Some readers yesterday rightfully pointed out the weird projections in Guelph where the model has the NDP closely winning against the Tories. While I acknowledge the fact it would require a major swing for the NDP to win this riding, I have to mention one technicality about the model.
In Ontario, my model (with all the variables) is able to explain around 70% of the riding-level swing for the CPC and LPC. This is not as good as Quebec but still pretty good (remember, the only actual source of information is the provincial swing). It means I would have made around 8-11 mistakes during the last two elections. However, the NDP is tricky because this party experienced almost no provincial swing. Indeed, the party of Jack Layton moved from 18.1% in 04, to 19.4% in 06 and back to 18.2% in 08. It's very small and not a good source of variation (which is required for the model). I can only explain around 30% of the variation. I'm not trying to find excuses, I'm just explaining clearly where the model works best and where it works less well.
When this is the case, I usually switch to a uniform swing model for the NDP (it is also the case in the Prairies). A linear swing model is les sophisticated but has been proven as providing an overall good approximation of the actual results (this is for instance what Ekos is using). It's more "extrapolation-friendly". Nevertheless, if you do that, the NDP would gain only one seat in Ontario, despite a provincial swing of 5-points (currently; wait two more days and if the trend goes on, it's gonna be 10-points lol).
So I'm left with two choices: I can go safe but have the NDP gaining almost nothing, or I can go with the model and have more gains. Since I believe the NDP will indeed gain more than one riding, I'm taking the chance to miss-identify those ridings. So yes, maybe it's not gonna be Guelph, maybe the 5-points swing will be concentrated in another region or riding, but the overall results for Ontario should be the NDP at 21-23 seats, not 18. It makes more sense to me.
Again, for the final projections I'll carefully look at various models and decide which one is best suited. But for now on, I'll stick to the original model and give Guelph to the NDP. You have to admit that it's not the most unlikely riding to switch! lol