Over more than four decades dating back to 1962, it has been a given that parties leading in public opinion at the outset of a federal election campaign, have declined in support by voting day. This time frame happens to coincide with a period when polls evolved beyond a novelty, into an important feature of Canadian election campaigns. Table 1 illustrates that during the fourteen federal elections inclusive of 2004, in only one instance did the party leading in popular support prior to the election call actually improve its standing on election day. That exceptional case was in 1974, a campaign focused upon the Conservative plan to combat inflation with wage and price controls, which led to Pierre Trudeau’s memorable put-down of Robert Stanfield’s proposal “Zap, you’re frozen!” There was one other election where the party leading at the outset finished with the same support level, the Mulroney Conservatives in l988. However, that contest, fought on free trade, was characterized by an early decline in Tory support, only to see the Conservatives bounce back at the end.

As shown in the table, 2004 follows the general pattern, but as any observer of the election knows, these data points mask a great deal. Between the Compas poll of May 15-19 and the June 28 election, the Liberal vote did not merely decline two percentage points, but was subject to a roller coaster ride. In fact public opinion polls showed a steady drop in Liberal support, both nationally and in the province of Ontario subsequent to the controversial Ontario provincial budget of May 18, which established health care premiums.

In attempting to monitor public opinion during the 2004 campaign period, it should be noted that there were over thirty national polls publicly released to the media by established companies, which provided regional divisions. Twenty of these were conducted following May 23 when the election writs were dropped. In addition there were a number of other regional samples collected during this period that were usable in assessing the state of the public mind in various parts of the country. This does not however include the daily rolling average collected by SES, except on those occasions where it provided internal regional splits.

The wish to organize and analyze this multiplicity of polls has led to the summary presentation in table 2 of a series of seat projections based upon the “regional swing” model outlined in the 1990 Canadian Political Science Association conference paper “Improving Upon the Cube Law.” These figures are based upon a weighted blend of polls conducted by Ipsos-Reid, Compas, Ekos, Leger, Environics and SES over fourteen-day periods of inclusion, except the final pre-election sampling compiled on June 27, which was drawn on a seven-day interval of polls. In addition a final projection is listed based upon the actual regional voting splits revealed on election night, or in other words a post-diction. This final projection, adjacent to the actual results in the top row, differs somewhat but shows how the same model performed with the regional voting percentages recorded in the election.

The estimate of 130 Liberal seats is based upon the election night vote total for each region (including a 44 to 31 percent Liberal advantage over the Conservatives in Ontario), while the 108 seat pre-election estimate was based upon the previous week’s polls (including a 39 to 34 percent Liberal edge in Ontario). The polling figures for Ontario are highlighted in this table not just because it was the largest region, but because it had a disproportionate number of competitive swing constituencies, and is particularly sensitive to the effects of regional swing. For example, the difference between the June 27 projection and the election night projection meant 17 seats for the Liberals in Ontario alone.

A closer examination of the Ontario party support figures reflects a linear trend of Conservative growth and Liberal decline through the week ending June 18. By that point in the campaign, the Conservatives had reached a narrow plurality among Ontario voters as reflected in the polls, and Stephen Harper made his first overt error in suggesting a Conservative majority was within reach. Until that time, Canadian swing voters seemed blithely unaware that punishing the Liberals because of scandal and arrogance might lead to a government with a very different ideological agenda. Moreover, within two days, Harper’s office was trying to exploit a brutal murder in the Toronto press by issuing a statement questioning if Paul Martin favoured child pornography.

Apart from being a laughable allegation, this showed that the Conservatives could run a negative lowball campaign themselves, and they were no more principled than the Liberals whose attack ads had apparently achieved little traction until that time. Harper had finally fallen off the pedestal of a mistake-free campaign, and the next poll by Ipsos-Reid published in the June 22 Globe and Mail showed that the month-long Conservative rise had been reversed both in Ontario and nationally.

