Making seat projections when a surge is occuring

As a polls and political junkie, I'm of course very excited by the recent turn of events we've seen during this campaign. However, I have to admit that making seat projections for this campaign will be difficult. First of all, polls give us a snapshot of the situation, but with 2-5 days of delay. This delay can clearly be seen by comparing polls from Ekos to Nanos for instance. Because Nanos uses a rolling panel, it takes longer to see the surge occuring. For projections, by using recent polls, there is the risk of being behind or outdated already. It is especially the case in Quebec where each new poll seems to give 1 more point to the NDP and one less to the Bloc. Right now, by simply updating the list of polls I'm using to calculate the average, I'm basically in a situation where the NDP gains 2-3 seats every day in Quebec.

So let's try to only use the four most recent polls: last one from Nanos, last two from Ekos and the recently released Angus-Reid. Ekos and AR have in common to place the NDP way ahead in Quebec and strong second nationally. They all place the NDP first in Quebec though. These four polls provide an overall sample size of above 8000, which is quite good. Here are the results:
And if it wasn't for Nanos who has the NDP lower in Ontario, the NDP could even reach above 90 seats. Notice how this scenario would lead to another incredibly unstable parliament where LPC+NPD would fall just short of a majority (by one seat actually if you elect the Speaker from the other party).

Ekos and AR also show a NDP surge outside of Quebec. In Ontario for instance, it's very interesting as the NDP is currently taking votes from both the Tories and the Grits. So again, NDP's surge doesn't mean Conservatives majority, not at all. We'll have to wait and see if it's confirmed elsewhere. But it seems more and more likely that the Liberals could well finish third, especially since they starting dipping below 30% in Ontario. For the Tories, if they can keep their support in this province around 38-39%, the NDP might split some seats for them. However, this is true only if the NDP grabs 2-3 points from the Grits and reach 21-22%. As soon as they go above that, every point switched from the LPC to the NDP leads to stable seat projections. For instance, try to input AR's numbers only and you'll see that the Tories would lose a lot of seats in Ontario.
So again, I said it many times on this blog but I'm not a big believer of the vote splitting. Those who say that are forgetting how many ridings are a race between CPC and NPD.

But you can see how difficult making projections can be right now. Simply compare these projections to the ones published yesterday, using other polls as well. If you believe the surge is mostly over, then by simply adding new polls this week, we should be ok. But if it isn't, we'll likely have the wrong percentages and therefore the wrong seat projections. On top of that, the swings are so large and weird that they push the model to its limit. The fact is that we never saw the NDP around 28% nationally or above 30% in Quebec. So my model, based heavily on estimation, might be in trouble lol rest assure though that for my final projections, I'll first compare my model to other ones. I have many excel files with many types of models (uniform, proportional, etc). So I can always see if one model predicts something very different and look into the details. But right now, I think the biggest challenge isn't that much to convert percentages into seats, but rather to have an idea of what the percentages will be. Hopefully things will settle down soon and people will spend the week-end watching the Royal wedding and not changing their minds!

By the way, I'd encourage everyone to publish, (by comments) their own projections (using the simulator under the tab "make your own projections"), based on the percentages you think will happen on May 2nd. That would make an interesting thread.