Those concerned about the influence of polls on voting behaviour have lots of evidence for their case in last week's election results. But they may show the opposite effect on strategic voting than is conventionally believed.
There has been an ongoing debate about whether polls influence citizens' attitudes and votes, or merely reflect their attitudes and voting intentions at a point in time. The knock on the pollsters is that their snapshots have one of two possible effects: They either encourage voters to get on the winning side, as telegraphed by the polls, or they encourage strategic voting to prevent an outcome that is being indicated by the polls.
The best recent example of the latter is what happened in the final days of the 2004 federal campaign in Ontario, where NDP supporters - concerned about a potential victory for Stephen Harper's Conservatives - moved substantially to the Liberals. The same shift happened again in 2006 to a lesser extent, holding Mr. Harper to a minority.
To the extent that bandwagon effect or strategic voting are deemed bad, there is a concomitant increase in the public's resentment toward pollsters. These critics argue they shift attention away from policy and leadership issues toward the "horse race."
In 2008, this resentment was particularly evident among Liberal supporters and especially Liberal campaign organizers, who felt the proliferation of polls (we had three conventional daily tracking polls and another daily survey which concentrated on potential swing ridings) smothered their attempts to focus on issues, especially the environment.
In Ontario this time, the numbers would suggest the polls had an influence on strategic voting - but to reduce its prevalence, not increase it. This contrarian switch may also explain why the main tracking polls captured the national vote intentions within their margins, but offered a misleading picture in Ontario.
Both Nanos Research and Harris/Decima Research had the Liberals ahead or statistically tied with the Conservatives in Ontario right up until the final weekend. Nanos had the gap 12 points in the Liberals' favour at one point, while Harris/Decima had it at eight points. By a week before voting, Ekos Research had telegraphed the Conservatives pulling ahead in Ontario, but then said its numbers indicated a bit of a comeback for the Liberals at the end.
Conservatives actually got 39.2 per cent of the Ontario vote, while the Liberals dropped to 33.8 per cent.
What went wrong? "Many Canadians are last minute political shoppers," according to Nik Nanos, especially since traditional party loyalties have declined.
Mr. Nanos says two things happened over Thanksgiving. The surveys indicated Ontario voters were more concerned about the economy than voters in other areas - so in Ontario, the economy became the ballot question. "Whether or not people liked Stephen Harper, he was competitively a better choice to manage the economy than Stephane Dion."
The second development was more tied to polling. Because the polls detected the Conservatives' precipitous decline in Quebec, it became conventional wisdom that Mr. Harper could not win a majority; hence a decline in the widely held apprehension about handing the Conservatives full control.
While a few previously declared Liberal supporters moved to Mr. Harper based on the handling of the economy triage, Mr. Nanos suggests that many Liberals just didn't vote. Nationally, the Liberals received 850,000 fewer votes than in 2006 - mostly because of the stay-at-home factor, since we know they didn't vote for other parties.
Nor was there much last-minute shifting from the NDP, unlike in 2004 and 2006. As for the Greens, earlier in the campaign polls were showing them running between 12 and 15 per cent in Ontario. They eventually got 8 per cent, but it is not clear from the numbers where those voters who had parked with them earlier went.
The perils of seat projections
There are good reasons why pollsters have shied away from running their percentage of popular votes through a seat projection model to predict seat totals on election day.
EKOS did offer seat projections on Oct. 9 and Oct. 13, which showed how minor changes in popular vote can skew the predictions.
On Oct. 9, the EKOS model predicted the Conservatives would get 152 seats based on 36 per cent of the popular vote. By Oct. 13, EKOS tracking had lowered the Conservatives support by 1.2 percentage points to 34.8 per cent. In the model, that brought them down to only 136 seats (Liberal seats ranged from 60 to 84 in the projections). But as we know, the final total was 143 for the Conservatives and 76 for the Liberals. Therein, a lesson in caution.
Hugh Winsor is an associate at the Canadian Opinion Research Archive at Queen's University, and a former columnist with The Globe and Mail.
Special to The Globe and Mail
