Naheed Nenshi was not supposed to become the mayor of Calgary in 2010. Polls predicted he’d finish a distant third and a Calgary Herald poll, taken a few weeks before the election, gave Nenshi only nine per cent of the popular vote.
“Recent history in polling has not been strong in this country,” he told students in Toronto recently. In fact, Nenshi won the 2010 Calgary election with 40 per cent of the vote.
Similarly, a poll in Ontario found the Liberal and Conservative parties neck-and-neck right before the provincial election earlier this year. On election day the Liberals won a majority. Part of the problem, according to Peter Kim, statistician and reporter for Global News, is that people don’t use their telephones – the key resource for pollsters – as often for verbal conversation.
“Most people I know don’t talk on the phone,” he said. “They use (phones) to email, text and surf. The talking function has become incidental.”
According to Kim, mock votes in schools before Ontario’s provincial election correctly projected a Liberal majority. He thinks pollsters are not properly predicting youth votes.
“Make it easy – think Tinder for political candidates,” Kim said. “Photo comes up. Swipe right for ‘like.’ Swipe left for ‘not like.’”
Ipsos Reid is the third-largest polling company in the world. During elections, the company gathers voter responses for media companies, but they have recently been required to change their methods. Sean Simpson speaks for Ipsos Reid.
“(Selectively sampled groups), panels, are becoming less and less viable,” he said. “It’s more accurate to randomize the online world.”
As younger voters move away from telephone-based communication, the margin for error in coverage bias increases, he said. Ipsos reports landlines in 82 per cent of households, but that number decreases to 67 per cent for younger users. Similarly, older voters are less likely to go online.
Simpson said another challenge is predicting which people will turn out to vote.
“A poll is a snapshot in time,” he said. “It’s (based) on opinion, not who’s going to vote. You measure 100 per cent, and 50 per cent show up.”
One solution may lie in aggregate companies such as Threehundredeight.com. They forecast results based on number of polling firms, Ipsos Reid included. Nancy Reid, professor of statistical science at the U of T, said aggregate polling tends to be accurate.
“If the turnout is quite different than usual, then polls have a lot of trouble capturing that … A single poll can’t, but aggregated polls are sometimes able to.”