PDF
Advance Polls and the "Late Switch" in the 2004 Election
Chris Cattle; Barry Kay

Every election produces considerable discussion among pollsters, academics and politicians about the accuracy of pre-election polls.  One analytical tool that has previously received little academic attention is the impact of voting in the advance polls. In 2004 they were held on June 18, 19 and 21 with the general election on June 28.  Does public opinion change dramatically during the campaign’s final days?  This article seeks to address two questions. Are advance poll voters reflective of the broader range of the Canadian electorate at the time they occur?  Did advance poll voters in 2004 perform distinctively enough from the general election voters so as to suggest a systematic change during the final days of the campaign? 

Perhaps nobody was more surprised by the election night results of Canada’s federal vote on June 28, 2004 than the array of pollsters who had been consulted at regular intervals during the preceding campaign. It was not  that the Canadian public had been free of volatility. In fact the election campaign had taken on the proportions of a roller coaster ride for much of its duration. For example, in the key swing province of Ontario, a weighted aggregation of polls indicated the Liberals dropping from a 23% lead over the Conservatives in the period ending May 15, a week before the writs were dropped, to a 2% deficit to the Conservatives five weeks later on June 18, only to have the Liberals rebound to a 13% advantage on election night.¹ 

Rather the reason for the surprise among practitioners of the public opinion industry, was the remarkable consistency that had been demonstrated in the set of polls released during the week prior to election day. Table 1 shows the results of seven separate polls published by six different polling companies during the final week of the campaign. They were based upon interviews conducted from June 17 through June 24. 

The discrepancy between these figures and the actual results, led to a round of articles following the election by eager skeptics in the press about how the pollsters got it wrong.² While pollster error is always a possibility, there are other explanations to be considered for the inconsistency between late polls and the final election results in 2004. This is particularly so after a substantially better record by late polls in other recent federal votes.³ The most obvious alternate hypothesis to pollster error was that public opinion, at least in certain regions of the country, might have been subject to late volatility after the polling for pre-election reports had ended on June 24. 

There were in fact other sources of evidence available from two companies that remained in the field, following the last round of pre-election poll releases. Ipsos-Reid reported a spike for the Liberals on election eve, Sunday, June 27, of approximately 15% over the preceding days in Ontario.⁴ 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.⁵ 

The election day survey was drawn from a sample of 1200, and one of its most striking features was 21.8% reporting that they made their voting decision on election day. These data are presented in Table 2, which also shows that over 40% of the sample had decided on their choice during the last week. It was further reported that of those deciding during the final week of the campaign, 45% voted Liberal.⁶ It can be deduced from the statistic, that among those deciding before the final week, only 31% of the sample voted Liberal. Within this remarkable degree of late choice for the Liberals, would seem to lie an explanation as to why the polls reported in Table 1 understated the Liberal vote.  Much of the change occurred, after they left the field. 

Advance Poll Results 

Another way to approach this issue is to analyze the advance poll results. As a rarely utilized research technique, the discussion of advance polls should be evaluated cautiously. Simply because a group of citizens choose to vote a week or so before the general election, does not necessarily make them an accurate reflection of the state of public opinion at the time. It is entirely possible that there are systematic distinctions in the advance poll electorate, that could bias any attempt to portray its representativeness of the general public at that time. 

We can speculate that some people who fall into this category are those who expect to be away from their normal residence, those who are busy working on election day activities, and those who because of physical challenges might find it more convenient to vote before the anticipated crowds of election day. The question then facing us is whether these people or others who prefer to vote early, are untypical of other Canadians in their partisan preference. 

In fact the 2004 advance polls brought a record turnout of over 1,200,000, a greater than 60% increase over the 2000 election experience of some 750,000 early voters. The timing of the general election at the beginning of the school summer vacation was undoubtedly a factor in this. It is difficult to determine demographic correlates of advance poll participation, but in 2004 Ontario was the region with the largest proportion of early turnout and the Prairies were the lowest, but no region varied by more than 1.5 points from the national average of 9.2%.⁷ Interestingly on an individual constituency basis, the three ridings with the highest advance poll turnout were in the Ottawa area, led by Nepean-Carleton. 

As there is a dearth of research about advance polls themselves, we know little about what motivates Canadians to participate in them, and vote prior to the general election. 

The most straightforward way to evaluate advance polls as a measure of Canadian opinion a week before the general election, is to take account of which party’s candidate won in each of the 308 federal constituencies. This information is provided in Table 3, accompanied by the percent change from the general election for parties in each region of the country. The most noteworthy observation is that the Conservatives would have won a bare plurality of five seats, just as was being mooted in seat projections during the last days of the campaign. This was particularly attributable to Conservative advance poll success in Ontario (17 additional seats) and British Columbia (6 more seats), and reflects the superior Conservative support of  4.5% nationally a week before the election.  Although this effect was evidenced in every province of the country, it was strongest in BC (7.0% greater in the advance poll) and Ontario (where it was 6.2% higher). As one might expect if the Conservatives declined between the advance poll and election day, the Liberals improved but not very dramatically nor uniformly. Nationally the Liberal increase was 0.7% on election day, but that figure included regional performances ranging from gains of  2.6% in BC and 2.3% in Ontario, to a 3.1% decline in Quebec. 

While the data is suggestive of voter volatility during the campaign’s final week, the question lingers as to whether this might be an artifact of something idiosyncratic about the advance poll electorate. Was that 9.2% of the voting population, whether motivated by impending vacations or some other factor, disproportionately inclined to support the Conservative Party?⁸ This matter cannot be established with certainty based upon aggregate data alone, but a comparison with previous elections can begin to address historical patterns to the phenomenon. 

