Some early reflections on the Ontario results

OK, I know I said I would be en route to Mexico City (I am) and, therefore, not newsletter-ing, but I can only lose at Solitaire so many times before the laptop comes out.

I have an unfortunate limitation: I’m a very poor bullshitter. So I’ll just say–I have mixed feelings about my polling results. On Wednesday, I presented 3 different turnout scenarios: one that is representative of the general population (my real prediction), one weighted to the age/gender proportions of the 2021 federal election, and one isolating only the voters who were most motivated to vote. 

My prediction had the OLP and NDP votes within the margin of error, had the Greens slightly outside the MOE (3.4%), and was high on the PCs (5.13%). I’ll be spending a lot of time with the data in the coming weeks, but I think this comes down to two things: 

1. As I wrote earlier in the campaign, I chose not to report people’s intention to vote for a party other than the PCs, OLP, NDP, or Greens because I think it’s often over-reported. In that article, I admitted that I might come to regret this decision and… I do. Approximately 3.7% of Ontarians voted for an “other” party/candidate. This probably led to some protest voters being forced to side with the Greens and New Blue Party voters (1.6%) parking with the PCs. Easy fix for next time. 

2. As my polling showed, by the end of the campaign, PC voters were less motivated to vote than OLP or NDP voters. Given the very low turnout (45.4%, compared to 2022’s 43.5%), I think this was a big factor. I’ll be chewing on how to integrate the likelihood of voting into my future predictions. 

On that note, I mentioned in my last newsletter that I don’t yet have a likely voter model that I feel confident in, but my stand-in crushed it. My average error when looking only at “very motivated voters” is .97%, which is just… very freaking accurate. It almost makes me wish that I had made that my capital P prediction, but it wouldn’t have been made in good faith. It would have been a quasi-guess and that’s not how I approach research. 

As a data scientist, I see all data as an opportunity for learning, so I’ll be spending some time working backward from the results to see how I could have improved accuracy. 

On a separate note, there’s a chance that I will do nightly tracking for the federal election. If so, what would you want me to keep consistent vs. my Ontario content? Is there anything you wish I had focused on more? Anything less? Shoot me an email if you have any thoughts!