Saturday, November 19, 2011

Blunt Object's Projections: Did I Do Well, At All?

This is a quick post, modeled somewhat after how Eric Grenier does it, for each of my projections this past provincial election season. First up, Ontario:

Ontario
Projection - 58 Lib, 29 PC, 20 NDP (38.5% Lib, 33.9% PC, 23.5% NDP, 3.0% Grn)
Actual - 53 Lib, 37 PC, 17 NDP (37.6% Lib, 35.4% PC, 22.7% NDP, 2.9% Grn)
Actual w/ Projection - 56 Lib, 31 PC, 20 NDP

Well, clearly it wasn't perfect, but including my ranges, it was pretty good. Even with the actual results inputted into the projection system, it still gave the Liberals a majority - but a difference of 3 seats for the governing party, and 6 seats for the main Opposition, I'm not too worried. And except for the PCs who were underestimated a tad by polls, the popular vote was kept within 1%. I consider that fairly impressive.

Next up, Newfoundland and Labrador:
Projection - 39 PC, 7 NDP, 2 Lib (54.0% PC, 29.6% NDP, 16.4% Lib)
Actual - 37 PC, 6 Lib, 5 NDP (56.1% PC, 24.6% NDP, 19.1% Lib)
Actual w/ Projection - 39 PC, 5 Lib, 4 NDP

This would have worked out really well, if not for most polls overestimating the NDP by upwards up 5-10%, and underestimating the Liberals by 5%. Had I the actual results, 39-5-4 is pretty close. Not the same ridings, however - I missed Humber Valley and Torngat Mountains for the Liberals, though both were very close, alongside some other mistakes - but the seat totals were close enough. The polls, however, had major issues.

Now on the stage, Manitoba:

Projection - 32 NDP, 24 PC, 1 Lib (45.9% PC, 42.6% NDP, 8.9% Lib, 2.3% Grn)
Actual - 37 NDP, 19 PC, 1 Lib (46% NDP, 43.9% PC, 7.5% Lib, 2.5% Grn)
Actual w/ Projection: 35 NDP, 21 PC, 1 Lib

My problem here was that I didn't let more recent polls in Manitoba carry better weight - so the old polls that showed the PCs with better leads meant the PCs ended up with a higher popular vote, and a higher seat count. However, as the actual results inputted through the projection show, had I gotten it closer to the mark I'd only challenge the NDP on 2 seats - St. Norbert and Southdale - which were close anyways.

Saskatchewan is last, but not least:

Projection: 44 Sask, 14 NDP (62.7% Sask, 32.6% NDP, 2.5% Grn, 1.7% Lib)
Actual - 49 Sask, 9 NDP (64.2% Sask, 32.0% NDP, 2.9% Grn, 0.6% Lib)
Actual w/ Projection - 44 Sask, 14 NDP

Well, I wasn't very good with this, missing the mark on 5 seats that were, frankly, all thought to be NDP strongholds. The popular vote was OK - the Liberals were over-estimated but not by a hilarious amount - so it was more just the previously-thought invulnerability of those five seats, which include Regina-Douglas Park, Saskatoon-Fairview, Regina-Coronation Park, Regina-Walsh Acres, and Prince Albert Northcote.

Overall, I'm happy with the results. A few tweaks here and there, and I'll get even better results. The worst practice was, obviously, the poor weighting on Manitoba; best was the ranges for Ontario. In the future I'll have to keep up on that, and my rolling average projection for Canada (which is currently 140-110-57-1, thanks for asking) does include better weighting on my part, and I'll make up ranges too.

No comments:

Post a Comment