Skip to main content

Liberal Leader Justin Trudeau is accompanied by his mother Margaret Trudeau (L) and his wife Sophie Gregoire, daughter Ella Grace and sons Hadrien (foreground) and Xavier (R) as he watches results at his election night headquarters in Montreal, October 19, 2015.CHRIS WATTIE

This time, the pollsters got the big picture right, although they missed some of the last-minute surge to the Liberals.

Although firms have done reasonably well measuring levels of party support and predicting the correct winner in most elections, the memory of two terrible failures – Alberta in 2012 and British Columbia in 2013 – still haunts the industry.

Once the votes were counted, the results of the 2015 election matched up well with the polls that preceded it in the past few days, although it ultimately underestimated Liberal support. Surveys released in the final weekend before the election reported a likely Liberal government, with some of the bolder pollsters even seeing signs of a majority. Polling firms surveying voters on a daily basis reported a jump in Liberal support in the final few days. This turned out to be what happened when the ballots were counted.

Full coverage of Federal Election 2015

The national average of the final surveys released by companies in the past few days foresaw the Liberals winning 38 per cent, the Conservatives 31 per cent and the NDP finishing third with 21 per cent. While the national numbers are still being finalized, it appears as if the real Liberal vote share was higher than their polling average by a few percentage points. This could have been the result of under-polling Liberals, but it is more likely representative of a last-day shift from the NDP.

There was some thought that the Conservatives might outperform their polling numbers, due to an anti-incumbent polling bias that has shown up in Canadian polling since at least 2004. This was not a significant factor this time around.

Polling methods did not appear to affect the parties' levels of surveyed support. Polls done online, by telephone and using interactive voice response (IVR), which allows respondents to reply to a recorded voice by entering numbers into a telephone, reported very similar figures to one another.

Two firms (Ekos using IVR, and the Angus Reid Institute, which polls online) had the Liberal vote share lower than others, but it is hard to say whether this was due to the method (Forum also used IVR, and had the Liberals quite high), or simply bad luck.

A real wild card going into election night was the province of Quebec. Pollsters could not seem to agree what was happening there. Forum had the Liberals with a 14-percentage-point lead over the Bloc Québécois, Nanos had a tight race between the Liberals and New Democrats, and Ekos foresaw a four-way race.

Ultimately, it seems like the Liberals came out on top by a decent margin in a race with four strong parties, which usually leads to a large seat share for the winner.

The Globe's election forecast, which used an aggregation of all publicly released polls during the campaign, performed well, giving the Liberals an 81-per-cent chance of winning the most seats in the last forecast released before voting day. It was also useful in identifying the rapidly increasing chance of a Liberal majority in the past few days, jumping from close to a 0-per-cent chance a few weeks before the vote to 18 per cent by Sunday.

The model was cautious because of some memorable and recent polling failures. While polling averages are often incorrect by a handful of percentage points, in recent Alberta and British Columbia provincial elections they were off by more than 10 in those cases.

This did not happen on Monday night, and the Liberals won the most seats, in line with the polls and the Globe election forecast's most likely outcome. A surging Liberal vote in the past few days cemented the Liberal's majority.

Paul Fairie is a political scientist in Calgary.

Editor's note: An earlier version of this story incorrectly said the Angus Reid Institute polled using interactive voice response (IVR). In fact, the firm polls online. This version has been corrected.