Methodology
Our prediction model relies on the principle of uniform national swing, the idea that support for parties rises and falls across all ridings by the same magnitude. Although the federal election will be decided in each of Canada’s 343 individual ridings, there are few polls and scant other data on individual candidates that can help us predict results at a local level. Meanwhile, plentiful nationwide polls give a good indication of the political mood overall. Based on 24 elections, from 1949 to 2021, our model uses those opinion polls to estimate how individual ridings will swing.
The most important part of our model is the nationwide average of public opinion polls, which becomes more predictive as the election approaches. The model also considers the partisan tilt of each polling firm. Using our dataset of historic polls, we compare how the averages tend to move over time, as well as the likely difference between firms’ final polls and the results on election day. We find that polls have a tendency to move back towards the previous election result, as the parliamentary term progresses. Each time we run our statistical model, we produce 10,001 estimates of plausible election results nationwide. The closer we get to election day, the smaller the range of plausible outcomes becomes.
From these nationwide forecasts, we estimate the likely range of outcomes in each of Canada’s ten provinces, incorporating province-level polls where they are available. Some, such as Canada’s three prairie provinces–Manitoba, Saskatchewan and Alberta–tend to move together. Others, such as Quebec, go their own way. For each of our 10,001 simulations, the sum of predicted votes across the provinces must add up to match our nationwide estimate.
Finally, we predict the vote share for each party in each riding by considering how they tend to swing from election to election. For each of our simulated scenarios, the ridings move in accordance with their province. Local dynamics, which we cannot account for, introduce substantial uncertainty. Using past elections as our guide, we simulate this uncertainty within and across ridings, to produce our final estimate of the election result.
