| Party Name | Seats (Current) | Seats Change | Percentage (Current) | Percentage Change | Majority Probability | Minority Probability |
|---|---|---|---|---|---|---|
| Liberal | 186 | +26 | 42.8% | +10.2% | 72.5% | 18.8% |
| Conservative | 129 | +10 | 40.1% | +6.4% | 1.6% | 7.2% |
| Bloc | 15 | -17 | 5.4% | -2.2% | 0% | 0% |
| New Democrat | 11 | -14 | 8.6% | -9.2% | N/A | N/A |
That’s a really good answer!
I just want to add the other thing about 538 & 338 Canada is that they take vote efficiency into account. A lot of people look at polls, see a percentage and take that as the person/party’s chance of winning. However, given in both our systems that it’s not a straight popularity contest, those percentages need to be given context. In the US, the Democrats vote is inefficient, because they get huge margins in states like New York and California. It doesn’t matter whether the winner of that state gets one more vote than their opponent, or 3,194,482 (the difference in California), most states are winner-take-all. Similarly, Canadian electoral districts are winner-take-all, and the Conservative party runs up their vote in the prairies (often winning 60%+ in a riding). This means their vote is inefficient so even even if the polls are right and they get close to the same number of votes as the Liberals, they won’t get anywhere near as many seats.
Something I’ve wondered is how the poll aggregators apply the vote % to regional breakdowns / seat estimates. Are they using the voting patterns from previous elections, or is there polling data broken down by seat?
So if voting patterns changed significantly without overall vote % changing much, the results could still be very very different, correct?
You can read about 338Canada’s Methodology in their About Page.
In answer to your questions, some polls break down their results by region and very rarely (but sometimes) there is riding specific data. Poll Aggregators have to careful about these as they sometimes involve small sample sizes. For example, I think Philippe was talking about one in the last couple years that broke down their results by province, but PEI only had 42 responses (that’s like 3884 Ontarians per capita, which is a good sample size) which is too small a sample to consider so he just aggregated the Atlantic provinces. Riding specific polls are also rarely useful due to small sample size, infrequent polling, and questions about their methodology.
The secret sauce of these poll aggregators is how they rate polling firms/methods and how they weigh ridings for past performance, shifting demographics, and current candidates. If you’re curious for more, I like listening to Philippe and Éric Grenier talk about polling on The Numbers podcast.