Crunching the prediction numbers

As the results finally become known in all 308 ridings (some recounts may occur, but none were apparently close enough to qualify for an automatic recount), the two big seat-by-seat projection websites did a self-analysis to see how they did:

DemocraticSPACE got 25 of 308 ridings wrong, for a 91.9% accuracy rate

Election Prediction got 27 ridings wrong, for a 91.2% accuracy rate

Those sound like impressive numbers, but I wondered how significant that is when so little changed. They both got Edmonton Strathcona wrong, for example, but then again so did all the pundits. They also didn’t predict three seats shifting from the Liberals to Conservatives in New Brunswick.

Looking at the election results (Wikipedia has a riding-by-riding breakdown on one page), I see that the incumbent (or incumbent’s party) won in all but 41 ridings. So if you blindly picked the incumbent to win in all 308 ridings, you’d have an 86.7% accuracy rate.

That makes 91% sound a lot less impressive.

Then again, in an election where only 25 of 308 winners got more than half the vote, predicting anything is a throw of the dice.

Leave a Reply