Tuesday, May 29, 2018

Ontario projection unchanged despite Ipsos poll

Despite the Ipsos poll, I am not changing my projection. You may wonder why. The answer is in the details of the poll itself. Note as well that I've checked their previous poll this election, and they seem to be doing some heavy weighting.

The most recent Innovative poll has males 18-34 weighted at 144, but only found 97 actual men between 18-34 to answer their poll. This means the answer of these 97 is "upweighted" as though 144 people answered. This is, of course, a gross over-simplification of how weighting works, but gives you an example of what it means. This means each respondent is upweighted to 148% vs the normal 100%.

Leger is more open with its weightings; allowing us to see they had 324 people tell them they were voting PC, and 304 for the NDP; but weighted this to 299 for the Tories and 295 for the NDP. This is not unreasonable, as, it is hard to contact those who work long hours, or are in school, etc, meaning the retired are simply easier to contact, and older voters trend right-wing.

Mainstreet also has re-weighted the parties, with the NDP moving from 518 to 522 and the Tories from 599 to 585; again, not unreasonable given the large number of older respondents in that particular poll.

None of these polls however compare to some of the things in the Ipsos poll. They managed to find 55 people without high school diplomas, but have up-weighted them to 177, meaning 269% vs the standard 100%. There is also a nearly doubling (55 to 101) of respondents in the "central" area.

Ipsos is the only firm of the bunch to not up-weight the NDP, but down-weight the NDP. Their may 15th poll had actual respondents break 310 to 309 for the NDP, but weighted this as 281 to 328. On the 22nd 349-306 became 311-306, and on the 29th, 405-376 is 355-391.

Lets pull back for a moment and go on a fun tangent involving math.

Each riding in Ontario is supposed to contain the same number of people. Sure this is not exact, ridings, by design, can very by up to 15%, some ridings are intentionally very unpopulated as they are physically large, and others can see high growth in between the time they were made and now; but for quick back-of-the-envelope calculations, especially those not involving Northern Ontario, you can assume each riding contains 1/124th of the population. This would mean a region containing 8% or so of the population of Ontario would be 10 or so ridings in size. This would be the size of the "Central" region in the Ipsos polls.

The "GTA" would have to be 59 or so, with some room to be fuzzy due to the aforementioned variances; East 15, and Southwest 33.

When you actually map it out, the divisions are all logical, and, in fact, match well with some of the divisions proposed by wikipedia.

I don't fully understand what the problem is. After staring at over a dozen Ipsos polls (those prior to the writ drop don't have this problem) I've not been able to figure it out. If they were accidentally assigning voters to the wrong regions, we would not see the breakdowns that we are.

Regardless, the weird weighting problem with the polls makes me have doubts. As such, unless I see a poll from another firm suggesting the tories are up from that firm's own previous poll, I will not be updating the projection for any PC "rise"

No comments:

Post a Comment