The Green Party of Canada will announce a new leader this evening.
Toward the end of the party’s 2020 leadership race, I used public donation data to project that Annamie Paul would defeat Dimitri Lascaris 54%-46% in the final round, with Courtney Howard placing third with 21%. This proved to be pretty accurate: Paul defeated Lascaris in the final round 55%-45%, and Howard placed third with 25%.
What about the 2022 race? Is it possible to accurately project the outcome this time?
Well… maybe. Read on for my tentative projection. After that, I’ll discuss some of the irregularities in this race and how I tried to compensate for them.
Elizabeth May wins in the final round,
defeating Anna Keenan 53% to 47%.
Here are the first round standings I’m projecting, along with the expected outcome for each candidate:
|PEDNEAULT Jonathan||24.6%||Eliminated 4th|
|WALCOTT Chad||17.6%||Eliminated 3rd|
|GNOCCHINI-MESSIER Simon||2.2%||Eliminated 2nd|
|BARON Sarah Gabrielle||2.0%||Eliminated 1st|
I recommend taking this projection with a grain of salt. If you’re interested to know why, read on. You can also subscribe to receive notifications of my future posts.
This projection was created using a Monte Carlo simulation, a way of simulating the outcomes of a system that has uncertain inputs. To use this method, you run a simulation an extremely large number of times, randomly varying the inputs each time. By adding up all of the outcomes, you can form an overall picture of what’s likely to happen.
For our simulation, the uncertain inputs are the level of support for each candidate. We can estimate support within a certain range by looking at donation data, but we can’t say for sure that we’re right. Running the Monte Carlo simulation helps us to see which outcomes are more likely, which ones are less likely, and which ones never occur at all.
This approach is a great way to deal with complex systems—for example, a multi-round election where a candidate is eliminated each round and their support transferred to other candidates.
However, a simulation is only as good as the data you put in, and this year’s data comes with some caveats.
To begin with, there’s simply less data available this time around. Since 2020, The Green Party of Canada Has Lost 11,000 Members and one third of its polling support, which means fewer people to make donations. This leadership race is also much shorter than the one held in 2020, and there are only six candidates, compared to ten—which means fewer candidates to attract donations and less time to receive them.
As a result, there are less than 1/5 as many donations as there were in the 2020 race. It looks like this is still enough to make a reasonable projection, but there’s no way it will be as accurate as 2020.
As an extra wrinkle, the campaign return for Simon Gnocchini-Messier hasn’t been posted by Elections Canada yet, leaving a gap in our data for October. To compensate for this, I’ve assumed that Gnocchini-Messier’s support in October was similar to his support in September. As a less prominent candidate, it’s unlikely that an error in estimating Gnocchini-Messier’s support will affect the outcome.
Two pairs of candidates are pitching themselves as “co-leaders”, campaigning together and in one case sharing a website.
However, the current Green Party constitution doesn’t allow for co-leadership. As far as voting is concerned, each candidate is being ranked individually by members, and a single candidate will be chosen as the winner.
This presents several challenges for our simulation.
First of all, we need to think differently about how votes are transferred between rounds. When one candidate from a pair is eliminated, how much of their vote will go to their “co-candidate” and how much will be distributed to others? Based on the data, I’m expecting upwards of 90% of transferred votes to go to a candidate’s partner, if they have one still in the race.
The second issue is that we would normally treat donations as a proxy for voting intentions. But with many people donating equal amounts to both members of a pair, how can we know who they’ll rank higher? What I’ve done is to look at members who only donated to one of two partners, use that to estimate which partner is more popular overall, and then assume that the 1st and 2nd vote preferences will be distributed in a similar way.
One final wrinkle is that Elizabeth May’s website explicitly instructs her supporters to rank Jonathan Pedneault first. However, her social media in recent weeks has not repeated this instruction. I also don’t know what was or wasn’t said at May’s events or direct communications to her supporters. So, how many people actually received this instruction, and how many will choose to obey it?
In fact, does it matter at all? To test that question, I ran some alternate simulations where I transferred varying amounts of May’s first-round support to Pedneault. In the simulations where more than one tenth of May’s supporters voted for Pedneault first, he takes her place as the winner.
Ultimately, I don’t expect that many of May’s supporters will rank Pedneault first. In an election where thousands of people are expected to vote, many of whom don’t interact with candidates at events or online, I don’t think May’s instruction will reach enough people to change the outcome. However, it’s impossible to know this for certain. This makes Pedneault something of a dark horse.
It also raises an interesting question: if Pedneault wins based on votes from people who actually prefer Elizabeth May, did he really win?
Finally, we have the issue of regional support. In 2020, the front-running candidates (particularly Dimitri Lascaris) attracted support from across the country, proportional to where the party’s members were located.
This time around, candidate support is much more regional, with both gaps and spikes. For example, Anna Keenan has received the most donations of any candidate, but a full 30% of those donations have come from Keenan’s home province of PEI, which has only a tiny number of party members compared to Ontario or BC.
