Tuesday, April 24, 2012

Alberta election aftermath

That was a surprising result. But in the context of how uncomfortable I felt when posting my final projection late on Sunday, I found myself unsurprised last night that something so astonishing occurred. As my wide ranges suggested, this sort of outcome was plausible. But even those ranges did not manage to contain the wild swing that appears to have taken place in the last 48 hours of the campaign.

Wildrose's support simply cratered, and to an extent that no model or method could have anticipated. The Tories picked up the majority of that lost support, but also took some support from the New Democrats and Liberals.

For an in-depth look at what happened with the polls, check out my article for The Globe and Mail here. For a perspective on what this means for Alberta going forward, you can read my article for The Huffington Post Canada here.

Did the polls call this election wrongly? That depends. If there was the kind of swing I talk about in my Globe article, it is difficult to fault the pollsters for failing to identify a radical swing in the last two days of the campaign considering they all (except Forum) did their final numbers sometime in the last week. If they had all opted to poll on the Sunday, would they have all identified that last minute swing? The odds that they could have all been so wrong so spectacularly are quite remote.

As my model is limited by the data I have available to put into it, it can't really be faulted for being so far off. Compared to what every poll said and what some other seat projectors forecast, my projection was the only one that did not forecast a Wildrose majority and voices (like mine) that put into question the likelihood of the expected Wildrose landslide were few.
But even my uncertainty model could not have predicted such a Wildrose collapse. At the very least, however, the results demonstrated that maintaining (and perhaps, going forward, even emphasizing) this sort of uncertainty modelling was the right thing to do.

In the end, the New Democrats did almost exactly as expected, the four seats they won having been projected to go their way. The Liberals won two of the three seats they were projected as being in the running for in Edmonton, but completely over-achieved in Calgary. It wasn't that their vote spiked in the city (they did exactly as well as expected), it was simply that their vote was super-efficient.

The Tories won all of the swing seats they needed to, and some that weren't even seen as being potential wins for the PCs. But their result of 61 seats was just inside the upper extreme of the projected ranges, a remarkable result considering that the Liberals also over-achieved.

Wildrose, however, simply did not deliver. That they could have won as few as 22 seats might have seemed ridiculous going in to last night's vote. Instead, they won 17! Not only did they under-achieve their polling by a significant degree, they under-achieved their worst case scenario.

Once again, and painfully so, the seat projection model would have churned out good results with the correct vote share being plugged into it. With the real regional results, the seat projection would have given the Progressive Conservatives 65 seats (24 in Edmonton, 19 in Calgary, 22 in the rest of Alberta), Wildrose 18 seats (1/8/9), and the New Democrats four seats (all in Edmonton, of course). While this is too late to do any good, it shows that seat projection models can work, even in strange elections like this one (i.e., Wildrose going from 7% to 34% in one cycle).

Interestingly, it still would have delivered zero seats for the Liberals. For a party that sank from over 26% to under 10%, it is remarkable that some of their incumbents were still re-elected with 40% of the vote or more. The level of attachment that voters (in Calgary, particularly) had to their Liberal MLAs was surprisingly strong, and something that would have been difficult to model.

Something that was included in the model - the assumption that the Progressive Conservatives, as the government, would be under-estimated in polls and that Wildrose, as the third party in the legislature, would be over-estimated - did turn out to be prescient. But it would have required a ludicrous amount of adjustment to have come up with this sort of result.

The Alberta electorate did a number on pollsters and myself, so kudos to them for demonstrating why democracy can be an unpredictable but important thing.

A statistical breakdown of how the projection model performed is available here.