As I posted in my previous entry, I think I made a fairly good prediction of the result of the election, based on the information I had available to me. For my final result model, I used a public opinion poll released on Sunday, May 1, 2011, as it sampled a very large number of people and provided excellent urban breakouts for popular support.
Unlike various projection projects, I chose not to aggregate the polling data. Each polling firm uses different methodology which produces different results, so, just like you cannot aggregate apples and oranges, you really cannot aggregate polling data.
|Unadjusted Ekos Data Projection:||150||116||37||5|
|Adjusted Ekos Data Projection:||159||114||29||6|
|Intuitive Based on Adjusted Ekos Data:||156||76||46||30|
Amusingly, the closest prediction to the actual results occurred after I adjusted the results by 3%, but left out my own intuition. My thought process was that, after the nature of some of the Quebec NDP candidates came to light (such as Ruth Ellen Brosseau in Berthier-Maskinongé, an assistant manager at Oliver’s Pub at Carleton University, who doesn’t speak French, went on a trip to Las Vegas halfway through the campaign, didn’t even tell her boss at Oliver’s Pub that she was the candidate for the riding, and has never set foot in the riding ) came to light, the voters would change their votes reducing the number of elected NDP MP’s. However, this did not happen. The desire for change was too great, and, as we have just learned, the public overwhelmingly votes for the party and/or the leader rather than the local candidate.
More analysis is needed on my part to determine what caused the incorrect predictions in 20% of the ridings, and build in any possible corrections for next time. Thankfully, I now have 4 years to figure it out.