After what will likely not be remembered as a great election, Justin Trudeau managed to win a surprisingly strong minority. This happened with a mix of polling failure - namely the NDP vote clearly not coming out - along with an incredible vote efficiency of the Liberals. The number of ridings won by a few hundred votes is astonishing. Multiple ones in the GTA and pretty much every single close ridings in Quebec against the Bloc ultimately went red.
So like Trump, Trudeau won while losing the popular vote. With only 33% of the vote, this is one of the lowest score for a governing party in a long time. Coupled with a weaker turnout - around 65% - this means the Liberals lost a little over 1 mio votes this time around. Trudeau can pretend he won a clear mandate but I think many would disagree. With that said, the NDP will clearly not want to go back in election any time soon - they don't have money - so I suspect they'll pass whatever the Liberals want to pass.
The Conservatives won the popular vote and actually got more votes than in 2015, but that wasn't enough to beat Trudeau. The GTA is one of the reason, although it only explains a few seats - usually close races. No the real difference between the projections and the outcome is the fact the NDP vote didn't get out and was incredibly inefficient. Maybe it's strategic voting, but if that happened, it happened at the very end really. Because even the polls done on Sunday were still good for the NDP.
The weaker NDP allowed the Liberals to win quite a lot more ridings, in Atlantic Canada, the 416, etc. Coupled with the close races in Quebec where the Liberals systematically beat the Bloc, that explains why the Liberals are over 150 seats.
It's pretty shocking how a somewhat small polling mistake, coupled with incredible vote efficiency, can make a big difference. I'm not sure what you can do. Some races could have been called better by using more strongly the Mainstreet riding polls, but they were still severely overestimating the Liberals. And looking at them, many were wrong, especially in Quebec.
Anyway, here are some numbers. The model made the correct call in 293 ridings, that's an accuracy of 87%. That's quite okay for projections models. I mean you sometimes get lucky and go over 90% but anything above 85% is perfectly decent. Using the actual vote percentages, I would have predicted 128 CPC, 139 LPC, 28 NDP, 3 Green and 38 Bloc. With the correct vote percentages, 90% of the ridings are called correctly (so 35 mistakes). 10 mistakes in Quebec, same with the polling average or the actual percentages.
This is an accuracy rate similar (actually better) than during the last Quebec election. The difference? Mistakes cancelled out in Quebec (mostly). Not tonight.
This is an accuracy rate similar (actually better) than during the last Quebec election. The difference? Mistakes cancelled out in Quebec (mostly). Not tonight.
So this isn't a systematic failure of the model like in 2015 where, even with the correct percentages, I still had 60 mistakes. No the model worked this year and I didn't miss a late surge in the polls or anything (see below).
So the polls caused a gap of about 10 seats for the Liberals (or NDP). The rest is mostly the vote efficiency of the Liberals in Quebec (and a few ridings in the GTA). The fact the Liberals are just ahead of the Bloc in votes (34.2% vs 32.5%) and yet won the most seat is surprising. You'd expect the Bloc to win more seats thanks to the francophone vote (or the concentration of the Liberal vote among the English community). So clearly the Liberals got their votes out where it mattered. They won Gaspésie, Sherbrooke -this mistake annoys me as it was super close and that would have been a nice correct prediction to get- Chateauguay or Quebec with, in each case, only a few hundreds votes lead. There as well it looks similar to Trump with his numerous wins by close margins.
I count at least 10 ridings currently going Liberals by a margin of fewer than a 1000 seats. Imagine if those ridings had flipped differently. The Liberals could easily be at 147 seats, which is quite in line with what the model would project based on the percentages.
I count at least 10 ridings currently going Liberals by a margin of fewer than a 1000 seats. Imagine if those ridings had flipped differently. The Liberals could easily be at 147 seats, which is quite in line with what the model would project based on the percentages.
Finally, let's talk polls. They did okay. They almost all underestimated the Conservative but somehow that didn't have a big impact - mostly because they underestimated them in Saskatchewan and Alberta. The overestimation of the NDP is systematic with only two polls having the NDP within the margins of error. And even there, one of them is only because the sample size was small, thus larger MoE. The NDP's actual result is below every single poll. So clearly something didn't work for them.
Leger did spectacularly well while Ekos was, to the surprise of no one, the least accurate. As mentioned above, my average was fine. Again I didn't miss anything or introduce any bias. It's just that all the mistakes pretty much all went in one direction. Another way to put it: the Liberals took advantage of the NDP under performing and also won pretty much every single close race they were involved in.
Alright, that's it for me. As I said before, I don't think I'll make projections for another election. It's getting boring and repetitive. It's also very derivative, no matter how some might pretend otherwise. I'll leave the blog open and might write once in a while. And who knows, I might change my mind. See you!
