The approach YouGov used to produce its projection of the results of the 2025 Canadian federal election is a statistical technique called MRP, short for multi-level regression with post-stratification. MRP combines survey data with riding-level information from the Canadian Census — such as population density and the proportion of the local population with a university degree – while incorporating data on previous voter behaviour and previous Canadian election results. YouGov’s MRP model is more than just a poll; it utilises leading-edge data science and is based on a substantial sample of more than 20,000 interviews. A sample of this size is essential to capture the full range of views across the Canadian likely voter population and to ensure the model is robust and reliable.
YouGov's approach to MRP modelling is market leading and has a proven track record of success including in recent British, Spanish, and German national elections. Our model uses a large sample of data collected from our panel to discover relationships between people’s characteristics and their answers to the voting intention question. It then combines these relationships/patterns with information about the characteristics of people living in different electorates to construct estimates of how voting intention would look in each riding. MRP models are extremely powerful and computationally complex and are capable of simulating and estimating thousands upon thousands of different configurations and patterns of voter behaviour and party support. For the 2025 Canadian federal election, each YouGov MRP model will explore over 25,000 different possibilities.
The sample size involved – more than 20,000 interviews — means we have collected the opinions of many different voters from many different places across all ridings in Canada. Working with such large sample data requires successive rounds of trimming down the overall pool of interview data to create the best and most robust sample possible which we then feed into the MRP model. The effective sample size after sample trimming each model will be around 6,000.
The modelling process includes many socio-demographic characteristics of individuals and the electorates they live in, past vote choices of individuals, share of votes for various parties within each riding, and information about patterns of candidacy and party contests across all of Canada’s federal electoral districts.
While we cannot reliably capture every local factor specific to every single seat, our large model sample sizes will help the Canadian public understand the attitudes and opinions of different types of voters in different places across Canada.
Both the seat winner and voting intention estimates have some uncertainty, as does any measurement using survey data and statistical inference. For each party in each seat, as well as the seat total for each party overall, we report credible intervals, which represent the range of outcomes we believe the true party shares (and seat totals) have a 90% chance of lying within, if the election were being held right now. The specific number presented for each party is the most likely value, which is typically very close to the middle of that range. So, when assessing the data it is best to remember that both the seat and voting estimates for all parties projected could very well be higher, or lower, than this midpoint.
Our central projected national outcomes (total seats and vote shares) are also midpoints of all the various possibilities we have explored, and are therefore themselves also subject to error and uncertainty. As with the riding-level projections, it is important to note that our central calls for the number of seats and percentage of the national vote won by each party could very well be higher, or lower, than those midpoints.
Image: Getty (Wong Yu Liang)
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.
