First projections for the 2019 federal election: A close race between Liberals and Tories

Alright, time to start making projections for the now on-going 2019 federal election. I apologize for starting a little bit later than expected. But you can expect regular updates from now on.

Also, I'm sure most of my regular readers will be interested in the new simulator. I have added the PPC (more details below). So have fun.

Alright, let's cut to the chase, here are the current projections of this site:

I'll need a little bit more time to get the probabilities going.

You can find the riding by riding projections here below.
1. What is the situation as the campaign is starting?

Believe it or not, we haven't had that many polls in the last two weeks, and especially not since the campaign officially started. So I don't believe we have a perfect idea of where things stand. Still, we can at least make an educated guess.

Basically, it's two-way race between the Liberals of Trudeau and the Conservatives of Scheer. This much is clear. There is absolutely no way any other party is in the running. This isn't your 2015 campaign with the NDP with a legitimate shot at power. If anything, the party of Jagmeet Singh might be lucky if it keeps its official status at the House of Commons. Yes, it's that bad for the NDP - although polls don't fully agree on the extend of the NDP's collapse, with IVR polls being much harsher than online ones.

The story of the last 12 months has been the rise of the Green party (they also won a second seat during a by-election on Vancouver island). The Green rising is a worldwide phenomenon and has been observed in BC in 2017 for instance. They aren't in a position -yet?- to win many seats but they could hope to finish of the NDP, at least in terms of votes. That is pretty nuts and we'll have to see if the Green vote is really solid as the campaign progresses. The current model doesn't have Green-specific adjustments. If this party stays as high as 10-12%, I'll look into that. I suspect the demographic of the ridings more likely to vote Green can be inferred.

Behind we have the Bloc. Under its new leader -Yves François Blanchet- this party appears to have regained some strength. After the collapse of the PQ last fall, it wasn't guaranteed that there would be a competitive Bloc Québécois. Number wise, polls put them slightly above their 2015 results (which was a record low). It's not great but again, they are still there. More importantly, now that the NDP has fully collapsed back to its pre-2011 levels (for many reasons, don't want to debate them), that offers a good opportunity to the Bloc to come back to 25 or even 30%. I personally can't imagine the Bloc becoming the number one party in la Belle Province again, but this party could at least go and challenge the Liberals in the 450 or in the more rural Quebec.

Finally, the new People's Party of Maxime Bernier exists and is alive. Actually it's doing not too bad. Multiple polls even puts this party around 5% which is quite impressive for a new party. I personally find it ridiculous not to include him at the leaders' debate, but that's another discussion.

Let's go back to the big two. The Liberals are clearly down from their 2015 results. But a strong resilience in Ontario and the NDP collapse in Quebec make it such that the LPC would be favourite to win the most seats at this point. Yes they wouldn't wipe the Atlantic like 4 years ago but they could make some significant gains in Quebec. This is where not switching to proportional representation could really pay off for Trudeau. There is a desire for change but it's not as high as what we observed 4 years ago, or in recent elections in BC, Quebec and Ontario. It appears that many people aren't the biggest Trudeau fans anymore but they aren't yet sure if they want to change.

As for the Tories, they seem to have made some gains but mostly west of Ontario. In Quebec they are stable and they still trail the Liberals in Ontario. In this province, any hope of becoming the first party again would rely heavily on improving the vote efficiency. I'm talking for instance of reclaiming some of the suburbs of Toronto.

The main issue for Scheer is that his path to victory requires him to gain in Ontario and hope that another party will prevent the Liberals from winning 55 seats+ in Quebec. I'm saying another party because I possibly can't imagine Scheer, with his fairly weak French, doing better than what Harper ever did in this province. In an ideal scenario for Scheer, the Bloc Quebecois would go back to 25-30 and cost the LPC many seats. Then Scheer would "only" have to do seduce the GTA. This last part is clearly the objective of the Conservative campaigns. The public transit tax credit is proof of that. It's very early in this campaign but it seems fairly obvious the CPC has decided to campaign conservatively (pun intended). No giant promise or anything, just one clear message ("hey middle class in the suburbs, do you want more money?") and voilà. Will it work? Well maybe. If the election were tomorrow, the Tories could well win. But the "Quebec problem" could prove insurmountable. And the soft-change voters don't seem fully convinced by Andrew Scheer. it might help him if his party didn't have a controversy regarding one of its candidates every day...

