The Accuracy of Public Polls in Provincial Elections

Abstract

This study extends work on the accuracy of polls released in the final week of election campaigns in Canada, using data from the nine provincial elections held in Canada between 2011 and 2013 to identify what might affect the accuracy of those polls. Specifically, we attempt to empirically test two arguments - mainly that some methodologies are better than others in measuring voter preferences and that the nature of the election itself might make accurate forecasts more difficult. We find that absolute change in voter turnout was the strongest predictor of polling accuracy. In contrast, sample size, survey mode, or electoral volatility were statistically significant predictors of polling accuracy.