and candidates, overestimated by the same proportion. Back to text

42. As will be seen below, it is above all important that the context of reservations be made salient in an election since these elements are more likely to remain somewhat latent until the end of the campaign. Back to text

43. On the other hand, this also suggests that outside of a referendum campaign, the first solution proposed would possibly be less important to measure vote intentions on sovereignty. Alternatively, and more likely, it is possible that the level of YES support was so low that the sovereignist camp was now reaching only convinced supporters, with little ambivalence, who therefore did not need to be placed in a context in which all considerations on this issue were made salient. Back to text

Political Ambivalence towards the Parti Québécois 309

What is however more significant here is that the relationship between the referendum and the provincial vote intentions was stronger in the CRIC poll than in the media poll. In particular, while 82% of the CRIC YES voters intended to vote for the PQ, the corresponding proportion was only 74% in the media poll, with 9% for the Liberals in each case. Conversely, among the NO voters, 68% chose the Liberals in the CRIC poll as against only 60% in the media poll. For the PQ, the corresponding percentages among NO voters were 10% and 19%. This means that in the CRIC poll, YES voters were more likely to choose the Parti Québécois than NO voters, without reducing the proportion of them favouring the Liberals (9% in each case). But the NOs in the CRIC poll were more likely to opt for the Liberals, but less for the PQ, something which is consistent with the hypothesis regarding the impact of reservations towards that party.

Moreover as there were more NOs than YESs in both polls, this favoured the Liberal Party. Let us add that among the undeclared voters on the referendum questions, 17% chose the Liberals in the CRIC poll, but only 11% in the media one, something which is again consistent with the hypothesis. The corresponding percentages for the PQ were 11% and 12%. Finally among those silent on the referendum question, a slightly smaller proportion remained silent on their choice of a party in the CRIC poll (63%) than in the media one (71%), something which suggests that the procedure employed rendered them more confident to reveal their vote intention. All in all, the prior evocation in the CRIC poll of considerations bearing on reservations towards the PQ, that is, referendum choices, produced a greater impact of these choices on party preferences, just as they would presumably do on voting day.[44]

The impact of the order of the questions can be observed in other instances. As mentioned, the Reid rolling poll obtained the best results among the polls carried out at the end of the 1998 campaign. The smaller proportion of undeclared intentions it yielded may, it was suggested, have contributed to that result. But the construction of the questionnaire may also have been involved. The vote intention questions came also very soon in that questionnaire, that is in sixth place, as in the other media polls. But contrary to the last CROP poll, for instance, the preceding questions did not bear only on attractive features of the PQ, but also on aspects related to reservations towards its option. At the very beginning there were two questions on the important issues of the election. The analysis revealed that the proportions mentioning sovereignty as an important issue were the only ones to increase during the campaign, particularly at the very end, when it became the second most important issue.[45]

Respondents were then asked which party would be the best to resolve the first

44. For a detailed analysis of the impact of the sovereignty issue in 1998, see Pinard, 2003a. Back to text

45. See ibid. Back to text

310 Canadian Journal of Sociology

issue they had mentioned and the Liberal Party turned out to be ahead of the Parti Québécois for the issue of sovereignty.[46] By calling those reservations to mind, the Reid poll thus possibly improved its measure of vote intentions.

Finally the much more accurate results of the Hamilton-Pinard poll at the end of the 1980 referendum campaign can presumably also be attributed in part to the fact that the referendum vote intention question (without reading the official question) followed questions on thirty-one other items covering, among others, both attractive features of the option (traditional grievances, attachment to Quebec, for instance) and reservations towards it (the economic consequences of sovereignty, for instance) (Pinard and Hamilton, 1984). By contrast, in the other polls done at the same time, the very long official question was read at the very beginning or only after a few questions, such as one on the degree satisfaction towards the government, which was then very high.

Conclusion

A new theoretical framework developed around the notion of political ambivalence was presented. The notion of ambivalence is central to Zaller's important contribution to the theory of mass opinions. Among other results, it led him to a reconciliation of Converse's "nonattitude" thesis and Achen's opposite view that most people have "true attitudes". In this paper, the notion of political ambivalence turned out to be essential to account for the very frequent failure of the Parti Québécois, unlike most other parties, to reap the electoral benefits which should normally be expected whenever the party's image is positive. To start with, in nine Quebec elections between 1970 and 2003, that party never, except for one election, gathered the net vote intentions which the degree of satisfaction towards it or of dissatisfaction towards its opponents should have generated. Subsequently it often failed to obtain the actual support which these already deflated vote intentions should have led it to expect. Conversely the Liberal Party often got more net vote intentions than what the degree of satisfaction towards it or dissatisfaction towards the Parti Québécois should have entailed and then always obtained a proportion of the actual vote larger than that which was anticipated by these vote intentions.

