The Alberta polls last time? They weren’t that wrong.
A piece of conventional wisdom in Alberta politics is that you can’t trust the polls. Just look at 2012: of the 22 publicly released surveys, all 22 showed the Wildrose leading the PCs, sometimes by as much as 17 per cent. Yet, come election day, the PCs won by nearly 10 per cent. So, the polls are totally useless, right?
No. Not really. (Honestly: it’s in the title, and titles are sometimes not misleading.)
How could that be? Surely, polls were just skewed! Perhaps opposition party supporters were super angry, as they are, and just quivering at the chance to answer a poll, while PC supporters, being a pretty satisfied bunch, couldn’t be bothered to answer the phone, preferring to busy themselves with something more important.
Still no. Still in the title.
Instead, the problem seems to be that we ignored undecided voters last time. In coverage and discussion of the polls, we simply discarded them and talked just about the intentions of decided voters. Basically, we assumed that undecided voters would split just as decided ones did. The problem: this was completely wrong.
Looking at the Comparative Provincial Election Project (CPEP) post-election survey of the election, people who decided in the last few days of the campaign split 58 per cent PC, 15 per cent Wildrose, 12 per cent NDP and 11 per cent Liberal. This is clearly quite a different spread from those Albertans who decided earlier.
So, let’s revisit these apparently awful polls. More to the point, and since there is only so much time in the world, let’s revisit the last poll released during the campaign. Forum [PDF] put out a poll taken on April 22, the day before the election, with the following proportion of decided voters: Wildrose 38 per cent, PCs 36 per cent, NDP 12 per cent, and Liberals 10 per cent. They even predicted a small Wildrose majority as a result of this. Pretty hilariously off, right?
Still not. (Notice how the title still says that it’s not. It honestly doesn’t change.)
What was the proportion of undecided voters in this survey? You have to dig to figure that out. It seems to be 23.3 per cent. How did I calculate that? The sample size was 1,949, and the sample size of decided voters was noted at 1,493. I then let our good friend division do the rest.
So, if we discount the 23.3 from the figures reported in the poll, it should read: Wildrose 29.1, PCs 27.5, NDP 9.2 and Liberals 7.7. Then, let’s divide up the undecided voters using the data from the CPEP survey. Of the 23.3 per cent, the PCs would get 13.6, the Wildrose 3.5, the Liberals 2.8 and NDP 2.6.
Now, let’s bid farewell to our friend division to make room for our pal addition, and then add the decided voters to the undecided ones. Doing this, we get the PCs at 41.1 per cent, the Wildrose at 32.6 per cent, the NDP at 11.8 per cent and the Liberals at 10.5 per cent. Here would be a good chance to remind ourselves of what the result was last time: the PCs won 43.9 per cent, the Wildrose 34.3 per cent, the NDP 9.9 per cent, and the Liberals 9.8 per cent. (You can do the same trick with other polls too — it works out to be about the same.)
Not so wildly useless, then. If there’s a lesson to be had in all this, it’s that the polls in Alberta might not be so wrong if we actually pay attention to the undecided voters. In fact, I’ll end with that repeated again but in a really big font:
