On election night, few people were more surprised by the final tally than the political experts and pundits. The final five national media polls, con- ducted during the penultimate week of the campaign, and completed at least three days before the vote, were all quite consistent and pointed to a very close outcome. Four of them showed the Liberals leading by only one percent- age point over the new Conservatives. As the first section of Table 1 indicates, these five polls predicted, on the aver- age, that the Liberals would obtain 33 percent of the vote and the Conservatives, 32 percent. Actually the respective figures were 37 percent and 30 percent. A predicted gap of only one percentage point had become one of seven points, enough to have a major impact on the results.

Two pollsters and one academic who, on the basis of these polls, ventured to forecast the distribution of seats in the new Parliament all accurately foresaw a minority gov- ernment. But in two cases they predicted a minority Conservative government, with about 116 seats, followed by the Liberals, with about 105 seats. Only in one case was there a prediction of a minority Liberal government. That forecast called for 117 Liberal seats " only eight more than the Conservatives. In fact the Liberals reaped 135 seats and the Conservatives only 99.

How can such forecasting errors be explained? The answer is astonishingly simple: Voting intentions changed at the very end of the campaign. The recently published results of the Canadian Election Study, conducted by uni- versity professors, indicated that, at least outside Quebec, an important surge in support for the Liberals occurred during the last few days of the campaign, especially during the days after the media’s pollsters had left the field. This coincided with a corresponding decline of both Conservative and NDP support. These results were consis- tent with those of a COMPAS poll carried out the day of the election and released the same night. Despite its rec- ognized shortcomings, the vote results obtained in that poll were within one percentage point of the actual results for each of the parties. More importantly, that poll found that 22 percent declared to have reached their final vote decision only on that day, and another 5 percent the day before. Finally Ipsos-Reid reported later net shifts to the Liberals in Ontario the night before the vote, according to its then on- going omnibus polling.

The results of the Canadian Election Study also revealed that there was a decline in Conservative support outside Quebec and possibly some Liberal gains, even before the closing days, which started around June 10. These results are also consistent with those of the national media polls, in particular those of SES Research.

To explain the late campaign changes, and implicitly the earlier ones, the authors of the Canadian Election Study, after analyzing the results, rejected many popular theories. For example, they concluded that the changes outside Quebec could not be explained by the English leaders debate, by changes in the image of the Conservative Party and of its leader, by late strategic voting among some NDP supporters or by decreasing anger over the sponsorship scandal. They found however that there was a late surge in Liberal support among those who were only somewhat angry over that scandal. But why?

In order to press further for an explanation of the mid-campaign and last-minute changes in voting inten- tions, it is important to identify the provinces or regions in which they occurred, something to which not enough attention has been paid.

In table 1, the average levels of party support in the last five media polls of the campaign are presented and com- pared to the actual results of the vote for six provinces or regions. For this analy- sis, the media polls are particularly use- ful, given that the regional averages computed ultimately rest on a com- bined total of almost 13,000 interviews.

The first thing to be noted is that the closeness of the voting intentions for the two main parties was hiding large differences in every province and region. The regional and linguistic cleavages of our party system continued to prevail, with the polls (and actual results) showing that the Liberals were leading in Ontario and the Atlantic provinces, while the Conservatives were leading in all of Western Canada and the Bloc Québécois, in Quebec.

The most striking finding, however, is the relative accuracy of the averaged results in all the provinces or regions of Western and Atlantic Canada. In British Columbia, Alberta, the Prairies and Atlantic Canada, the final polls estimat- ed on the average the support of all par- ties quite accurately. All differences with the actual results were two percent or less. This is an excellent poll perform- ance by any standard. This suggests that there were no important last-minute net shifts in favour of, or against any one party in these provinces. They were therefore not responsible for the surpris- ing election night results.

Conversely, the significant poll mis- estimations were all located in Ontario and Quebec. In each case, there were on the average serious underestimations of Liberal support, by five and six percent- age points, respectively. In these provinces, all other parties were overesti- mated, although each less than the underestimations of the Liberals. Separately, each of the five final polls consistently underestimated Liberal sup- port in Ontario and four of the five did the same in Quebec.

To explain the last minute surge of Liberal support, it is imperative to con- sider these two provinces separately rather than Canada or Canada outside Quebec as a whole. Notice that these two provinces turned out to be the only ones where the Liberals did less well in 2004 than in 2000. Each had therefore a large pool of former Liberal supporters who may have felt pressures to return to the fold at the last minute.

It is important to examine develop- ments in Ontario and Quebec starting at the beginning of 2004. As the year began, the members of the Progressive Conservative Party and Canadian Alliance had approved a merger of their two parties and the country had a new prime minister, Paul Martin.

Consider Ontario first. As table 2 indicates, at the turn of the year, the Lib- erals enjoyed the very comfortable sup- port of close to 60 percent of the voters in that province. But they soon faced a first setback with the release of the auditor general’s report on the sponsorship scan- dal. As it did throughout Canada, sup- port for the Liberals in Ontario slipped immediately by more than 10 percentage points. It then remained at 46 to 47 per- cent for more than three months. The deserters went about equally to the new Conservatives and to the NDP.

