How did polls perform for the 2015 Alberta election?

First of all, the good news: the polls did much better this time around than 3 years ago when they were all predicting a Wildrose win and the PC ultimately won (easily). Polls this year were predicting the NDP to finish way ahead and this is exactly what happened.

The table below show you all the latest polls. For each, I calculated the average absolute deviation (so if a poll had the NDP at 44% and this party got 40%, the error is 4) as well as how many parties were within the margins of error. This last part is obviously subject to sample sizes (and I used the same margins no matter if it was an online poll or not. I don't feel like having this debate all over again).



Forum and Ekos released each two polls in the last week. I included both even though I really should only count their last one.

As I've said above, polls in general did well and very much in line with the Quebec and Ontario elections of last year. However, not everything is perfect. The polls in particular clearly underestimated the PC again. The margin is less than in 2012, but we are still talking a good 4 points in average. Only one firm, Leger, had the PC above its actual result. Still about the PC, most polling firms had this party finishing third, with a score sometime as low as 21%. Only two polls had the incumbent finishing 2nd ahead of the Wildrose. Make no mistake, this is a major miss. People will be quick to forget because the orange wave actually happened, but a miss is a miss. I wouldn't say that yesterday was that great of a showing for pollsters, it was an acceptable one. They had the PC 4 points too low (so outside the margins of error) and the NDP too high as well.

It's especially disappointing because it was predictable. The PC is underestimated by polls systematically in Alberta, election after election. And sure pollsters will use excuses such as "people changed their mind" or "the PC has a good machine to get the vote out", but I personally find them a little bit weak. My adjusted average was spot on and as usual outperformed the polls (which is expected, an average should do better. It's just that this time, my average did really better). And I indeed had the PC second. How did I do that? By making very simple adjustments. In particular, I simply allocated undecided differently. Instead of distributing them proportionally (like all the firms do; Which is stupid because it means we are assuming undecided are similar to decided voters), I only allocated them to the main three parties. And among those three, I gave half to the PC and 25% each to the Wildrose and NDP. Yes this is a guess, but it's an educated guess. I do the same in Quebec with the Liberals. I also adjusted for the fact that the Liberals and the Alberta Party were not running in every riding. There again, why is it that professional pollsters don't do that? In the last BC election, Angus-Reid was accounted for that and was only offering respondents to choose among the parties running in their ridings. Unfortunately, A-R didn't poll this time around.

The winner is easily Leger. Not only do they have the smaller error in average, they also had the correct order with the PC finishing second. They only had 3 parties within the MoE and therefore missed on the two smaller parties, the Liberals and the Alberta Party. To be fair, they had the AP really low and relatively close, but MoE are really small for a party polled at 1%. At least they had all three major parties right and they were the only one (since everybody else missed on the PC)! So congratulations Leger!

As the seats projections, yesterday was a very good night for me. I had 12 PC, 19 Wildrose, 3 Liberals and 53 NDP. It means that I was only 1 seat off in average. For me, this is very good news before moving on to the next federal election.