Tuesday, October 20, 2015

i mostly called it, i just want to point out how i got to where i got to and what went wrong where it went wrong. my predictions were..

liberals: 160-170
conservatives: 95-100
ndp: ~45
bloc: ~35 - but i saw this as a max.

the liberals overperformed in quebec, so that's the major thing that i got wrong. otherwise, it would have been dead on.

error for quebec: the quebec polls were just impossible, near the end. the correct answer was "there are four parties in the margin of error, and the result is unpredictable". but, you can't do that. you need to pick something, even if you're aware it's really just a guess. aggregates break when you have a lot of movement near the end because they're designed to be conservative (as in not change a lot). i thought i picked up a bloc surge over the last 36 hours, and was guessing it would be something like libs 30, ndp 25, bloc 25, cons 15. i actually correctly pulled out low conservative turnout (or unchanged turnout since 2011), which was in direct contradiction to the media narrative; it struck me as absolute bollocks. so, i was thinking that the split would let the bloc come up the middle in many ridings, and leave the ndp with ~15. so, my quebec predictions were: bloc ~35, libs ~25, ndp ~15, cons ~5. what we actually got in quebec was libs 35, ndp 25, bloc 20, cons 20 - which let the liberals sweep. NOBODY picked that up.

almost nobody. i actually picked it up, in noticing that the ndp decrease seemed to primarily moving to the undecideds. but, then the polling firms stopped publishing that, and i was stuck in the dark.

i think that, when the dust settles and the numbers are crunched, what we will find out is that the liberals got a boost on a low turnout in quebec amongst left-leaning sovereigntist voters. people just rejected everything and tuned out. that is the reason that none of the polls seemed to have them so high in their published numbers; if you look at their unpublished numbers, i think it's something that will come out in a distinct swing from ndp to "they're all terrible, i'm staying home and watching the habs game."

errors for ontario: i predicated 90+ seats for the liberals in ontario and -25 for the conservatives. we ended up with 80 for the liberals and 33 for the conservatives. this is better than the models, but still off by about ten seats.

i got to 90 seats by looking at the 2004 federal election, which had the following numbers:

libs: 44.7
cons: 31.5
ndp: 18.1

that seemed to be very similar to what we were seeing in the polling. it netted the liberals 75 seats in 2004; there have been 15 seats added since, mostly in urban spaces, so i added it up to get to 90+. i also suspected the liberals were going to crest over 45, which made me feel ok about the + part. if you check a map of 2004, that would give the liberals a nearly clean sweep all the way around toronto (including brantford, milton, durham, thornilll, markham-unionville, barrie and kitchener-conestoga to name a few), some of hamilton and most of northern ontario.

actual numbers were:

libs: 44.8
cons: 35.0
ndp: 16.6

so, we see the cause of error here: the conservatives polled higher than expected in the gta. nobody had them as high as 35. northern ontario mirrored the 2004 election well, and they actually did better this time around in the 613 than they did in 2004. but that 3.5% difference for the conservatives (clearly at the expense of the ndp) was the ten point difference.

i still beat the models, though.

errors for alberta: i was arguing that the uniform swing was naive and overestimating conservative support; based on the polling, most of calgary and all of edmonton should have been in play. this was based on the idea of the conservatives polling at 53% province wide, which is what the aggregates were suggesting. in the end, they polled at 60% over the province. if you had plugged 60% into those models, they would not have predicted any seats for the liberals (and may have even handed over linda duncan's seat).

6/66 = 9%. doubled is 18%. this, ironically, recreates numbers that are closer to what the models were producing - but it's a total fluke & coincidence.

so, conservatives were being underpolled in alberta. it could have been a last minute swing back, even. i based a calculation on it. it didn't work out because the data was inaccurate.

that will be it for me, here, for another four years.

deathtokoalas, dvlghgfgjhjhfkhghjklka and all the other pseudonyms are signing out until the next cycle. track down my blogs if you're really upset by this.

www.cbc.ca/news/politics/canada-election-2015-majority-liberal-regional-narrative-1.3279126