Who is winning the June 2 provincial election?
With the Ontario campaign officially starting next Wednesday, the Star is launching its election predictor, The Signal, to try to answer that question.
Developed with Vox Pop Labs, an independent research organization, The Signal is an online tool that gathers polling information from across Ontario and aggregates the results.
It then extrapolates that data to determine how many seats in the 124-member legislature each party would win if the election were held that day.
As of Wednesday, The Signal forecasts Doug Ford’s Progressive Conservatives, with 37.7 per cent, would win another majority with 70 seats while Steven Del Duca’s Liberals, at 28.4 per cent, and Andrea Horwath’s New Democrats, at 22.1 per cent, would each win 26 seats. Mike Schreiner’s Greens, at 5.4 per cent, would win two seats.
“We will run an update daily so long as there is a poll that’s been released within that 24-hour cycle,” said Clifton van der Linden, a McMaster University political science professor and the CEO of Vox Pop Labs.
The Signal uses what’s called a Bayesian model — similar to U.S. aggregators like FiveThirtyEight — that estimates the “bias” in a polling firm’s past results, then adjusts future forecasts to account for it.
“We’re the only ones in Canada to use the Bayesian model, which is very complex. We run hundreds of thousands of simulations on a super computer. It is a demanding model,” van der Linden said Wednesday.
“We crunch the numbers overnight. There are so many calculations, it takes hours.”
Unlike some poll aggregators that may treat all public-opinion research surveys the same, The Signal examines a polling firm’s historic “house effect” that can influence its results.
“There are different processes and procedures that pollsters use … and of course, there is always a margin of error,” he said.
“The house effect or house bias may be based on the sampling method.”
Some firms might only include political parties’ names when they survey respondents, which can skew results, while other researchers include leaders’ names in their questions.
With federal and provincial parties sharing similar brand names and colours, lines can be blurred during an Ontario election campaign.
“We’re talking about a statistical bias, not a normative bias,” stressed van der Linden, noting pollsters are striving to best reflect public opinion when they do their snapshot-in-time surveys.
“That bias, or framing effect, should get picked up and controlled in the (computer) modelling.”
Drawing upon hundreds of publicly available polls dating back to 2009 — and including those from the 2011, 2014 and 2018 provincial elections — The Signal accounts for such trends and adjusts for them to generate its poll of polls.
Because of that, van der Linden said his team can “produce more reliable forecasts.”
Indeed, in the September 2021 federal election, The Signal’s prediction was closer to the final results than any other polling aggregator in Canada.
JOIN THE CONVERSATION