Okay, so I described the methodology I used to build a disaggregate microsimulation model of the election using multinomial logit discrete choice techniques. If that sounds intriguing, here’s the post with the details.
If that sounds confusing, I made a virtual Canada with 1.8 million people (1 for every 20 real ones); they have different likelihoods of voting and prefer different parties based on their demographics (like age, income, education) and where they are (mostly how the vote went last time, what current regional polls are, etc.) And then I made these guys have a thousand elections, under slightly different assumptions about what would effect people’s choices.
I call it a bad election prediction because I did it basically in the past weekend and so I didn’t have time to put in useful information like if candidates are incumbents or well known (or even if every party is running a candidate); and because I didn’t have access to current election poll response data so I used 2019 poll data and banged it roughly into shape; see the details post linked above.
But even if it’s not a good prediction, it’s still a prediction. So here goes:
The Top Line
The most likely result (in this model, not necessarily in reality — assume this caveat throughout) is a Liberal minority government, with a fairly reasonable chance of a Conservative minority. The ties would presumably go to the Liberals, since they’re the sitting government and since the NDP has pledged to work with them but not with the Conservatives. Although, intriguingly, in the 12 tie scenarios none of them had enough Liberal and NDP votes for a majority without the Bloc participating as well; 11 of the 12 had enough Bloc votes for them to choose a government on their own.
Here’s the overall results. The median shares should match the CBC Poll Tracker, since that’s what they were calibrated to match. But the range of seats and votes is my own. The Green and People’s parties are shut out in this simulation; probably in part because the incumbent Green candidates and party leaders don’t get any special benefit, which they should (but, again I ran out of time). It’s not necessarily a likely outcome that they get seats; but it is much more likely than 1 in 1000+ implied here. The box and whisker plots help show the ranges:
Seat by Seat
In the interest of having these election projections out before the election is actually over, here’s a link to download an Excel sheet with a summary of the seat-by-seat results; each seat has the mean vote share by party, as well as the proportion of the time each party wins the seat.
Here’s a tiny figure (which can be downloaded here) showing a selection of 50 ridings; those with the closest margins including at least a few from every region:
Again, the note at the top of the page calls them ‘bad’, and they’re certainly not the best; for instance, former Green leader Elizabeth May should get a higher vote share in Saanich-Gulf Islands as a high-profile incumbent, and the People’s Party should get less than 12% of the vote in Calgary Centre, since they did not run a candidate.
Maybe next time I’ll start two weekends before the election.