Note: si vous voulez une version plus courte mais en français, cliquez ici.
The Conservative Party members are choosing their new leader in 4 days. Polling data has been rare (and
expensive to access). Some have even said this race was unpollable because of the closed nature of the race (only members can vote) and the electoral system chosen. This is why I decided to use the same indicator as for when I covered the
PQ leadership race last year: fundraising. Mixed with the available polling data, I think Maxime Bernier is by far the number one favourite to win on the 27th.
Predicting leadership races is always tricky and people need to keep that in mind. With that said, I'd go as far as saying that his win is almost guaranteed. There are mostly three reasons for this prediction:
- Maxime Bernier is ahead in both the polling and fundraising data. And the only threat he was really facing (Kevin O'Leary) dropped and officially supported him.
- The electoral system (giving a 100 points to every riding no matter how many members it has) is giving Quebec a very important role (6% of the membres but 23% of the points!). Bernier will take advantage of it as he'll easily win his home province.
- Second choices are unlikely to allow one candidate trailing to climb back up because members have to indicate all their choices at once and no formal deal was made between candidates. Also, we have no reason to believe Bernier isn't doing well for the 2nd, 3rd, etc, choices. On the contrary.
Before going into the details, here are the projections for this race. You have the chances of winning as well as the confidence intervals (at 95%) for the first round.
|
Percentages of points |
The entire field is here, in order:
Compared to the Mainstreet numbers, Bernier is higher. His lead in the fundraising has always been bigger than the one he had in the polls. Also, Bernier will benefit from the efficient votes from Quebec (worth a lot of points). Scheer dominates his province (Saskatchewan) but nowhere else. Actually, if I was only going by the fundraising data, Scheer wouldn't even be top 2. But I can't ignore the rising trend observed in the Mainstreet poll for him since early April. Leitch would be the main contender to Bernier based on the fundraising data, but she has been polling consistently low with Mainstreet and she's decreasing over time. Lemieux is here and this is a little bit surprising. But he has more donors than Scheer (although he raised less money). Also, the temporal adjustments based on the trend in the the Mainstreet polls (see detailed methodology below) has him increasing since April. Still, I wouldn't be surprise if the model overestimates his result but I won't modify the numbers arbitrarily. Finally O'Toole is currently in the top 5. His performance is similar to Scheer: polling better than fundraising and on the rise since early April.
Here are the possible results after the first round. This graph should convince you of the lead of Maxime Bernier. Even in his worst case scenario, he's still ahead. And that's with simulations including a ton of uncertainty, way more than what I usually do for a typical election.
If you think the advantage of Bernier isn't as big as what is shown here, think about this: thanks to Quebec, he'll likely receive at least 4000 points from this province alone, in the first round alone (he'll get 50% of the votes if not more). That is more points than all the other candidates are projected to receive Canada-wide, except Andrew Scheer! Quebec gives Maxime Bernier a huge advantage, one that is, quite frankly, most likely enough to give him the leadership. With the redistribution of the votes of the candidates once they are eliminated, we could imagine Bernier reaching 6000 points in Quebec alone, more than a third of what is required to win overall.
Also, O'Leary was polling higher than Bernier and raising just as much. But he officially threw his support behind Bernier and it seems that at
least 40% of his supporters are following through. Something that isn't surprising since the fundraising data regarding multi-candidates donors was indeed showing a strong link between Bernier and O'Leary supporters.
Still, Bernier is highly unlikely to win in the first round. His polling/fundraising numbers are too low and there are too many candidates that split the vote. This means the candidate who finished last will be eliminated and his/her votes redistributed. And so on. So, could Bernier "pull an Ignatieff" and win the first round but lose at the end? It's possible but highly unlikely for multiple reasons. First of all, all the
data available shows that there is no "anybody but Bernier" sentiment among the CPC members. Bernier is actually quite popular as a second choice. Fundraising data of multiple-candidates donors (people who gave money to more than one candidates) shows that. Polling data too. Secondly, Ignatieff lost to Dion for the Liberal leadership because the LPC was (back then) using actually delegates to vote. They had to physically be in Montreal and vote multiple times. So when Gerrard Kennedy threw his support behind Dion, he could literally tell his delegates who to vote for. But the CPC race this year isn't like that at all. Members have to indicate their subsequent choices right away. And we haven't seen
any formal alliance between candidates. So I think it's highly unlikely that we'll see the second choices leaning heavily for one candidate. When Dion won, over 90% of the delegates of Kennedy followed through and supported Dion. This is the type of rate you need if you finish 15 points behind and are hoping to climb back. But even if the second votes (by second, I also mean 3rd, 4th, etc) were to favour Scheer 60-40, that would likely not be enough to catch Bernier.
