Over the next few weeks, Canadians will encounter a dizzying number of poll results predicting how the Oct. 14 federal election is likely to play out.
Problem is, most of the numbers will relate to popular vote rather than seats in the House of Commons, the only tally that really matters.
And when the polling companies try to extrapolate seat shares from their data, well, it turns out the masses may do a better job.
“All the big pollsters produced seat projections that were completely off the mark,” said UBC professor Werner Antweiler of the 2006 federal election.
The Strategic Counsel’s absolute error – the total difference between their prediction and the actual outcome for all the parties – was 94 seats, EKOS’s was 100 and Ipsos-Reid’s was a whopping 114 out 308.
In contrast, the UBC Election Stock Market, which Antweiler co-directs, was only off by 15 seats.
Whereas polls ask individuals for their voting intentions, Antweiler says the UBC project turns every participant into an expert as they buy and sell “shares” on four markets including one where they predict seat totals.
Traders set prices by deciding how much they are willing to pay or receive for different stocks.
Antweiler puts the market’s predictive power down to “the wisdom of crowds,” saying prices reflect a balancing of diverse opinions.
“A lot of people put the information together and the market is basically the channel through which this information is aggregated,” he said.
The project first started during the 1993 federal campaign and although it produces no revenue for UBC, participants use real money.
“Each individual trader has a monetary incentive to get it right,” according to Antweiler. “They’re going to put their personal beliefs aside when it comes to predicting the outcome.”
As well as the Election Stock Market performed in 2006, the Election Prediction Project did even better, only missing the final outcome by 12 seats. Milton Chan, a securities lawyer in Toronto who started the project while studying at the University of Waterloo in 1999, describes it as a “gossip poll.”
He and a small evaluation panel read about local races in emails from the public and then make riding-by-riding projections, a process Chan calls “absolutely, completely unscientific.”
“I’m not saying there’s any magic to it,” he said. “Anyone can do it. Anyone who has the patience to comb a gazillion emails can come up with the same conclusion, I think.”
Chan expects the current flow of 400 daily emails to rise and the number of “too close to call” ridings to dwindle and ultimately disappear as the election date approaches.
“The success of such a project all depends on the quality of the input,” according to Chan who points to the lack of francophone contributors for lower-grade predictions in Quebec.
For all of Chan’s modesty, Antweiler remains impressed with the unorthodox project’s 2006 results.
“Between Milton Chan and us, you probably would have gotten the best predictions anywhere.”