matthewwhatever
Member
If the polls prove inaccurate again this election then I think it is time for the pollsters to really start looking at their methodology because clearly something is wrong.
I think part of the problem is that polling doesn't work so well in a system like ours. Public opinion polling is great where everyone is voting for the same party or the same candidate. If you're looking at a Westminster-style system, where an "election" is really a whole bunch of elections running concurrently, it doesn't work so well.
Look at that Forum Research poll I posted above as an example. It's from a sample of 801 voters. There are 87 ridings in Alberta, which means that, in the absolute best-case scenario, they around 9 people per riding -- probably less than that in some ridings and more than that in others, but that's the general number. There's no way a polling company can get an accurate read on a riding based on the responses from 9 voters.
Not only that, as far as we know it didn't differentiate between eligible voters and likely voters. Alberta averages just over 50% turnout, so if the poll was just looking at eligibles, they measured a lot of people who aren't planning on voting. Even assuming that people who responded to the poll may be slightly more likely to vote, that still drops the number even further, so you're looking at...what, 5-6 people per riding?
This isn't meant to pick on Forum Research at all -- I imagine it's the same way for all provincial polling in Canada. That doesn't excuse the lousy job pollsters have been doing for the last several years, but I think it explains why there's been so little disconnect between polling numbers and actual ones.