With the federal election just four weeks away, political pollsters, Ontario sportsbooks, and predictions market traders are all scrambling to figure out how Canadians plan to vote. What they can all agree on is that the Liberals are currently the favourites to win, but the size of that lead depends on who you ask. According to 338Canada, an independent political poll analysis site, a Liberal win seems near-certain. However, those with financial skin in the game currently see the Liberals’ edge as closer to 2-1 or 3-1.
Any sort of Liberal lead is a massive reversal of fortunes. Pierre Polièvre’s Conservatives had held a significant polling lead since mid-2023 and would have almost certainly won a majority had there been an election in late 2024. However, more recent data reveals that overall sentiment has shifted rapidly toward the Liberals since Justin Trudeau‘s resignation and his subsequent replacement by Mark Carney. Also factoring into things are tensions with the United States since President Donald Trump’s inauguration.
Bonus spoke with Phillipe J. Fournier, the Montreal-based post-secondary physics and astrophysics professor behind 338Canada’s projections, about the difference in potential outcomes. He said that part of the reason is that his model projects what would happen if the election were to take place today. Bettors, on the other hand, have to factor in the uncertainty of what might happen between now and the actual election on April 28.
I do not make predictions; I make projections. And I know it’s semantics, but the difference, of course, is prediction is in the future, and projection is in the now.
Polls and betting markets indicate shifting sands
Until recently, Pierre Poilievre’s Conservative Party of Canada (CPC) dominated polls, prediction markets, and Ontario sportsbooks as the most likely election winner.
According to 338Canada, on Jan. 19, the odds of a CPC majority government were 99%. By March 16, the Cons’ odds plummeted to 13% (neck-and-neck with the Liberals 14%). As of yesterday, 338’s data shows a 0% chance of a Conservative-led majority in Parliament.
Conversely, the Liberal Party of Canada (LPC), previously on track for an embarrassing defeat, now have an 88% chance of forming a majority according to Fournier’s model. Further, 338 gives the LPC a 99% chance of winning the most seats.
Over at Kalshi, the markets gave the CPC a 92% chance of achieving a majority result when they opened in December 2024. By mid-March—amid President Trump‘s tariff war—Kalshi traders saw a Conservative majority, Conservative minority, Liberal majority, or Liberal minority as all about equally likely.
Today’s market share prices suggest more than a 72% chance of Mark Carney winning as Prime Minister, and a 61% chance that his government will hold a majority. Betting lines at Ontario sportsbooks vary slightly from site to site, but bet365 currently offers the Liberals at -200 and the Conservatives at +150, suggesting a belief that the Liberals are around 63% to win.
Notably, while not exact, the Canadian Broadcasting Corporation’s (CBC) projections fall within a similar range and pattern to 338’s findings. In both cases, data from multiple polling sources feed the projection models.
Politicos should expect further changes
Fournier stressed that the main reason that pollsters and bettors may disagree is that he and the CBC are making projections based only on current and historical data. Conversely, sportsbooks and market traders are also trying to predict events that are yet to come.
He said his numbers indicate the odds of each party’s results if Canadians voted that day. So, with projection models, the analysis ignores future uncertainties. Alternatively, prediction markets require people to guess what will happen on election day, so buyers must consider the unknown.
In January, my model had conservative odds of winning over 99%, but that was if the vote was held in January. We know that in the future, events can change things. If I had given the odds of winning for the Liberals at one to 10,000, I think a lot of people would have said, ‘You know what, I’ll throw two bucks.’ That’s why they’re so different.
Meanwhile, he said, there’s still a month to the election.
The numbers have moved so much in the last few weeks that it would be a fool’s game to say they won’t change again.
Demand for Canadian prediction markets remains low
Something else to consider is betting volume.
Right now, robust Canadian political polling data is easy to come by. The more data fed into the model, the more likely the projections accurately reflect the current political climate. With prediction platforms, the more money wagered on a particular market, the more likely it is to reflect popular sentiment.
At Kalshi alone, buyers spent $132 million on predictions for the outcome of the recent US election. In comparison, the prime minister market has a volume of only about $782,000, with the overall election result commanding about $215,000. At the same time, the volumes for individual ridings are tiny, so there’s little more than $1 million riding on our upcoming vote.
Considering this, the low volume would likely make the predictions less reliable than projections based on robust polling data. However, over at Polymarket, the volume of wagers on who will become Prime Minister is significantly higher, at over $28 million. Despite this, the results align with Kalshi’s (Carney with a 65% chance vs Poilievre’s 36%).
While he acknowledged he’s not an expert on prediction markets, Fournier noted that outside actors can sway those outcomes. Meanwhile, polling data over the last decade, particularly at the federal level, “has been really good,” he said.
So, while betting markets are always interesting, Fournier said he tends to place more weight on the polls.
Projection model has almost 90% success rate
So far, Fournier’s method has served him well during the 18 general elections covering 2,039 electoral districts that 338 has modelled.
To date, the equation has correctly identified the winner in 1,821 districts, equating to an 89.3% success rate. Of the 218 remaining districts, 132 (6.5%) received a vote share within the projection’s margin of error. Only 4.2% of his results (equivalent to 86 districts) were entirely off-base.
Even with these respectable results, Fournier said that judging by his social media mentions, people expect perfection. But these models are not supposed to be perfect, he said.
It’s incomplete data and a moving target.
Asked whether he recalls any specific unexpected shakeups, Fournier said some individual riding results surprised him because hyperlocal effects are hard to detect and model.
But overall, he said there have been no huge surprises.
I can’t think of one, really. Not since I began in 2017.
For 338, it all started with a spreadsheet
Regardless, winding up in the business of political projects was something of an accident, Fournier added.
During the 2016 US election, he created a “humble” spreadsheet to monitor polling stats. He soon realized the data and American media commentary were “completely dissonant.”
When Donald Trump won ‘unexpectedly,’ he realized he might be able to help the media make polling and projections more accessible to its audiences. So, the teacher of 20 years started the 338 blog—named for the number of seats in the House of Commons—and it took off.
Ever since the media searched him out, he said.
Lots of media wanted to talk to me, and now I talk to you.
About the Author
