The field work for our opinion data was conducted every day starting on August 26, 2015 and ending on October 18, 2015, the day before the election. Our sample was provided by Research Now, a North American leader in public opinion sampling. They worked with a consortium of public opinion sampling companies to produce the most representative samples possible. Respondents completed the survey online, using the Qualtrics platform. The total sample size is 37412 with an average of approximately 700 respondents per day. Sampling was set up so as to achieve nationally representative samples every three days. We further apply iterative proportional weights to improve representativeness. The instrument can be accessed here.
While we have a large number of respondents in many constituencies, we employ a second strategy in order to improve statistical precision. In short, we group ridings together that are very similar, using a statistical method called matching. We use information about all ridings from the most recent census, specifically, the population size, percent immigrant, percent unemployed, and mean income. We also use the vote shares for the Bloc Quebecois, Conservative Party, Liberal Party, NDP and Green Party in the 2015 election. We use the R statistical package blockTools. blockTools creates a measure of multivariate distance (using the above mentioned variables) between all possible pairs of units. We use these measures to create groups of between seven and ten ridings at a time. The result is that each riding is grouped together with the six to nine ridings that are the most similar across those characteristics. We then observe average opinions on various issues in those riding groups.