This will likely be my last update before the final projections on Sunday. The race remains tight but the Liberals have a clear advantage thanks to their vote efficiency. With that being said, the chances of a LPC majority are half that of the chances of a CPC plurality. Both 'surprises' would only require a minor polling upset but the Tories do seem closer to finishing first than Trudeau is from a majority. Vote efficiency works both ways: it makes you win a ton of seats by small margins but it also means the next list of winnable ridings is further away.
So here are the projections and I look at what numbers we'd need in Quebec and Ontario to see a Liberal majority on Monday (or Tuesday I guess since mail-in ballots won't be counted until the next day).
The wild card is the PPC who is clearly enjoying some momentum -- and the events in Alberta will help. Can the PPC hurt the Tories enough to split the vote? Or maybe straight up stealing seats in Alberta or Southwestern Ontario? Not impossible. But I admit that a model such as mine will almost surely miss this. However, looking at other sources (Google trends, etc), I personally think the odds for one PPC seat are not bad. Ekos has also indicated that those voters are motivated and committed, so it's unlikely they'll switch to the CPC on Monday.
Map
Detailed projections
Proj Canada 17 September 2021 by bryanbreguet on Scribd
What would it take for a Liberal majority?
I had looked at this previously, back when the Liberals still had a pretty massive lead over the Bloc in Quebec. Since then, the lead has mostly evaporated and the lead over the CPC in Ontario has also shrunk a little bit. This means a majority is far further for the Liberals. See the new probability curves.
You might be wondering why the chances aren't higher at +4-5 (which is where the Liberals currently stand in Quebec and Ontario) since my projections above show a 15% chance of a majority. The reason being that the simulations in the projections use +4-5 as a starting point and then apply the effective margins of error. That means the 15% chances mostly come from the simulations where the Liberals outperform the polls. But if they really do win Ontario by 5 points and Quebec by 4%? The chances of a majority are almost zero.
We also see the limited usefulness of Ontario for the Liberals. Outperforming the polls there wouldn't help Trudeau much as he is already winning most of the seats. He would just prevent some losses. Quebec, on the other hand, is more promising. If the Liberals could win Quebec by a comfortable margins (10 points or better), then a majority is all of a sudden a lot more likely. And just to be clear, the orange line does indeed show that even winning Ontario by 18 points over the CPC wouldn't guarantee a majority. But you need to interpret this as the Liberals staying the same everywhere else (including Quebec) but winning Ontario like 45% vs 27%. Again, the Liberals already win a ton of seats in Ontario, at some point there are diminishing returns.
What this shows is the Liberals need a systematic polling error. They can't hope to simply outperform the polls in one big provinces. They have fallen too far in the West, especially in BC. Looking at the simulations, the Liberals need (in average), at least 35.7% Canada-wide for a majority. That means a polling error of about 3 points. That is more than the average error observed during federal elections (note: I'm not talking of a 3 points deviation from any given poll but from the polling average). Given that a good share of the ballots have already been cast (and we have no reason to believe the Liberals did better than the Tories during the advance voting), the Liberals need a clear national lead in the polls by Sunday. Or they'll need some crazy vote efficiency but we should be reaching a point where the Liberals can't really improve on their efficiency from 2015 and especially 2019.