By the way, it seems the ballots were so complicated that as
many as 20% of them will not count because they weren't filled correctly. This is absolutely insane and it could potentially create some surprises (maybe one candidate will do better than projected because this candidate will have spent a lot of time making sure his/her supporters were filling the ballots correctly!).
You can see the detailed methodology below but let me simply say this: in order for Scheer to win some of the simulations, I had to make some adjustments. In particular I adjusted my fundraising numbers (so data going up to March 31st) based on the trend observed in the polls between early April to last week. According to Mainstreet, Andrew Scheer has been increasing quite a lot during that time (note: Scheer has always polled higher than what the fundraising data suggested). I used the trend in the polls not only to boost Scheer (well I adjusted everybody but Scheer is the one who benefit the most, along with Lemieux) in the first round, but in the other rounds as well (since I also adjusted the intentions among the 2nd, 3rd, etc, choices).
For Andrew Scheer to win, here's what needs to happen:
- Scheer to be underestimated in the polls and fundraising data. We are talking of a systematic bias here where Scheer overperforms the polls/fundraising in multiple provinces.
- Scheer to receive way more second votes than what the data suggests (and Bernier way fewer).
The path to victory for Scheer relies entirely on the polls/fundraising data being wrong. And quite wrong actually. So while it's possible, especially when we remember how unpredictable leadership races can be, it remains unlikely.
Methodology
1. Using the fundraising data available on Election Canada, I looked at the share of donors and shares of amounts for each candidate in each province for the first quarter of 2017. Ideally I'd have looked at the riding level but then most ridings would only have a couple of data points. Province level isn't perfect but it captures some of the disproportions introduced by the electoral system. Based on my research for the various leadership races in Quebec, it's not clear if the amounts or the number of donors is the best indicator. I thus averaged the two.
2. Given the number of points in each province (number of ridings x 100 points), I then allocate these points proportionally. As mentioned above, we would technically need to do it riding by riding. But unless one candidate has a crazily concentrated support in a few ridings in a province, my method should give us a good idea of where each candidate stands.
3. For the redistribution of the votes of the candidates that are eliminated: I used the information from the data of the donors who contributed to more than one candidate. This gave me a big matrix of cross-voting intentions. For instance, I could see how many donors contributed to both the Bernier and O'leary campaign. Again, far from a perfect measure but it's better than nothing.
4. Adjustments based on polls. As I said, the only good polling data was from Mainstreet. But I wasn't gonna pay $110 a month to have access (actually $1000 if I wanted to then use the data on my blog). With that said, Mainstreet published the data publicly on some occasions. One of them was for the period going from April 11 to 13 (thus not too far after the end of the fundraising data going up to March 31st). Another was
April 29-30 (thus providing us with a good idea of what happened after O'leary dropped). Finally, they published one last week, conducted between
May 11 to 14.
Maybe surprisingly, the numbers I got with the fundraising data were actually quite close to the ones from Mainstreet. Bernier would usually be higher using the fundraising data, so would Leitch, while Scheer would be a lot lower for instance. Again though, overall it was quite similar. I thus adjusted my numbers partially to the polling ones. More importantly, in order to account for the trend between early April to now, I adjusted the numbers based on the trend in the polls. I used the same adjustments for the second choices. These adjustments help Scheer quite a lot. As a matter of fact, without them Bernier was winning 100% of the simulations.
5. Simulations: the share of a candidate in a given province is randomized. For instance for Bernier in Quebec, his share is around 50%. In some simulations it's only 40% while in other it'll be 60%. The margins of error used were of 7%, thus much wider than for my typical simulations for an election. Why? Because leadership races are more unpredictable.
For the 2nd (and subsequent) choices, I also randmonized. So in some simulations many supporters of Leitch would then vote Bernier while in another simulation they'd go to Scheer. Here the margins of error are of almost 10% because we really have limited information regarding these second choices.
I repeated this process 10,000 times, each time until one candidate reached 16901 points. I then counted the number of wins to get the probabilities of winning.