Ahead of the Canadian federal election (held on April 28, 2025), YouGov produced the first publicly available projection of a Canadian election using a statistical technique called MRP, short for multi-level regression with post-stratification. Overall, we are pleased with how the model performed on the three key measures that any such election model is judged:
- Calling the correct winner of the election
- Each party’s estimated seat total falling within the expressed bounds
- The final overall result — control of government — also lying within the reported potential outcomes
This Canadian election cycle was remarkable. As recently as the start of the year, the opposition Conservative Party seemed on course to dislodge the incumbent Liberals in a potential landslide.
However, on Election Day newly minted prime minister Mark Carney was returned to Canada’s highest elected office and his Liberal Party saw a significant increase in both its vote share and the number of seats it won compared to the previous election.
Even as there was enormous swing back toward the incumbent government in a short space of time, the Conservative opposition also registered an impressive result, securing more than 40% of the popular vote for the first time since 1988. In fact, the 2025 election was first since 1930 in which both of Canada’s main parties each took more than 40% of the national popular vote.
Our MRP told the story of the conclusion of this dynamic race. Overall, for each party projected, the YouGov model had an average vote share error of 1.8 percentage points — less than the standard margin of error surrounding vote-share estimates of around 3% — and an average seat projection error of 4.9.
However, while the central projection of our final MRP was a Liberal majority, the Liberals fell short of the 172 seats required for this, winning 168 seats and challenging for a 169th. This error in the central prediction is something that counts against the success of the model. It is clear that the model overestimated the extent of the Liberal win — not in terms of vote share, but in terms of the party’s ability to turn its narrow popular vote lead into more than 171 constituency wins.
The final election result was even closer to a Liberal majority than that three-or-four-seat gap; the Liberals lost in 10 tossup seats by a hair’s breadth. A few hundred votes across just a handful of seats were, ultimately, the difference between Carney returning with a governing majority and instead having to rely on the support of individual MPs or parties to pass legislation.
Our MRP model — as with any projection — came with a range of intervals within which we believed there was a good probability that the election result would land. While a Liberal majority was our central forecast, the intervals included a hung parliament, as we reported when we published the final model update on April 25.
Each party's seat projection also had lower and upper bounds, and we can be pleased that the final seat total for each party was inside those published estimates.
In order to improve our model for future Canadian elections, we will be spending time over the coming weeks to understand exactly why our central projection overestimated the Liberals. It is likely that the nature of this election and the closeness of so many races will have been factors — as they are in any model’s ability to call races with tight outcomes.
We will also be looking to see what impact any late swing in voter preferences had on our estimates. Looking at the figures over the final days from what was ultimately a strong performance from the Canadian polling industry, there certainly appears to have been some last-minute closing of the gap between the Liberals and the Conservatives, which is likely to have affected the Liberals’ ability to get over the line in the marginal contests they needed to build a majority.
Image: Getty (Andrej Ivanov / Stringer)
What do you really think about politics and everything else? Share your reality, join the YouGov panel, and get paid to share your thoughts. Sign up here.