Talk:Opinion polling for the 2019 Canadian federal election

Leger “GTA” poll with Toronto-only breakdown
https://leger360.com/wp-content/uploads/2019/10/Federal-Politics-Oct-16-2019.pdf

Is there anywhere we can fit in this Oct 10-14 Leger poll of the GTA that breaks down Toronto and Rest of GTA (or at least get in the Toronto part)?

It may be their last poll of the campaign other than their Quebec-only one Mikemikem (talk) 11:05, 18 October 2019 (UTC)

Advance Polls
So, it seems that poll firms have a seperate section for advanced polls. Do we separate by itself, since it's good info? JonathanScotty (talk) 14:24, 19 October 2019 (UTC)


 * This happens towards the end of every election campaign, when pollsters inevitably start reaching respondents who have already voted in advance polls. Personally, I don't think it would add much if any value to the page to start yet another new table of poll results among respondents who voted in advance polls: first of all, the sample sizes of advance voters are so small that the results are hardly statistically reliable (the report for the latest EKOS poll says as much); second, it's well known that the results of advance voting don't tend to be representative of the final overall election results (notably, there tends to be a significant overrepresentation of "hardened partisans" among advance voters), and this page is ultimately concerned with polls capturing overall voting intentions, nationally and regionally. Cheers, Undermedia (talk) 14:43, 19 October 2019 (UTC)

Rolling polls in the graph
Just to explain how the graph deals with rolling polls: If each daily update was inputted at full sample size—e.g. 1,200 for each daily 3-day Nanos poll—it would be as if each actual day of polling was counted 3 times in the calculation of the trendlines, which would give undue weight to the rolling polls. The solution is to give each poll a weighting proportionate to the amount of new, unique data it adds compared to the previous poll; in the case of Nanos polls that’s 400 new survey respondents per daily poll. This is why a fraction is put next to the sample size: the graph script multiplies the sample size by the fraction to proportionally reduce the inputted sample size of rolling polls with overlapping data.

So similarly with the Campaign Research polls, the polling done on Oct 16–18 is added a first time in the table with a sample size of 1,987, then those same data are incorporated a second time in the following day’s poll (1,987 + 1,554 = 3,541), and a third time in the last poll along with a second time for the Oct 19 data (1,987 + 1,554 + 1,498 = 5,039); so if all those polls are inputted at their total sample sizes, it’s going to give far too much weight to them in the graph. Cheers, Undermedia (talk) 02:19, 21 October 2019 (UTC)