Election Noise 

 My home country is in chaos - of a sort. With the dissolution of Parliament on November 29, Canada is heading into a federal election.  

 As a multiparty parliamentary democracy, predicting political outcomes in Canada isn't simply a matter of reading a thermometer. Of course, it isn't even that simple in a two-party system, but it gets me thinking about prediction methods. 

 I've been working with Gary on JudgeIt, a program used to evaluate voting districts for a variety of conditions, designed for a two-party system. With an emphasis on Bayesian simulation, its methods make use of uniform partisan swing -- a shift in the percentage of voters moving from one party to the other, and in the same proportion in each district -- to determine the likely outcomes given a set of covariates and a history of behaviour in the particular system. 

 What caught my attention was a series of election prediction websites, making use only of previous election information, which allows the user to input what they expect to be either the vote shares or swings in support. This by itself is mathematically unremarkable, but may keep political junkies up hours. 

 The real question of interest remains: by what process can a system predict who will gain whose votes in a shift in support? In most Canadian ridings (districts), seats are contested by three parties: from left to right, the socialist New Democrats, the incumbent Liberals and the opposition Conservatives. For the most part, votes lost by an outer party would naturally flow to the Liberals. In this election, however, a scandal which led to the election call may prove to cost the Liberals a good deal of support. 

 Since geography -- and hence, demography -- dictate much of the Canadian political climate, I have no doubt that the appropriate covariates are out there, waiting to be measured and/or analyzed. In the meantime, I'm keeping my head away from election speculation and looking to see if this problem has already been solved.  Anyone out there have any suggestions?