Tuesday, October 28, 2014

Tory, polls, big winners in Toronto's election

Though at one point in the count it looked like the result would be far closer than anyone expected, in the end John Tory prevailed over Doug Ford and Olivia Chow comfortably. The polls performed remarkably well and, on average, estimated each candidate's support to within two percentage points.

Tory took 40.3% of the vote last night, beating out Ford at 33.7% and Chow at 23.2%. The dozens of other candidates combined for 2.8% of the vote.

ThreeHundredEight.com's weighted average was only slightly different, at 42.8% for Tory, 32.2% for Ford, and 22.2% for Chow. Support for other candidates was exactly right at 2.8%, for a total error of five points. That equates to an error of 1.7 points per candidate, or 1.3 points if we include the estimate for 'other' support.

But the average performed well not because it found the middle-road among a group of disparate polling results, but rather because each of the pollsters in the field in the final days came very close to the mark.

The closest was Mainstreet Technologies, which polled four days before the vote and had a massive sample of about 3,320 decided voters. In the first public test of its polling, its overall error was 3.4 points, or 0.9 points per candidate (including others). It was the only poll to nail one of the candidates' support levels exactly, estimating Ford to have the support of 34% of voters. It was also the closest for estimating Chow's support. Tory was over-estimated slightly and Chow was under-estimated a little, but all reported results were within the poll's small margin of error.

Ipsos Reid, which polled the furthest out from Election Day of the three (October 21-23) was the next closest, with a total error of 6.3 points (or 1.6 points per candidate, including others). Ipsos was the closest for Tory, at 41.7%. It under-estimated Ford and over-estimated Chow, but all results were again within the margin of error of a random sample of similar size.

The most active pollster on the municipal scene, Forum Research, was the furthest with a total error of 7.8 points (or two per candidate, including others). Its estimate for Tory was just outside of the margin of error, over-estimating his support by almost four points. Ford and Chow were both under-estimated slightly, but overall it was still a decent result.

It is interesting to note that without Forum's final poll, the averages would have been 41.7% for Tory, 32.4% for Ford, and 23.1% for Chow. That would have given a total error of 2.8 points, better than either the Mainstreet or Ipsos surveys.

But overall, the polls told the story of the campaign in Toronto accurately, and those surveys that had Ford at a relatively high level of support were by no means implausible. In fact, Ford ended up with more of the vote than the polling averages (with the exception of one update on October 6) ever gave him. We can reasonably conclude then that the polls served the electorate well, with little misleading information being published.

Other cities

As far as I am aware, Forum was the only pollster active outside of Toronto with the exception of Mainstreet in Brampton (if I am wrong here, please correct me).

The polling in Mississauga was done closest to the vote out of these non-Toronto polls, and the result was not exactly stellar. The final Forum poll of October 24 surveyed 308 people in Mississauga, estimating Bonnie Crombie's support to be 52% against 34% for Steven Mahoney. The result was actually 64% for Crombie and 29% for Mahoney. An earlier poll by Forum done on October 15 with a larger sample of 769 respondents was closer, with 56% for Crombie to 31% for Mahoney. The winner was right here, but the margin was not.

In Hamilton, Forum did better. The last poll of October 17 (751 surveyed) put Fred Eisenberger ahead with 37% to 25% for Brad Clark and 22% for Brian McHattie. The result was 40% for Eisenberger to 32% for Clark and 20% for McHattie. Not a bad performance, but Clark was well outside the margin of error.

Mainstreet was last in the field in Brampton, reporting a poll of October 20 with 1,602 respondents that showed 41% for Linda Jeffrey, 34% for John Sanderson, and 13% for Susan Fennell. Forum was in the field earlier on October 16 (surveying 1,020), putting Jeffrey ahead with 42% to 27% for Sanderson and 14% for Fennell. The result: 49% for Jeffrey, 22% for Sanderson, and 13% for Fennell. An over-estimation of Sanderson's support to the detriment of Jeffrey by both Mainstreet and Forum.

In London, Forum was out of the field almost three weeks before the vote (Oct 8-10, surveying 782) so it is hard to blame it for the size of its error. But it did have the winner right, giving Matt Brown 35% to 27% for Paul Cheng. In the end, Brown won with 58% of the vote to Cheng's 34%.

The one real upset, if we can call it that, was in St. Catharines, but again that is based on an old poll of October 7 (729 surveyed). The poll gave Jeff Burch the edge among decided voters with 32% to Walter Sendzik's 24% and Peter Secord's 22%. The result was instead a Sendzik victory with 40% to 35% for Burch and 20% for Secord.

Finally, there was the poll in Ottawa. This was done well before the election on September 18, surveying 1,096 people. But the campaign here was very dry and low key, which explains why Ontario's second-largest city went virtually unpolled. The Forum survey gave Jim Watson the support of 63% of decided voters, followed by Mike Maguire at 24%. In the end, Watson won in an even bigger landslide with 76% to Maguire's 19%.

So the polling outside of Toronto (featuring lower profile campaigns with fewer voters, so there is more potential for error) was hit or miss. Of the polls done within two weeks of the vote, the one in Hamilton was a success with those in Mississauga and Brampton being of mixed quality. The older surveys in London, St. Catharines, and Ottawa were only somewhat indicative of what the result would end up being. The winners were identified in London and Ottawa, and that it could be a close race was hinted at in St. Catharines. On that score, the polls outside of Toronto would get only a passing grade, compared to the straight-As the polls receive in Toronto.

This is likely to be the last major election to be held for quite some time. Newfoundland and Labrador will probably not head to the polls until next year, and it remains to be seen what will happen in Alberta. But if Jim Prentice decides to take his strong by-election results as a mini-mandate, there may be no high-profile election held before the federal vote next fall. That leaves Toronto as the 'what have you done for me lately' election for the polls, one in which they were vindicated. I naively hope that will quiet the critics for now.