But contrary to what happened elsewhere at the time, a second important slippage in Liberal support occurred following the presentation in mid-May of the Ontario budget by the governing provincial Liberals. The sub- stantial new taxes it contained, con- trary to a campaign promise of no tax increases, greatly irritated the voters of that province. Many decided to vent their anger against the federal Liberals, who again lost more than 10 percent- age points, recuperated mostly by the Conservatives. For a month, this left the Liberals neck and neck with the Conservatives in Ontario, with their support averaging 35 percent and 36 percent, respectively. This early phase of the official campaign turned out to be, for the Liberals, their worst overall.

The data suggest that such irra- tional reactions finally gave way to more sober considerations, and sup- port for the Ontario Liberals started to rise again. It increased by five percent during the following week and, finally, by an additional and unforeseen five percent on voting day. This occurred equally at the expense of the Conservatives and of the NDP.

It is quite important to realize that the Liberals’ recovery simply brought all parties in Ontario back to roughly the same levels of support as those that prevailed for more than three months before the provincial budget. Without the budget contro- versy, Ontarians’ voting intentions might well have remained at the level at which they were before the budget throughout the campaign.

It is worth pointing out that this would have possibly led to relatively accurate vote forecasts in Ontario, as they were in most other provinces. However, it remains to be seen whether the return of some waverers to more rational considerations may have been facilitated by their opposi- tion to the social conservatism of Harper’s party and the blunders on social issues by some of his candidates during the campaign, if these blunders caught their attantion at all. The Liberals did not pull their punches on these issues. All in all, this could account for the fact that some of those who were only somewhat angered by the sponsorship scandal returned to the Liberals, as mentioned before.

As in Ontario, the Liberals started the year in Quebec with a com- fortable lead over their opponents, although their lead in Quebec was less pronounced than in Ontario. Yet close to half the voters supported them on the average, while their main oppo- nent, the Bloc Québécois, was getting the support of barely more than a third of the voters. Indeed many thought that the Bloc was in serious trouble only a few months before the election. But as table 3 indicates, the auditor general’s report also had devastating effects on the Liberals in Quebec. Their plunge of about 15 percentage points was even larger than that suffered in Ontario at the time. Two-thirds of those points went to the Bloc Québé- cois. This produced a complete rever- sal, the Bloc now replacing the Liberals as the leading party in Quebec. Very much like in Ontario, the new division of the vote then remained stable for no less than four months.

As the election campaign stepped up in early June, the Liberals however managed to suddenly lose another six points, possibly as a result of the increased salience of the sponsorship scandal. It left them in a disastrous sit- uation, with the support of barely more than a quarter of the Quebec vote. But this time the lost points went largely to the Conservatives.

Then came the leaders debate in French, which, according to the polls, Gilles Duceppe won hands down. The authors of the Canadian Election Study reported that this was the decisive moment in the Quebec campaign, with the Bloc support then jumping by about 10 percentage points. To be sure, the data of table 3 also show gains by the party after the debate, but the gains, at four points on the average, do not appear to be as pronounced as in the Canadian Election Study. Moreover the gains by the Bloc, according to the data of table 3, were not made solely at the Liberals’ expense. Indeed, as we have seen, the important losses by the Liberals at that time preceded the debate.

From there on, the Bloc Québécois maintained its position for the rest of the polling period. But in the week preceding the vote, Liberal fortunes turned for the better. As in Ontario again, they started then to regain some of their lost support, a process that con- tinued after the last media polls until election day. They ultimately took a third of the vote, leaving the Bloc Québécois just under half the vote. As in Ontario once more, these results brought the parties back to about the same levels of support as those they had obtained from February to March.

To explain the Liberal gains in late June and during the last days of the campaign in Quebec, a different hypoth- esis than that suggested for the Ontario situation seems warranted. To be sure, as has been suggested, the under- representation of non-Francophones in polls in Quebec, which most polls did not correct, may have contributed to the five percent underestimation of Liberal support in the last media polls of the campaign. But if this were the main explanation, one should have observed an equal overestimation of Bloc support, which is not the case. The overestima- tion was only two points.

What presumably best explains the Quebec case is political ambiva- lence, which exists mainly toward the sovereignist option and the Parti Québécois, but also to some extent toward the Bloc Québécois. This ambiva- lence stems from oppos- ing forces that pull voters between attraction to these parties and reserva- tions about Quebec sovereignty. At the very end of election campaigns, this often leads some to back off from supporting the sover- eignists or even minor parties. For example, since 1970, support for the Parti Québécois has been greatly overestimated in polls when the party was very attractive, while sup- port for the provincial Liberals has always been underestimated, whether the Parti Québécois was attractive or not.

While understandably the Bloc Québécois does not arouse as much ret- icence, as it cannot achieve sovereignty, such an ambivalent situation did indeed prevail, for instance, with the major sweep of that party in the 1993 federal election in Quebec. Again last June, in order to contain what appeared to be another major Bloc sweep, some marginal supporters, dissatisfied with the Liberals, but favorable to federal- ism, presumably felt compelled, toward the end of the campaign, to vote Liberal anyway. This occurred not only at the expense of the Bloc, but also of the Conservatives and the NDP, thus rein- forcing a two-party polarization in Quebec and the underestimation of Liberal support in the final days.

An ambivalent electorate is all too often an unpredictable one.