We’ve compared our predictions of the 2019 ๐จ๐ฆ election to the actual votes. Overall, we were within 5% with no obvious geographical biases, though we did slightly overestimate support for the NDP at the expense of the Liberals. I think weโre on to something good here!
With an agent based model you can explore interesting scenarios. Our latest post models the ๐จ๐ฆ election with another Liberal scandal, new Conservatives climate change policy, or proportional representation. The results are not obvious, showing benefits of non-linear modelling.

Spatial analysis of votes in Toronto
Our predictions for the recent election in Toronto held up well. We were within 6%, on average, with a slight bias towards overestimating Keesmaat’s support. Now we’ll add more demographic richness to our agents and reduce the geographical distribution of errors www.psephoanalytics.ca/2018/11/r…

We’ve completely retooled our approach to predicting elections to use an agent-based model. Looking forward to comparing our predictions to the actual results tonight for the Toronto election!
