All of our Canadian Provincial and Federal models use a similar methodology in how it translates seat projections from vote shares.
At Mainstreet, we conduct polling in Canada using IVR(integrated voice response) meaning that people are reached through phone call, either to cell or landline, and are played a previously recorded message (recorded by a real person) which prompts them to press buttons corresponding to their vote preferences, demographics, and their answers to various survey questions.
All of the responses in a survey are then weighted to match the demographics of a specific province/city/region, and the final report is generated and shared publicly. The demographics from that poll, specifically vote intention by demographic and by region, are collected and used to model election outcomes for all of our models.
By knowing how an area tends to vote, stronger for one party compared to how it does in other ridings, is the basis for the model. By knowing how it has voted in past elections, we can project how it will vote in an upcoming election given the vote shares, regional and demographic breakouts from our most recent poll.
However, the strength of the model is not just limited to change from the past election. As the model uses the demographic information from the polling itself, it is able to identify areas where a party will overperform or underperform due to strength with various different demographics. For example, while a party might tend to better in the urban areas of a city, if, in new polling, that party is doing much better with higher income voters and voters that have a university degree, then they will make gains, larger than expected if using conventional swing, compared to the last election. Conversely, if a party traditionally does well in rural ridings, but in their polling they are losing ground with older voters and voters without a college or university degree, then the model will account for that and give them a lower vote share in ridings where those voters are more prevalent.
This is different than most other more swing based models, as the Mainstreet Research Canada Models are better able to identify differential swing, where parties will make larger than expected gains or losses in ridings based on their demographic profile. Whenever we update our polling in Canada, we will update our model accordingly.