To be sure these patterns are in part due to the fact that those who do not reveal their political preference tend to vote disproportionately for the Liberals, something which has long been recognized. It was however hypothesized that these patterns are mainly caused by a more critical factor, the degree of political ambivalence which the Parti Québécois generates. Even when that

46. There were no questions bearing on PQ attraction before the vote intention question, but as mentioned it is more important to cover reservations during an election campaign, while the attraction aspects of the parties are then more likely to be salient at any rate. Back to text

Political Ambivalence towards the Parti Québécois 311

party is perceived as very attractive by large numbers of voters, it remains, hypothetically, handicapped by strong reservations among many voters towards its sovereignty option. The evidence presented indeed showed that the degree of ambivalence prevailing at the end of electoral campaigns is strongly and positively related to the size of the overestimations of Parti Québécois support and, to a lesser extent, to the size of the underestimations of Liberal support. The analysis also showed that permanent ambivalence prevents the Parti Québécois from translating satisfaction or dissatisfaction into vote intentions for the party in all elections, but more so when ambivalence is high, while the Liberal Party does better than the degrees of satisfaction or dissatisfaction would entail, but mainly when ambivalence is at a high level.

It is important to mention that in trying to account for misestimations in party support, a critical substantive process of theoretical significance was uncovered. That is, while political ambivalence accounts for poll errors, it also constitutes a central dimension in the analysis of the motivational factors underlying political preferences in the Quebec electorate, and this, not only during electoral campaigns, but also long before.

Finally, on the basis of that analysis, the paper presented new solutions to the problem of misestimations so often encountered by pollsters in their efforts to forecast party support in Quebec.

It remains to be investigated whether political ambivalence should also be given close consideration in other electoral contexts. In general, it should prove to be important whenever a party, otherwise quite popular, arouses strong reservations towards a radical portion of its platform. The most immediate potential examples which come to mind are those of secessionnist parties in other countries. It could also be possibly relevant in the case of some popular radical right parties in Europe.

At the federal level in Canada, the Bloc Québécois seems to have been the object of ambivalence which, while possibly weaker, engendered electoral consequences similar to those observed with the Parti Québécois in at least the elections of 1993 and 2004, during which it was extremely popular (Pinard, 2004). Similarly, again at the federal level, the reservations of many Ontario voters to the social conservatism of the new Conservative Party, born of a merger with the right-wing Alliance Party, may have been responsible for its decline during the late phases of the 2004 campaign and for the underestimations of the support for the federal Ontario Liberals (see ibid.).

References

Achen, Christopher. H.
1975
"Mass Political Attitudes and the Survey Response." American Political Science Review 69: 1218-1231.

312 Canadian Journal of Sociology

Belli, Robert F., Michael W. Traugott, Margaret Young and Katherine A. McGonagle
1999
"Reducing Vote Overreporting in Surveys: Social Desirability, Memory Failure, and Source Monitoring." Public Opinion Quarterly 63: 90-108.
Bolstein, Richard
1991
"Comparison of the Likelihood to Vote among Preelection Poll Respondents and Nonrespondents." Public Opinion Quarterly 55: 648-650.
Boy, Daniel and Jean Chiche
1999
"La qualité des enquêtes d'intention de vote: le cas des régionales de 1998." In: Sofres. L'état de l'opinion 1999. Paris: Sofres, pp. 237-253.
Breen, Richard
2000
"Why Is Support for Extreme Parties Underestimated by Surveys? A Latent Class Analysis." British Journal of Political Science 30: 375-382.
Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes
1960
The American Voter. New York: John Wiley.
Converse, Philip E.
1964
"The Nature of Belief Systems in Mass Publics." In: David Apter. Ideology and Discontent. New York: Free Press, pp. 206-261.
Crespi, Irving
1988
Pre-Election Polling : Sources of Accuracy and Error. New-York: Russell Sage.
Curtice, John
1997
"So How Well Did They Do? The Polls in the 1997 Election." Journal of the Market Research Society 39: 449-461.
Curtice, John and Nick Sparrow
1997
"How Accurate Are Traditional Quota Opinion Polls?" Journal of the Market Research Society 39: 433-448.
Drouilly, Pierre
1997
Indépendance et démocratie: sondages, élections et référendums au Québec, 1992-1997. Montréal: Harmattan.
1998
"Sondages. Il faudra ý l'avenir attribuer tous les votes des non-francophones au PLQ." La Presse, December 16.
Durand, Claire
2002
"Are the Polls Biased Against Quebec Liberals." Policy Options March: 51-56.
Durand, Claire and André Blais
1999
"Why Did the Polls Go Wrong in the 1998 Quebec Election? The Answer from Post- Election Polls." Bulletin de méthodologie sociologique April, no.62: 43-47.
Durand, Claire, André Blais and Sébastien Vachon
2001
"A Late Campaign Swing or a Failure of the Polls? The Case of the 1998 Quebec Election." Public Opinion Quarterly 65: 108-123.
2002
"Accounting for Biases in Election Surveys: The Case of the 1998 Quebec Election." Journal of Official Statistics 18: 25-44.
Fox, John, Robert Andersen and Joseph Dubonnet
1999
"The Polls and the 1995 Quebec Referendum." Canadian Journal of Sociology 24: 411-424.
Gagné, Gilles and Simon Langlois
2002
Les raisons fortes. Nature et signification de l'appui ý la souveraineté du Québec. Montréal: Les Presses de l'Université de Montréal.
Giguère, Alain
1998
"Une tempête dans un verre d'eau." La Presse, December 5.