Also, there is the whole issue of what winning really means. If Scheer wins a minority, does he actually become Prime Minister? I don't know, I guess it would depend. While a Liberal minority would likely require the support of the NDP and the Green (and maybe the Bloc), it also means a Tory minority would need 1 or more of these parties to defeat Trudeau. Because I can't imagine Justin Trudeau just giving up the power even if he wins fewer seats. So Scheer either needs a majority (he's far from it right now) or he needs to get close enough. My guess is it'd make a huge difference if Trudeau would only need the support of the NDP (and Green) or if he'd also need the Bloc.

Anyway, we'll need more polling and data before making a better call.

2. How does the model work?

All projections models -and there are a lot nowadays- use basically the same principle. You take the results in 2015, you add the provincial or regional swing and you get the numbers for 2019. You can tweak this swing, make some adjustments, be pretentious and write that you "account for socio-demographic characteristics" (without explaining how of course) but that's pretty much it.

My model isn't fundamentally different. I do have regional coefficients estimated using past elections. Do they help? I think they can. They usually don't hurt at least. I also have an incumbency effect but this is super weak -and it's always the case in this country. Don't believe anyone telling you about a strong incumbency effect in Canada. I also account for when a long term incumbent retires. Data shows that costs the party around 5 points (sometimes more but that's the average).

I believe I do two things differently however. The first one is how I aggregate the polls. I do NOT distribute the undecided proportionally. Instead I currently allocate 40% of them to the Tories, 40% to the Liberals and 10% each to Green and NDP. Other parties don't get anything (well the Bloc in Quebec does). Why do I do that? Because redistributing the undecided proportionally is a bad assumption. And yes this is an assumption, even if it looks like the natural thing to do. It tends to overestimate the smaller parties. It also assumes that decided and undecided voters will ultimately vote the same, which is dumb. So yes my assumptions of 40-40-10-10 are subjective but I'm clearly letting you know. I believe the undecided will ultimately go for one of the main two parties. The CPC vote is older -so higher turnout- while the Liberals benefit from the usual underestimation of the incumbent. Again, you are perfectly allowed to disagree but remember that using a proportional redistribution is also making a big assumption. My track record on polls aggregation is that it has always helped me compared to simply averaging the polls. Yes it hurt me partially in 2015 because I didn't see the late surge for the Liberals. My bad. But my overall polling average was still better than the CBC one for instance.

I also don't really waste my time trying to weigh each poll differently based on a slightly bigger sample size or whatever. My experience is that giving equal weight (except for obvious exceptions) works perfectly fine. So my rule during the campaign is every poll within two weeks is included as long as I don't have more than 2 polls from the same firms (and the second one is heavily discounted). Polls from the last week get a weight of 1, polls of the week before are at 0.5.

The second thing is that I'll use the riding polls a lot. Not only will I use them to adjust at the riding level -usually if my current projections are really off- but I'll also aggregate them and calculate the swings using these polls only. I'll then do an average of the swing calculated using the riding and provincial polls.

You might not trust the riding polls (especially the ones from Mainstreet which has a mixed reputation -wrongly if you ask me) but collectively, they have proven to be gold. They had the large victory of Ford over the NDP in Ontario. They also pretty much had the perfect results in Quebec, while provincial polls ended up being so wrong we can add Quebec 2018 to Alberta 2012 and BC 2013 in the list of giant polling failures.

We currently only have a few of these riding polls, so I don't have any adjustment yet. But it'll come.

2.5 How did I add the People's Party?

It wasn't simple. Adding a party to a projections model is always tricky. Ultimately, after carefully looking at polls and correlations, I decided to add the PPC the following way: I assume that 40% of the votes of the PPC will be taken from the Conservatives. Then 15% each from the Liberals and NDP (polling correlations did indicate a relationship with these parties, so did an analysis by Abacus). The remaining 30% is taken from new voters or uniformly. This creates a situation where the distribution of the PPC vote resembles the CPC's but it isn't a perfect copy.

Maxime Bernier is assume to keep half of the votes he got as a CPC candidate. This assumption makes it such that the projections match the two Mainstreet polls done in the riding. Personally, I'm fairly convinced he'll win his own riding. 

Alright, that's all for now. Expect daily updates from now on, in English and French. Have a nice campaign!