Wednesday, October 14, 2015

my issue with alberta is that you're distributing the numbers uniformly. i'm going to throw some data at you, regarding a mainstreet poll that was released yesterday in edmonton mills that had the conservatives at 36%. they got 61% in that riding last election.

now - i know - the polls could be wrong. the respondents could be lying. but let's take them at their word for a moment.

the conservatives got 66% in alberta last election, and grenier's model had them at 53 last i checked. that's a 20% drop. but what all of the models do is then distribute that 20% across the whole province. that would suggest they should be polling around 45% or so - which is what the models do say. so, how do you explain the discrepancy, while taking the poll at face value?

i would argue that the uniform swing is the error. and that this is something that we have more than enough data to come to good conclusions regarding.

if you look at the data in rural alberta going back many years, it doesn't really change much. they win seats by huge numbers. 70. 80. it's big wins. and there's not a lot of variability.

but, if you look at the numbers in the cities, you see a lot more variability. they even elect the odd liberal or new democrat from time to time.

that suggests to me that when you see a 20% drop, province wide, you should not distribute it equally across the whole province but proportionally in the areas most likely to swing. this is an observation that is...not exactly unique to alberta, but certainly not applicable country wide.

now, we don't have to do any algebra here because, by chance, the combined populations of calgary and edmonton are close to half of the province. so, i might suggest that rather than say the conservatives are down 20% across the whole province, they're probably down 0% in their rural strongholds and 40% in the cities of calgary and edmonton. that balances out to 20% province wide.

as it turns out, that would predict a result of around 36% in edmonton mills - which is consistent with the polling. and, this relation also holds in the other recently polled alberta ridings.

i would consequently challenge the idea that the conservatives running at 50% suggests they'll perform well province wide. i might rather suggest that that would imply they're running at the same 65+ they always run at in the rural regions, but probably an average of around 35 in the cities. and that could open up around ten seats.

50 v 49 might not make much of a difference. but 66 v 53 could be pretty substantial.

something similar would be true in ontario, but perhaps less drastic. but this particular election is seeing a 30% swing in ontario. a uniform model would suggest 30% everywhere. do we really think the liberals are up 30% in haldimand-norfolk, or simcoe grey? that's maybe a little hard to believe. maybe it's more believable that they're up 40-50% in cambridge and brantford and barrie, and maybe 5% in simcoe grey, if that.

if it was a smaller swing - 7-8% - the differences in projections from a uniform swing model and a more targetted one would be minimal. but, at 30%? anything using a uniform swing is going to be broadly inaccurate in ontario in this specific election. that applies to all models that i am aware of.

rather, if you want to know what ontario is going to look like, i would suggest consulting a map of the last provincial election, which had the liberals at 39% and the conservatives at 31%. that gave the liberals an almost clean sweep all the way around toronto - including brant, cambridge, kitchener, newmarket, barrie, durham, peterborough and northumberland. these are the likely candidates for the very large local swings that a 30% province-wide boost is likely to produce and, it seems likely that the liberals will poll higher than 39%, while 31% for the conservatives looks about right. that might mean an extra seat or two around kitchener.

i'm not even going to try and guess how you deal with quebec and bc, other than to point out that the most substantial movement in quebec appears to be from the ndp to the liberals, and that will both help the liberals where they are competitive and help the bloc where they are not.

i am suggesting extreme caution regarding all models in this election, due to their use of uniform swing. it can be useful in an election with small swings. but, big swings break this kind of model.

i believe that calgary and edmonton are in play, and that the liberals above 40% will sweep not just the gta but into the furthest reaches of communities around toronto, from brantford to kitchener, newmarket and barrie, and peterborough south to the lake - with the perhaps sole exception being around whitby and oshawa.

www.huffingtonpost.ca/2015/10/14/canada-election-polls-ontario-liberals-conservatives_n_8294430.html