Polling and by-elections

Forum Research has been regularly polling all the federal by-elections in the last few years but how seriously should we take this polling?    The big difference between a general election and a by-election is voter turn-out.

The polling does not seem to capture this reduction in voter turn-out which I think means the majority of people answering the poll are not likely to vote.    The other option is that people who are unlikely to vote are not answering the poll.   Either one means that statistically the sample used in the poll is flawed and may not be much of a good reflection for what will happen on election.   The sample Forum used may be a realistic reflection of the public but it could also be far off.    Their current polling trends in the four by-elections are very stable so may be the sample is a good reflection of reality but I could not answer why the sample is correct and how one would duplicate it.

The reason I raise all this is because three of four by-elections Forum has polled have results that are broadly expected based on past electoral performance but one is out of sync.  Forum has found the Liberals leading in Brandon Souris by a measurable margin in each of the three polls they have done.  Not only are the Liberals leading, they are polling at levels in the riding significantly higher than they have ever achieved in past elections.    I can not tell how the voter turnout in this by-election will impact the results.

When we look at the voter turnout of the last 60 federal by-elections we get an average of 37.16% turnout Expressed in terms of the previous general election, 62.47% the turnout in the by-election compared to the last general election.    If the race in the by-election is close, the turnout trends higher but still significantly lower than the previous general election.   Realistically Brandon Souris will see between 24,500 and 28,000 voters instead of the 35,095 voters in 2011.

What the data says is that we should expect the turnout in the current federal by-elections to be down, this means a lot of people that voted in 2011 will not be voting in the by-elections, in Brandon Souris it means between 7,000 and 10,500 fewer voters .  Who are these non-voters?

Data sources I can access indicate that age, education and wealth are the best predictors of who is most likely to vote or not.   A 65 year old university educated home owner is much more likely than a 30 year old waiter.  It means that the population that votes is not    Older people seem to be still treat voting as a societal norm they should do.

As it stands, Forum polling tends to skew very strongly towards older voters, David Coletto of Abacus Data has some issues with this tendency and what it might mean.  The latest poll they have done in Brandon Souris had 68.7% of their respondents aged 55 or older which is not demographically representative of the general public but might be a reflection of the actual voters.

If I could understand who the non-voters are and how they are captured in the polls, or not, and why the polling consistently skews old, I would have more trust that the data from the polls are a reflection of the likely outcome in the by-election in Brandon Souris.  As it stands, I take the poll results with a grain of salt.