Quebec’s Incredibly Volatile Election

On Chaos, And The Nature Of It

With the news that Quebec has formalized the boundaries for the next election - including an increase in seats from 125 to 127 - the question of seat projection is on my mind. And, having spent many hours I won’t get back trying to figure out how to do this, I have come to a conclusion - this election is going to be such a shitshow that I wouldn’t be able to put out a forecast I actually have any faith in.

With my Federal model, I believe it is basically as good an effort at predicting seat outcomes and translating polls to seats. It’s not perfect, but I put it out because I think the information in the model is both useful for people to see, and is solid enough to justify the legitimacy that calling something an output of a statistically reliable model conveys. When you call something a model, people take it seriously. And I don’t think this Quebec election is going to be able to be modelled in any way more reliable than a guess.

The forces that make this the case are long and varied: we can start with the Liberals, whose extremely efficient vote in 2022 has left them bereft of many obvious targets, as their vote disproportionately fell away from them in non-Liberal seats. Does their vote rise this time also disproportionately return in those seats, or is the QLP vote rise going to be wasted winning safe seats by more and then getting back to ~25% in the next band of seats where they flirted with 10% last time - a respectable shot but not enough to flip seats.

The PQ’s vote efficiency is also in question. With their vote being horrendously inefficient last time, that gives them a wider range of seats they can win, but their real gains come when their vote gets into the 30s, and when it is more of a head to head with the Liberals. With the rise of the CAQ back into the 20s, there are something like two dozen seats where the gap between the top 3 parties is less than 10%. There will be many seats where the winner gets in with less than 30% of the vote, because all three of the PQ, CAQ, and Liberals are in the 20s, and then the Quebec Conservatives and Quebec Solidaire take between 14-20% of the vote between them. I’m not really looking forward to putting out a forecast where I’m essentially praying that I get a dice roll right across the 20 most important seats.

Seat projection has become a thing people want because as our politics has become more multi-party and our voting system has stayed the same, it’s become a lot harder to plot vote shares that pollsters are releasing onto who will form government without a model. Trudeau winning two comfortable minority governments from popular vote losing positions, Bonnie Crombie beating Marit Stiles by 11% and winning half the seats, and Ford gaining vote share and losing seats in 2025 all makes this link harder. Even in a two party province, Danielle Smith won the popular vote by 8% in 2023 but only won the tipping point seat by less than 5%, meaning that Rachel Notley could have lost the popular vote by up to 3% and still won a seats majority.

Quebecers know the difference between seats and votes well - the Quebec Liberals were 4th in votes and 2nd in seats last time, the Quebec Conservatives broke double digits and didn’t get a single seat, and the PQ were third in votes and got 3 seats. The general existence of modern Quebec’s three party politics, whether in the guise of the ADQ under Dumont or now the CAQ under Legault, has meant the relationship between votes and seats has always been wonky. Certainly Federal politics, where you’ve had a rotating cast of third parties fighting the entrenched BQ/Liberal fights since the 90s (Charest’s PCs, Jack and Tom’s NDP, even the Poilievre Conservatives) has made things harder to parse. But this year has gone from hard to parse to impossible to know anything about.

There are two types of uncertainty when it comes to elections. There’s the uncertainty inherent in there being time between now and Election Day, and therefore that things could happen that would change the public mood, and then there’s the uncertainty of a polling error. That risk is real, yes, but at the end of the day there is usually a fairly narrow band in which election results are going to finish in that’s predictable the night before the election. Sometimes - say, BC 2024, where the polls showed an incredibly tight race (which ended up being true) - that narrow band either side of the polls and the models could see either side win. Most of the time, however, it’s the last three Federal elections, where we knew the Liberals would win and the only question was majority or minority. Had Poilievre (or O’Toole, or Scheer) won those elections, given the polls we had and what the models said on election day, it would have been utterly shocking. Here? No matter what the model says I’m not going have any fucking idea what’ll happen. Hell, even if the polls are correct at a provincial level, there could be 20 seat differences in what every party ends up with.

Could the CAQ win 30 seats on 20% of the vote? Oh yeah, easily, if their vote holds up well and non-separatists pick them over the PLQ in seats the Liberals can’t win. Could they also be in single digits on that same 20%? For sure - they get ~5% in Anglo Montreal, 25% basically everywhere else, and they’re within 10% of winning in 35 seats. The Liberals could win anywhere from 25 to 50 seats on their current mid-to-high 20s polling, again depending on how insanely weird three way splits fall. The PQ could win 80 seats right now or 45. We don’t know, and very little that could happen between now and October will make this clear.

Yes, it’s possible the polls break in such a way that the picture does clarify, but in that case then I also don’t really care about the seat projection. I haven’t posted a Federal model update in weeks, because there’s no point. Carney would win a big majority and the CPC would be in a worse place than they were before Poilievre became leader. Whether the Liberals are at 202 or 212 doesn’t actually matter, because in majority governments polls only really matter insofar as they influence the government’s decisions. That the Liberals will win any election held now easily doesn’t need my model to communicate, nor does the difference between 205 and 220 Liberal seats matter.

But in Quebec these distinctions matter quite a lot. And I can’t credibly claim that what I am seeing and building will bring more signal than noise to the conversations. So I’m out of the model wars in Quebec. I’ll cover the campaign closely, and I’ll do so enthusiastically. But part of doing data science is knowing when what the data is giving you is not a science, but a guess.