When we use donations to estimate support, what should we make of a situation like this? Does Keenan’s absolute dominance in PEI imply that she should be seen as the favourite in Ontario and BC as well? That seems unlikely.
Other candidates also have regional patterns, with May’s support concentrated in BC, Walcott’s heavily concentrated in Quebec, and Baron’s in Ontario. (It also appears that May and Pedneault may have divided up their travel to cover more provinces.)
To compensate for regionalized support, I used two different methods of estimating support, and ran one million simulations with each method.
For the first method, I used the data as-is: every donation is treated as equally important, regardless of where it comes from.
For the second method, I adjusted donations on a province-by-province basis. For each province, I scaled each candidate’s support according to the number of members in that province. With this method, dominating PEI can only get a candidate as many votes as there are party members in PEI.
So How Reliable is this Projection?
I don’t have any way to quantify how reliable or unreliable this projection might be. It’s definitely less reliable than the projection I made in 2020.
Compared to the data we had available in 2020, we have far less data overall, “co-leadership” slates, and regionalization. Each time we do something to compensate, we’re introducing a human judgment call into what started off as a purely mathematical model.
Having said that, I do feel pretty confident in the broad strokes: the race will go to the final round, and team May–Pedneault will win.
If you’re interested in a few extra details, including a round-by-round projection, read on. You can also subscribe to receive notifications of my future posts.
Just for fun, here are a few more details from the Monte Carlo simulation. These should be taken with a bigger grain of salt than the main projection. (The finer the detail, the better the data you need.)
Here’s how things will play out in each round, according to the simulations.
Rounds 1 and 2: Sarah Gabrielle Baron and Simon Gnocchini-Messier are eliminated. Each of these candidates is projected to capture around 2% of the vote. It appears somewhat more likely for Sarah Gabrielle Baron to be eliminated first, but the data for these two candidates is extremely sparse, so it’s almost meaningless to give an expected order. Having said that, out of two million simulations that I ran, there was no outcome where either of these candidates remained in the race after Round 2.
Round 3. Chad Walcott is eliminated with about 20% support. In a very small minority of simulations (2.5%), Walcott remained in the race, and one of the other candidates was eliminated instead, usually Jonathan Pedneault. For this to happen, Walcott would need to greatly overperform compared to his donations, and another candidate would need to greatly underperform.
Round 4. Jonathan Pedneault is eliminated with about 25% support. In about 10% of simulations, Elizabeth May was eliminated instead.
(What about Anna Keenan? It’s extremely unlikely for her to be eliminated in Round 4, having just inherited the majority of Chad Walcott’s votes. In fact, in about 3% of simulations, Anna Keenan wins by a hair in this round, thanks to Walcott’s votes. There is also an extremely small chance, less than 1%, that Elizabeth May wins in Round 4—but only if Jonathan Pedneault was eliminated in Round 3.)
Round 5. Elizabeth May wins the race with 53-56% support. Other outcomes are possible, but very unlikely. In about 6% of simulations, Jonathan Pedneault survives Round 4 instead of May, and goes on to barely win the race. In about 1.5% of simulations, Anna Keenan wins in Round 5. (Keenan is actually more likely to win in Round 4, as described above. Overall, she wins in <5% of simulations.)
Weighted vs. Unweighted Results
I mentioned earlier that I ran one million simulations without compensating for regional support, and another million with compensation. What do those results look like on their own?
Here are the winning chances for each candidate using the unweighted data:
|Candidate||Overall||Round 4||Round 5|
|BARON Sarah Gabrielle||–||–||–|
And here are the winning chances with the data adjusted according to the number of members in each province:
|Candidate||Chance||Round 4||Round 5|
|BARON Sarah Gabrielle||–||–||–|
Yes, you’re reading that correctly. Even after one million simulations, the only outcome using the weighted data is an Elizabeth May win in Round 5. Keeping in mind that May’s strongest competitor is Keenan, and Keenan’s support is strongest in the smallest province, it makes sense that the weighted data would strongly favour May.
Regardless of who wins the leadership race, the Green Party of Canada still faces an existential crisis. The party has lost 11,000 members, laid off half its staff, nearly gone bankrupt, and closed its head office.
Many members are familiar with these financial and reputational troubles. However, the real threat to the party is something entirely different, which I’ll discuss in an upcoming post.
To be notified of new posts, please subscribe—it’s free.
4 thoughts on “Projection: May to Win Green Party Leadership in Final Round”
Well, that’s depressing 😉 . I hope you are wrong because that would mean that the voting members would choose for the party to stay irrelevant.
LikeLiked by 1 person
Absolutely adore the pic! And I think you’re right on the money.
If it’s any consolation, I feel significantly less certain about this year’s projection than I did about the 2020 projection. 🙂 I will have a post in a few days discussing the GPC’s relevance, reputation, and $/member woes in detail.
LikeLiked by 1 person
Thanks! A team effort between the DALL-E AI and myself. 🙂
LikeLiked by 1 person