WHAT CAN THE POLLS TELL YOU?

PERFECT PREDICTION IS IMPOSSIBLE, EXPERT SAYS, BUT RACE WILL BE CLOSE

If there’s one consistency about the election campaign over the past six weeks, it is that the Conservatives and the Liberals have barely budged in the polls. Unlike past elections, there hasn’t even been much of a discrepancy between polls from individual pollsters like Abacus Data, Angus Reid and Nanos Research, which all show the popular vote for the Conservatives and Liberals hovering between 29 and 35 per cent, each, at best.

Thus, the consensus is a minority government will most probably be the result on Monday. As of Thursday evening, the CBC’s poll tracker run by Eric Grenier, which aggregates all national polls on a daily basis, predicted that the Liberals have a 46 per cent chance of winning most seats, but not forming a majority, and the Conservatives have a 42 per cent chance of doing the same.

But how much should Canadians trust the polls before they cast their ballots? For those who intend to vote strategically, should you base your vote on the latest polling data? The National Post interviewed Dr. Faron Ellis, an expert on polling and public opinion research at Lethbridge College’s Centre for Applied Research, Innovation and Entrepreneurship to figure out just how accurate election polling really is.

(This interview has been condensed for length)

NP I just want to start with what methods most pollsters use to come up with their predictions. So for instance, Angus Reid’s latest poll shows that the Conservatives are at 33 per cent. How did they get that figure?

Dr. Ellis There are a variety of methods. What is still considered the gold standard is live interviews and telephone calls. But of course, a lot of people don’t have land lines. So some polling companies use random digit dialing, which means the computer randomly dialing a combination of numbers. The problem with this is you’re catching people at a bad time — they are at a Tim Hortons’ drive-thru, for example, they might not want to talk. The other thing to keep in mind is that a lot of people are going to be less straightforward on the phone and more honest online.

NP So internet polling is better?

Dr. Ellis Internet polling involves a great deal of work for pollsters but can be very accurate. What you have to do there is generate a panel of say five or 10 or 20,000 people who agree to respond to constant surveys sent by the polling company. It is quite a mammoth task to keep that panel representative of demographics. Let’s say I’m too busy, and I don’t respond to the surveys, they will kick me out. They will then have to find another middle-aged white man to replace me. But when done properly, internet polls are a good approximation.

NP All right so this election. Pollsters are pretty aligned on popular vote predictions (30-something per cent each for Conservatives and Liberals), but how does that translate into seat projections?

Dr. Ellis It does not. A more sophisticated model is needed to project seats. Eric Grenier at CBC uses a model that includes more than just polling numbers to project a seat. It includes … the history of the riding, how people have voted in the past. For example: Trudeau senior and the Liberals won three to four per cent more of the popular vote than Joe Clark and the Conservatives in 1979. But

Joe Clark and the Conservatives won more seats.

NP Wait, but what’s the difference between an individual national poll like Nik Nanos, and Eric Grenier’s poll tracker or Philippe J. Fournier’s 338Canada.com predictions?

Dr. Ellis What Fournier and Grenier do is they take whatever public data from the polling companies that is available — Angus Reid or Ipsos or Nanos — they throw them into their model and they weight each poll. It’s an aggregate poll. So for instance, Grenier has weighted the most recent CTV/Nanos poll at 25 per cent. Mainstreet gets a lower weight because he doesn’t like the interactive voice polling method. The older the poll, the less weight it has.

NP Would you say aggregate polling is more accurate as opposed to looking at data from individual polling companies?

Dr. Ellis That’s hard to say. In an election where one party is ahead by five or eight per cent, individual polls are very reliable. In a close election, you have to take a look at what the aggregate polls are saying about seat range. I think it is very reliable to say that the Conservatives, for instance will get between 84 and 172 seats. If you want something that predicts exactly how many seats a party will get, no model can do that.

NP If I wanted to vote strategically, how much should I rely on polls to figure out how to cast my vote in my riding?

Dr. Ellis My advice to Canadians is use the national polls to get the overall national picture. Don’t use any national averages to help you cast your vote locally. Strategic voters need to ask themselves: Am I in a riding where it is going to be close enough that a strategic vote will even matter? For example, in a vast number of Alberta ridings, you can vote as strategically as you want but the Conservatives are going to win 50 to 80 per cent of the vote so it doesn’t matter. But if you are in a riding where there are local polls and you see the party leader being there four times, then sure, you might think of voting strategically. NP I have to ask you … what on earth happened in the 2016 U.S. presidential elections? There’s this rhetoric that the pollsters got it all wrong.

Dr. Ellis Where the pollsters got themselves into trouble was in not being more cautious in telling journalists how to use their polls. The pollsters were right! They predicted, using national polls, that Hillary Clinton would get 1-3 million more popular votes. That is what happened. In a close election, national polls cannot account for the 500+ electoral college votes and how they are going to be distributed. All that we should be taking away from our national polls right now is that Monday will be a very close race.