Table 3 also shows the advance poll seat totals and vote change by region for each of the three preceding federal elections dating back to 1993 when the political party configuration took on its present character. In each case, there is incidence of some change between the advance polls and the general election vote, but the one inescapable observation to be drawn is that Quebec Liberals consistently overperform in the advance polls, and then decline in general election support. This pattern prevails throughout the period of study, but is especially marked for the 1997 and 2000 elections. 

One might even be tempted to observe that there is a typical Quebec Liberal overvote in advance polls. It was 3.0% in 1993, 5.6% in 1997, 7.7% in 2000, and 3.1% in 2004, corresponding to greater gains at the advance poll of 9, 23, 19 and 4 seats respectively. The only other consistent pattern of regional change from advance polls to general elections occurs in BC, with Reform/Canadian Alliance and their descendant, the newly united Conservative party. They also draw more votes in advance polls, but not by the same proportions as the Quebec Liberals, and apart from 2004 it makes little difference in seat switches. 

*Prairies region includes Manitoba, Saskatchewan and Territories. 

A further examination of the 1993 through 2004 elections shows substantial seat changes between the advance and general votes, but apart from 2004 this is largely attributable to Quebec.⁹  Excluding Quebec for example, there is never a change exceeding two net seats for the Liberals in any election from l993 through 2000. There is however one other noteworthy switch in seats, also possibly attributable to a late change in public opinion. For those who recall the 1997 election, one of the chief surprises at the time was the unexpected NDP surge in Atlantic Canada, and the advance poll differential suggests that it was also a late development, not fully caught by late pre-election polls. 

What does the data tell us about regular discrepancies between advance polls and general elections? Four elections might not constitute a sufficient sample from which to generalize conclusively, but apart from the pattern of Quebec Liberals voting disproportionately in advance polls, it is dubious to suggest that there is systematic evidence to question the use of advance polls in 2004 as an indicator of public opinion a week before election day. The pattern of data from other sources cited, as well as from advance polls presented here seems to be consistent with the hypothesis that late voter decisions and changes in party preference were responsible for the switch between late polls and the election night results. It should be remembered that the last of the interviews included in the final round of pre-election polls were conducted on June 24, four days before the election. 

There was however, one final set of data examined and presented in Table 4. These figures provided a comparison of the final round of seven pre-election polls summarized in Table 1, categorized by region and compared with the election night and advance poll percentages organized by party. The regional patterns are somewhat mixed and conclusions to be drawn are inconsistent, but both the national numbers and the figures for Ontario show that the late polls weighted and aggregated by region closely parallel the advance poll results, which were held nearer the time most of those sample were interviewed. For example, the national pre-election poll numbers averaged to a 1% lead for the Liberals over the Conservatives, while the advance poll numbers showed a 1.9% Liberal lead, and the election night result was a 7.1% Liberal lead. In Ontario, the pre-election polls showed a 5% Liberal lead over the Conservatives, while the advance poll vote reflected a 4.7% Liberal lead, and the election night result was a 13.2% Liberal margin. The combined sample of these pre-election polls was over 15,000 nationally and approximately 5000 in Ontario, but the smaller regional subsamples were between 1000 and 1500, and the associated sampling error could account for the more uneven pattern elsewhere. The Quebec data on the other hand, confirm the widely observed trend of polls undercounting Liberal and federalist voters.¹⁰ 

Without intending to conclude too much from the mixed evidence in Table 4, to the extent observations can be drawn, they do tend to confirm the previously reported findings. The cumulative pattern is that the polls based upon interviews conducted from June 17 to 24 were generally in line with the advance polls of June 18 to 21, and where this didn’t occur, particularly in Quebec, it could be attributed to the historical trend of depressed Liberal support in polls, and the observations newly identified in this paper of disproportionate Liberal turnout in advance polls. The June 28, 2004 election gives substance to an adage frequently repeated by students of public opinion, polls are not predictive of the future but are only a snapshot in time. It is unusual for polls conducted as little as four days before an election to be at variance with the actual result, but it is not unprecedented and should remind us of what we already know, that public opinion is volatile. 

Notes 

1. Data presented in Barry Kay “Polls, Projections, Pundits and Prestidigitation”, Policy Options, Vol. 25, no. 8 (Sept. 2004) p.71. 

2. For a particularly zealous illustration, see Peter Calamai in the Toronto Star, June 30, 2004. 

3. For example, an analysis of the 2000 pre-election polls for six national polling companies showed that their regional accuracy ranged from 1.8% to 2.8% per party. See also Claude Emery, Public Opinion Polling in Canada, Library of Parliament: Parliamentary Research Branch (1994). 

4. Information gathered from interview with Darrel Bricker, Ipsos-Reid CEO on June 30, 2004. 

5. An additional data source confirming this late trend to the Liberals appeared in “How the Race was Won”, E.Gidengil, A.Blais, J.Everitt, P. Fournier and N. Nevitte, The Globe and Mail, July 14, 2004. 

6. Global Television election broadcast June 28, 2004. 

7. At the provincial level however, New Brunswick consistently had the highest turnout and Newfoundland the lowest for advance polls over the past four elections. 

8. The comparable advance poll turnout figures for previous elections were 6.0% in 2000, 5.4% in 1997 and 4.6% in 1993. 

9. A count of each actual seat change between 1993 and 2000 shows that 53 of 90 (58.9%) occurred in Quebec, while in 2004 only 8 of 40 (20%) took place in that province. 

10. Claire Durand, Andre Blais and Sebastien Vachon, “A Late Campaign Swing or a Failure of  the Polls? The Case of the 1998 Quebec Election”, Public Opinion Quarterly, Vol. 65 (Spring 2001), pp.108-123.