Estimating voter migration in Canada using generalized maximum entropy

As voters switch political preferences from election to election, understanding the magnitude of voter flows among parties and transitions between voters and non-voters is an essential element of political analysis. As exit polls are uncommon in Canada, voter migration can also be estimated using suitable statistical techniques. Backing out micro-level voter migration probabilities from macro-level election data is a problem of ‘ecological inference.’ This paper uses the method of generalized maximum entropy (GME) to estimate voter migration patterns for the two most recent Canadian federal elections (2004 and 2006) and two most recent provincial elections in British Columbia (2001 and 2005). The estimation results answer important questions about voter behaviour in Canada. These results will be of interest to political scientists, historians, and politicians, as well as econometric practitioners who wish to estimate voter migration.