Political Ambivalence towards the Parti Québécois 313

Howell, Susan E. and Robert T. Sims
1994
"Survey Research and Racially Charged Elections: The Case of David Duke in Louisiana." Political Behavior 16: 219-236.
Jowell, Roger, Barry Hedges, Peter Lynn, Graham Farrant and Anthony Heath
1993
"The 1992 British Election: The Failure of the Polls." Public Opinion Quarterly 57: 238-263.
Katosh, John P. and Michael W. Traugott
1981
"The Consequences of Validated and Self-Reported Voting Measures." Public Opinion Quarterly 45: 519-535.
Lau, Richard R.
1994
"An Analysis of the Accuracy of 'Trial Heat' Polls during the 1992 Presidential Election." Public Opinion Quarterly 58: 2-20.
Lemieux, Vincent
1970
"Trois hypothèses sur le vote des indécis." Le Soleil, April 18: 18.
1988
Les sondages et la démocratie. Québec: Institut québécois de recherche sur la culture.
Marsh, Catherine
1985
"Predictions of Voting Behaviour from a Pre-election Survey." Public Opinion Quarterly 33: 642-648.
Nadeau, Richard and éric Bélanger.
1999
"L'effet Charest… Une conclusion." Le Devoir, April 6: A9.
Noelle-Neumann, Elisabeth.
1984
The Spiral of Silence. Public Opinion - Our Social Skin. Chicago: University of Chicago Press.
1991
"The Theory of Public Opinion: The Concept of the Spiral of Silence." Communication Yearbook 14: 256-287.
Pinard, Maurice
2000a
"Ambivalence et référendums au Québec, 1980, 1992 et 1995." In: Jack Jedwab. ¿ la prochaine? Une rétrospective des référendums québécois de 1980 et 1995. Montréal: Les éditions Saint-Martin, pp. 119-126.
2000b
"Sondages, élections et référendums au Québec, 1970-1998. Les effets de l'ambivalence." Paper presented at a joint meeting of the Canadian Political Science Association and La société québécoise de science politique, Québec, August 1st.
2003a
"Ambivalence politique et comportement électoral au Québec en 1998." In: Jean Crête. La science politique au Québec: le dernier des maîtres fondateurs. Québec: Les Presses de l'Université Laval, pp. 271-308.
2003b
"Anatomie d'une campagne: comment les libéraux ont repris le pouvoir ý Québec." Policy Options 24: August, pp. 75-79.
2004
"The Election Night Surprise." Policy Options 25: September, pp. 82-85.
Pinard, Maurice and Richard Hamilton
1976
"Quelques hypothèses sur le comportement des indécis." Le Devoir, November 10: 2.
1980
"Quelques hypothèses sur le comportement des 'discrets'." Le Devoir, May 16: 7.
1984
"Les Québécois votent NON:le sens et la portée du vote." In: Jean Crête. Comportement électoral au Québec. Chicoutimi: GaÎtan Morin éditeur, pp. 335-385.
Presser, Stanley and Michael W. Traugott
1992
"Little White Lies and Social Science Models: Correlated Response Errors in a Panel Study of Voting." Public Opinion Quarterly 56: 77-86.
Schuman, Howard, and Michael P. Johnson
1976
"Attitudes and Behaviour." Annual Review of Sociology 2: 161-207.

314 Canadian Journal of Sociology

Smelser, Neil J.
1998
"The Rational and the Ambivalent in the Social Sciences." American Sociological Review 63: 1-15.
Therrien, Gilles
1999
"Constats sur les dernières élections au Québec et les sondages." Presentation at the Ottawa Professional Marketing Research Society, January 28.
Thompson, Megan M., Mark P. Zanna and Dale W. Griffin
1995
"Let's Not Be Indifferent About (Attitudinal) Ambivalence." In: Richard E.Petty and Jon A. Krosnick. Attitude Strength: Antecedents and Consequences. Mahwah, NJ:Lawrence Erlbaum Associates, pp. 361-386.
Traugott, Michael W.
1987
"The Importance of Persistence in Respondent Selection for Preelection Surveys." Public Opinion Quarterly 51: 48-57.
Traugott, Michael W. and Vincent Price
1992
"Exit Polls in the 1989 Virginia Gubernatorial Race: Where Did They Go Wrong?" Public Opinion Quarterly 56: 245-253.
Zaller, John R.
1992
The Nature and Origins of Mass Opinions. New York: Cambridge University Press.

Political Ambivalence towards the Parti Québécois and its Electoral Consequences, 1970-2003

Maurice Pinard

Canadian Journal of Sociology/Cahiers canadiens de sociologie 30(3) 2005: 281-314