two suggestions on the model:

1) when you have riding polls, you should use them.
2) instead of taking just the 2011 results, you should take a weighted mean of the last several (say, five) in order to find the "centre of gravity" in the riding. this will help to identify true bellweathers and balance out fluke results from the last election.

http://www.theglobeandmail.com/news/politics/how-our-election-forecasting-model-works/article25371747/