Overall Riding Prediction Performance

Overall, my predictions for Ottawa ridings were generally not that accurate.  The Brier score for my initial round of predictions was 0.61, versus an expected Brier score of 0.28 (lower is better for Brier scores).  This suggests that I was overconfident in some of my predictions.  In fact, if I had just given 50% odds to the top two parties, I would have gotten a Brier score of 0.56 (I initially put the Liberals as the third most likely party to win in Gatineau, so the average Brier score would be above 0.5).  Having said that, since my riding errors were so heavily correlated (i.e. I consistently underestimated the Liberals), my first round of predictions were closer to a single prediction underestimating the Liberals than a bunch of independent incorrect predictions.  This limits the strength of the conclusion I can draw.

I did much better in my second round of predictions, achieving a Brier score of 0.33 versus an expected score of 0.31.  My second round predictions were evidently much better calibrated and more accurate.  I evidently underestimated the importance of changes in the polls during the campaign.