The State of the 2025 Election Based on Aggregate Poll Modelling

It’s a common expression to say that “the only poll that matters is election day,” regarding opinion polls for a given election cycle. While this is absolutely true, it is also true that we can look to aggregate polling for a good idea on how an election can go – especially in Canada, where opinion polling has frequently proven incredibly accurate. In the 2021 Federal Election, 338Canada correctly predicted the winner in 92% of seats; in 2019, this number was 88%.

With this in mind, we can look to what aggregate polling (particularly 338Canada) is saying about the 2025 election in order to have a pretty good idea of what’s going to go down on election day – or at least where polling would need to go wrong for us to see a result different from what is expected. And the writing on the wall is not ideal for the Conservative Party of Canada.

As of the writing of this article, 5 days before election day, 338Canada projects that the Liberals will win a majority of 188 seats, with the Conservatives coming in a distant second at 123 seats. Yes, polling could be off here, but it would need to be off by a magnitude we’ve never really seen in order for Conservatives to win this election. To illustrate this, I’ll use the benchmark figure of 90% seat accuracy in a given election. If 10% of seat projections in this election are incorrect, this amounts to 34 seats. If every single one of these incorrect projections is at the expense of the Liberals and the gain of Conservatives, the CPC still only wins the election by three seats, 157 to 154.

Of course, this is almost certainly not how the 90% accuracy would play out. Even if it was mostly entirely at the expense of the LPC – which is very possible given a tendency of pollsters to overestimate Liberal support – some of these losses would benefit the NDP rather than the CPC. Even then, there is still little to no chance that every single error would hurt the LPC. In some ridings, 338Canada cannot account for the fact that the LPC is running a popular city councillor, to use our very own Stormont-Dundas-Glengarry as an example. There are almost certainly stray ridings that have too many local factors at play in favour of the LPC for some unexpected unilateral rightward shift to occur.

So where does this leave us? The way I see it, it leaves us in a position where we need to make a few key assumptions in order to justify a CPC win; assumptions that are definitely possible, albeit very unlikely.

Assumption 1: The polls are incorrect.

There is simply just zero chance that the Conservative Party can win with the current polling numbers. The Liberals won the last two elections without even winning the popular vote. If they win this year’s PV by around 5% as the polls are almost universally predicting right now, they win the election – simple as.

Assumption 2: The polls are reaaaally incorrect.

This second assumption is certainly harder to justify. In 2021, Leger’s final poll correctly captured the election’s results to a T, coming within 1% of the vote share for every party. Outside of Leger, groups like Research Co. and Abacus also came quite close to the final results. In order to make the leap to this second assumption, we need to believe that pollsters like Leger (who has just released another +4 LPC poll) are incorrect; not only by a few percentage points, but by well outside of the margin of error. This is partially because of the final assumption:

Assumption 3: The model itself is incorrect.

In my opinion, this assumption is the hardest to reconcile with. The truth is that the CPC’s voting concentration is a demographic disaster. Where the party does well – such as in SDG, or even moreso in rural Alberta, the party does very, very well. But where they need to make up ground in order to win an election – such as in the GTA or in Quebec, they’re just short of winning in too many key seats. This is also to say that the Liberals’ vote is much more efficient. They eek out comfortable (but not overwhelming) wins in many ridings and do very poorly in ridings they’d never win anyways.

So, the third and final assumption we must make to predict a Conservative victory is that, due to a major demographic shift or some other reason, the model itself is incorrect now. As we’ve established in the previous assumptions, the Conservatives need to win the popular vote by a large magnitude to win the election. In order for this to fully be actualized, the way they win the popular vote needs to be different than the ways they’ve done so before – they need to win the GTA; they need to see some support in south Quebec; and they need to absolutely crush everywhere else.

The models project that there’s no way that this happens; that Quebec has rejected Conservatives altogether and that only the most suburban and white parts of the GTA have any chance at electing a Conservative. So, in order to predict a Conservative win, you must predict that, be it because of a wave of young Conservative voters or some other reason, the very polling models themselves are wrong.

Where does this leave us?

I am not here to say that the election is over. That would be incredibly naïve. I will, however, say that the information we have now is pretty damning; that, as I’ve illustrated, you need to take several prior leaps before you can jump to the conclusion that Pierre Poilievre will be our next Prime Minister. It’s possible, but it would require the ever-accurate institution of Canadian polling to be off by several magnitudes.

I opened with this, and I’ll close with this: it is true that the only poll that matters is the poll on election day. So if you don’t like the picture I’ve painted with the information that we have, then get out and exercise your civic duty to vote on the 28th.

And please remember: We are one nation greater than our political divide.