Until this juncture, there had been little evidence that Paul Martin’s campaign was resonating with any new voters. To mix a metaphor, there had been a series of “backbench bimbo eruptions” from sitting Conservative MPs that should have known better than to allow their personal agendas to distract the media from their leader’s carefully controlled message of mainstream moderation. Comments by Rob Merrifield on medicare, Scott Reid on bilingualism, Cheryl Gallant on abortion and later Randy White on the Charter of Rights now started to raise questions about what hidden agendas lay behind the pleasant and, until this time, unthreatening demeanour of party support figures reflects a linear trend of Conservative growth and Liberal decline through the week ending June 18. By that point in the campaign, the Conservatives had reached distract the media from their leader’s carefully controlled message of mainstream moderation. Comments by Rob Merrifield on medicare, Scott Reid on bilingualism, Cheryl Gallant on abortion and later Randy White on the Charter of Rights now started to raise questions about what hidden agendas lay behind the pleasant and, until this time, unthreatening demeanour of Stephen Harper. It could certainly be argued that the Liberal caucus had its own nest of social conservatives, but this was not a problem they had to deal with. After more than a decade in power, nobody could be concerned that Liberals such as Dan McTeague or Tom Wappel had the clout to impose their personal views. Perhaps more serious were Alberta Premier Ralph Klein’s veiled comments about medicare, at around the same time. Whichever of these self-inflicted wounds from the Conservative troops it was that became the straw that broke the leader’s back, it is doubtful that the old Progressive Conservative Party would have been subject to such doubts.

The last week of the election campaign seems to have represented a sea-change in public opinion. Momentum now clearly shifted to the Liberals, particularly Ontario. When one reviews each of the seven polls from six different companies published during that week as presented in table 3, what seems most noteworthy is their remarkable consistency, particularly regarding national totals. Individually, the national differences varied from a tie in the Environics poll to a 6 percent Liberal lead in the first of two Ipsos-Reid polls, but four others showed a 1 percent Liberal lead, including the largest of the samples conducted by Ekos. When a weighted average of all the polls was aggregated on the basis of a combined grand sample of over 15,000, the Liberal plurality over the Conservatives also rounded to 1 percent nationally and 5 percent in Ontario. In Quebec, the Bloc Québécois lead averaged 22 percent over the Liberals.

The difference between these figures and the actual election results was attributed by some eager skeptics to the polls being wrong, but the consistency of data provided by six separate polling companies seems to belie that. Such an interpretation also seems to be challenged by findings provided by two companies who remained in the field following the last round of pre-election poll releases provided in table 3. Ipsos-Reid reported a spike for the Liberals on the evening of Sunday, June 27, of more than 10 percent over the previous days, occurring disproportionately in Ontario and Quebec. This observation was based upon a relatively small sample, but it did portend the findings of an election day poll undertaken by Compas for Global television.

The election day survey was drawn from a sample of 1,200, and one of its most striking features was the 21.8 percent reporting that they made their voting decision on election day. These data are presented in table 4, which also suggests that over 40 percent of the sample had decided on their choice only during the last week. It was further reported on the Global election telecast that of those deciding during the final week of the campaign, 45 percent voted Liberal. It can be deduced from the statistic that among those whose voting decision preceded the final week, fewer than 30 percent voted Liberal. Within this remarkable degree of late choice for the Liberals seems to lie the answer to why the polls reported in table 3 understated the Liberal vote. Much of the change occurred after they left the field. What is less clear is how many of these late decision-makers were changing their preference, and how many were previously undecided.

Traditionally public opinion analysts have tended to dismiss the likelihood of late change attributable to undecided respondents. This was because the undecided seem to vote in smaller proportions, and typically didn’t appear to move collectively in any one direction, but rather to offset each other’s impact. At this point, there isn’t sufficient evidence to conclude that the change noted over the last weekend of the campaign can be explained by this group, but change there clearly was. Another finding of the Compas election day poll was substantial party switching among voters from the 2000 election to the 2004 contest. Among other observations from that source was that over one-third of 2000 Liberals and PCs defected from their corresponding party in 2004. For other parties the switching was of a lesser order, but nonetheless significant. In short, the 2004 election featured substantial partisan change, but in multiple directions.

