A follow-up to Thursday's Ontario election

This post will take a look at how my seat projection model turned out, as well as how the election results would have turned out under other electoral models.



Seat Projection

A few days prior to the Ontario election, I posted a seat projection on this blog.  After a couple of further polls came out, this was my final projection:

LIB: 58 (-13)
PC: 31 (+5)
NDP: 18 (+8)
Required for majority: 54
Odds of LIB majority: about 2/3

The actual result was as follows:
LIB: 53 (-18)
PC: 37 (+11)
NDP: 17 (+7)

This makes my overall error 12 seats, or 4 per party.  Under normal circumstances, I wouldn't call it that bad, but in such a close election it means the difference between a majority and a minority.

In terms of the individual seats, my accuracy level was also acceptable.  I projected 94 of 107 ridings (87.9%) correctly.  On the whole, not bad, although saying nothing would change would have been 83.2% correct.

Accuracy breakdown by region:
Central Ontario: 10/11 (90.9%)
Eastern Ontario: 12/14 (85.7%)
Greater Toronto: 17/18 (94.4%)
Hamilton/Niagara: 10/10 (100%)
Nothern Ontario: 9/11 (81.8%)
Toronto: 20/22 (90.9%)
Southwestern Ontario: 16/21 (76.2%)

I put Southwestern Ontario last because this was where most of my errors were found.  There were five seats in this region that I incorrectly called for the Liberals (4 went PC, 1 went NDP), and only one of them had the eventual winner within 5%.  Chatham-Kent-Essex in particular was WAY off: the projection had the Liberals ahead by 13%, and the PCs won it by 9%.  YIKES!

So where did the projection fail?  It wasn't the opinion polls, because even with the exact poll data I would have had a very similar result.  Rather, the failure was in one specific assumption in the model, which is that swings across a region are roughly uniform.  This was, with a couple of exceptions, the case in each of the other regions, however Southwestern Ontario was all over the place on the swingometer.  The PCs did well in some seats, pulling off swings in the 17% region in Chatham-Kent-Essex, whereas if we go a couple of ridings east to Haldiman-Norfolk, the swing was a much more modest 2.5%.  In Kitchener-Waterloo, there was actually a 1.3% swing from the PCs to the Liberals.

I modified the swingometer from last time to show the actual results and to fade all seats other than those in Southwestern Ontario.  The swing in this region came to around 7.3%, but if you looked at how the seats fell, you'd suspect it to be closer to the 9.5% range.  Normally, that would suggest tactical voting going on in a number of ridings, and indeed that appears to have happened in Perth-Wellington.  But the swing in Elgin-Middlesex-London (which is a huge 20%) cannot be explained by tactical voting entirely.  This will require more analysis that I have time for immediately, but when I get around to it, hopefully it will allow me to improve the model.

Other Electoral Systems


As I do after any election I track, I looked at how the results would have panned out under other ways of electing our MPPs.

First Past The Post (current method):
LIB: 53
PC: 37
NDP: 17

Alternate Vote:
LIB: 56
PC: 31
NDP: 20

Preferences are based on a Forum Research poll commissioned by the Ontario Federation of Labour.  Admittedly not the most objective source, but that's the only preference information I could find.  It shows that NDP and Liberal voters are more likely to preference each other, while PC voters are slightly more likely to preference NDP than Liberal.  None of this, though, I consider surprising.

Proportional Representation (separating regions):
LIB: 41
PC: 41
NDP: 25

Proportional Representation (province-wide):
LIB: 41
PC: 38
NDP: 25
GRN: 3

Proportional Representation (province-wide, 5% vote required):
LIB: 42
PC: 40
NDP: 25


Are you happy with the results of Thursday's election?  Do you think the legislature will survive the entire four years?  Comment below!