Qi (Emma) Wen

Prediction for Canadian Federal Election Aided by Canadian Community Health Survey

This project aims to develop predictive models for the Canadian federal election. We begin with explanatory analyses of two sets of data: (i) publicly accessible election related data, including the election outcomes, voter characteristics, opinion polls, and post-election survey information; (ii) extracted data from the Canadian Community Health Survey (CCHS) 2007-2018 on life satisfaction and other potentially associated social-demographics. We propose to predict for federal election outcomes using the information on longitudinal Canadian life satisfaction. Specifically, we model the federal election outcomes for each riding in change from its previous election jointly with its longitudinal life satisfaction since the previous election. Election data of years 2008 and 2011 and the CCHS data of 2008-2011 are employed to fit the model via both the two-stage estimation and the maximum likelihood estimation by the Monte Carlo EM algorithm. The analysis results indicate life satisfaction plays an important role in modeling election outcomes. Also young adults are more likely to vote for a new political party but male voters are less likely to do so. Using voter information or CCHS respondents information to model the election outcomes produce different estimation results. Two additional applications are presented to further illustrate the proposed joint modeling approach.

Keywords:

General linear mixed-effects model; Joint modeling; Logistic regression model; Monte Carlo EM algorithm; Two-stage estimation.