Another interesting finding was the negative impact that the election campaign had upon the esteem in which the various leaders were held. During the process, Paul Martin, Stephen Harper and Jack Layton all declined in the number of voters who thought they were most trustworthy. Gilles Duceppe’s figures held steady, while the number who didn’t know or refused to answer jumped precipiously.

While among Liberal and NDP voters health care was easily the most important issue, for Conservative and BQ supporters it was Adscam and Liberal corruption that was the most frequently stated campaign issue.

Apart from the question of voter change both short-term and long-term, was the prominence of urbanization in the election results. This of course has long been prominent in the Canadian west, where only in urban areas is the Liberal Party competitive. However, the Ontario results were particularly intriguing in this regard. Rural Ontario is normally less Liberal than the cities, but there was an ever-increasing skewness in the regional patterns of the province in this election. The province-wide difference between the Liberal and Conservative vote in this election was similar to the margin in 2000 (when one combines the Alliance and PCs) but the regional trends vary dramatically. In most urban areas, particularly greater Toronto, the Liberals trended upward compared to 2000, but in most rural constituencies the opposite was true.

One of the less-noted features of the voting result was the deleterious impact of the Green Party upon the NDP. The Greens, seemingly elated with having reached the exalted heights of 4 percent of the national popular vote, don’t seem to have fully contemplated the damage they have wrought. Their best constituency performance was in Saanich-Gulf Islands where the candidate placed fourth with 17 percent support, less than half that of the winning Conservative. To assess the tradeoffs involved in this accomplishment, one should also note that the Green Party represented the margin of difference in sixteen constituencies lost by the New Democratic Party, nine in British Columbia alone.

It is undoubtedly an exaggeration to think that the Green vote could be automatically transferred to the NDP. However, even if only half of the Green support had been directed to the NDP it could have taken eight of these ridings, including Trinity-Spadina and Oshawa. Given the exquisite cliff-hanging result of election night, where only one more seat would have provided the New Democrats the parliamentary balance of power, the impact upon the political left caused by the Green Party’s “excellent adventure” cannot be overestimated.

The parallel to Ralph Nader’s influence on the Florida result during the 2000 US presidential election is difficult to miss. In the closest constituency, New Westminster-Coquitlam, 114 votes, less than 5 percent of the Green total would have furnished that extra NDP seat that would have greatly enhanced its parliamentary leverage. The first response typically made by Green Party apologists to charges that it merely divides the political left, is that if Canada had a fairer proportional representation electoral system, the party’s negative effect would not matter. While true, such an observation ignores the more relevant irony that the Greens, by harming the NDP’s prospect to influence Parliament, also set back the NDP’s primary agenda item of electoral reform, hurting the Green Party itself. Paradoxically, the Green Party is the chief victim of its own “success,” if 4 percent of the popular vote is to be defined as successful. By not electing any members, the party will have difficulty distinguishing itself from the Christian Heritage Party, the Marxist-Leninist and the Marijuana Party in pressing its claim for debate participation in the next election.

One of the more exotic election analyses was performed by the mentalist Kreskin on CTV’s “Canada AM,” during the first week of the campaign when he transcribed seat predictions in code that were only revealed to the public after the election was over. The disclosed numbers showed a remarkable prognosticative capability for someone ostensibly unfamiliar with Canadian voting patterns, and who had the additional handicap of divining numbers a month before the election occurred. However, the seat projection presented for that same week ending May 28 in table 2 provides numbers just as accurate as practitioners of prestidigitation. Such observations might lead to the mirthful hypothesis in this unusual election year, suggesting that forecasting accuracy increases with distance from the actual election.

More seriously, it is educational for readers of polls to be reminded that they are not predictive of the future, but only a snapshot in time. It is highly unusual for polls conducted as little as four days before the election to be out of sync with the actual result, but it is not unprecedented, and shows that public opinion can be volatile even a day or two before an election.