I just want to point out a few things quickly, and I'll go more in-depth later on tonight.
First off, while the projection failed, it didn't fail too spectacularly - it got 16 ridings wrong, five of them which went Liberal, nine which went PC, and two which went NDP. Otherwise, it made the correct call in 91 ridings, with varying degrees of accuracy on the margins. The full list is posted below this post.
The reasons for this sum up my second point: last night was a case that no projection system could have predicted without major tweaking of its regionals. The fact is, the Liberals rose in urban areas, but dropped like a stone in rural areas; most of our rural ridings were wiped out. But we made boosted our urban ridings against the Tory and Dipper tide remarkably well, especially within the GTA.
What does this mean? It means that the election fell upon the urban-rural divide more heavily than usual, and I thought that would happen when I started noticing polls trending. Problem is, I have zero way of imputing that kind of trend into my projection, minus doing it all by hand.
So, you'll notice right away where my projections went wrong - the rural ridings shifted heavily to the Tories, and these are the ones I attributed as giving McGuinty his majority numbers. I got five urban ridings wrong, and those were special cases of local influences and star candidates; the other 11 were all rural ridings. It's interesting to notice the shifts for sure, and how localized they ended up being.
Thirdly, were there ever some wild swings last night in particular ridings. Essex, Bramalea-Gore-Malton, Ottawa Centre, and the two Thunder Bays, and many others, were insane bucked-against-the-trend ridings that it's made them impossible to predict without messing around, and for B-G-M I did mess around, and pumped up the NDP 10% over what my base projection said; it still didn't give them an advantage.
Fourthly - we should all be very afraid of last night's results. Very afraid. I'll explain why later.