If a federal election were held today, with the Liberals and Conservatives so close in national polling, predicting the outcome in the House of Commons would be little better than a coin flip.
But that doesn't mean we're completely in the dark.
- Read more analysis from Éric Grenier
- Milewski: Polls bring bad news for everyone as MPs return to work
As Canada's political leanings differ from region to region so significantly, it is useful to be able to turn national and regional polling numbers into reasonably accurate seat estimates. With our first-past-the-post system, it is possible for parties that garner the most votes nationally to not win the most seats. There have been multiple examples of that in Canada's political history.
ThreeHundredEight.com's updated seat projection model suggests the Conservatives could do just that if an election were held today, squeaking through with a thin minority government despite trailing the Liberals in the popular vote.
From polls to seat estimates: how it's done
ThreeHundredEight.com has been making seat projections since 2008, and the model used to make these estimates has gone through many tweaks, adjustments, and wholesale changes since then. You can read more about that here.
The model that is now being used to project the 2015 federal election, based on national and regional polling averages, first appeared in a slightly different form in 2011. Since then, over a dozen federal, provincial, and municipal elections have been covered.
Success on correctly projecting the outcome relies on several factors. The first is, of course, the seat projection model itself. Again and again, it has proved its ability to accurately project elections — if the polls are close to the mark. With the correct regional support levels plugged into the model, for instance, the 2011 election would have been pegged to within a handful of seats.
The second factor is the accuracy of the polls themselves. Averaging out the polls and weighing them by their age, sample size, and the track record of the polling firm can reduce the potential for error. But there will always be cases where the voting public does something unexpected. This is where the projection's ranges come in, as they try to estimate how the polls could be off.
The last factor is the most important one: the voter. No model can always predict the behaviour of individuals, or capture every local particularity. But a model like this one can get it close, and give those voters an idea of what the polls are saying might happen on Election Day.
At that point, it is up to them.
ThreeHundredEight.com's vote and seat projection model aggregates all publicly released polls, weighing them by sample size, date, and the polling firm's accuracy record. Upper and lower ranges are based on how polls have performed in other recent elections. The seat projection model makes individual projections for all ridings in the country, based on regional shifts in support since the 2011 election and taking into account other factors such as incumbency. The projections are subject to the margins of error of the opinion polls included in the model, as well as the unpredictable nature of politics at the riding level. The polls included in the model vary in size, date, and method, and have not been individually verified by the CBC. You can read the full methodology here.