Much ink was spilt over polling in the Ontario campaign. Though the polls generally did well in this election, in the end the same problems that emerged in the federal campaign were repeated. But whereas the federal results turned an expected Conservative minority into a majority, the provincial results turned a Liberal majority into a minority.
According to the unofficial results of the election last night, the Liberals won 37.6 per cent of the vote, the Progressive Conservatives 35.4 per cent, the New Democrats 22.7 per cent, and the Greens 2.9 per cent.
The gap between the Liberals and the Tories stood at only 2.2 per cent of the vote, far less than the 4.8 point gap that the average of the campaign's final polls expected – and certainly less than the six to 10-point gaps that some polls suggested.
There were six firms that reported in the final days of the campaign. Two used the traditional telephone survey, two used the interactive voice response method, and two used online panels. Simply averaging these final six polls puts Liberal support at 37.9 per cent, PC support at 33.1 per cent, NDP support at 24.5 per cent, and Green support at 3.5 per cent.
The polls did a good job estimating the vote share that Dalton McGuinty's party would receive, with almost all of them indicating that Liberal support stood at either 37 or 38 per cent. The polls also did a decent job of identifying Green supporters, with the difference from the result being only 0.6 points.
But the New Democrats were over-estimated across the board. No single final poll put NDP support lower than what they received, and the average was 1.8 points higher than what the New Democrats actually earned last night.
The real culprit, however, was the under-estimation of Progressive Conservative support. As in the federal campaign, the Tories were under-estimated by over two points (2.3 points last night, 2.2 points on May 2). This turned a slim Liberal majority into a strong Liberal minority, as the Tories hung in closer with the incumbent than expected. Only one firm put Tory support higher than reality, and only by 0.6 points.
No one methodology stood out above the others. The top three firms all used different methods.
Ironically, in light of the controversy stirred up in the opening days of the campaign, the two most accurate polls came from the newest firms: Forum Research, with a total error of only 1.6 points and using the IVR method, and Abacus Data, with a total error of 4.4 points using an online panel. Nanos Research, The Globe and Mail's official pollster, had a total variance of 6.2 points across the four parties on their last three day sample, but that dropped to 5.5 points on their final two day sample. Nanos does its polling using the traditional telephone method, and as in the federal campaign was among the top performers.
The final results were within the reported margins of error of all three of these firms. From a statistical perspective, Forum, Abacus, and Nanos accurately predicted the outcome.
Angus-Reid (6.4 points total error), an online pollster, EKOS Research, using an IVR method, and telephone pollster Ipsos-Reid (10.2 points total error) placed in the bottom three, with all of them having one or more of their results outside of the margin of error. Interestingly, however, EKOS Research's "expected" outcome based on the most likely voters dropped their total error from 7.6 points to 6.3, with the firm upping PC support among likely voters, while Angus-Reid's results were within the margin of error on their sample of "absolutely certain" voters. Perhaps, considering the abysmal turnout last night, the likelihood of decided voters actually turning out to vote is something Canadian pollsters should report in future campaigns.
Generally speaking, the polls performed well Thursday night. They accurately estimated Liberal support and those reporting on the very last days of the campaign indicated that the Tories were closing the gap in the final hours. Some of the best performers from the federal campaign did less well this time, and some of those that struggled in May nailed it last night. All of this points to the role that probability plays in good public opinion polling – sooner or later that 19 times out of 20 catches up to you.
Éric Grenier writes about politics and polls at ThreeHundredEight.com