In both cases, many were quick to conclude that “the polls got it wrong.” In the British election, this appears to be the case: The polls underestimated the strength of the Conservative Party. But in Canada, this wasn’t the case at all. The polls provided a very accurate forecast.
Instead, the problem was the media, not the polls. Many members of the media misread the polls and misled their audiences. The same could be happening in the United States right now regarding the presidential campaign and, in particular, the candidacy of Donald J. Trump.
One problem is that the media often places far too much emphasis on polls taken weeks (or in the United States, months) in advance of an election. In reality, voters do change their minds at the last minute. Many people are not as interested in politics as commentators who write or talk about politics for a living. People with less interest in politics are also less likely than others to agree to participate in election polls and, if they do participate, they are less likely to state opinions about who they prefer.
But, like students who are not that interested in school, they may decide to “study” at the last minute and then vote. This appears to have been the case in Canada. Throughout the campaign, two of us (Loewen and Rubenson) polled a random sample of Canadians every day. These data suggest many voters remained undecided through the campaign. At the outset, 27 percent of likely voters were undecided. Nearly one-fifth, 18 percent, said they were undecided the day before the election.
Here is a second problem: the media often fail to appreciate “strategic voting.” In both Canada and Britain, four parties with significant funding and campaign organizations contested the election. Yet in each parliamentary district, only the top candidate is awarded a seat in parliament.
Now consider the situation of a voter who comes to realize that their favorite party will not win the seat in their district. If they really dislike one of the stronger parties, they may have an incentive to switch their vote to their second-favorite party. In a tight election, this type of realization and decision is most likely to come in a campaign’s final days.
Again, this appeared to be the case in Canada. In many parts of the country, Canadians who initially leaned toward one left-of-center party (the NDP) appeared to switch their vote to the other left-of-center party (the Liberals) as it became clear that only one left-of-center party had a real chance of defeating Stephen Harper and the Conservatives.
Another factor that can lead pre-election polls to miss the actual result is how votes are allocated across electoral districts. Many organizations report on national poll numbers, but these don’t always provide an accurate sense of what will happen in individual districts.
In both Britain and Canada, the leading party earned over 50 percent of the seats despite getting less than 40 percent of the nationwide vote. So if commentators do not understand how votes are allocated across seats, they can misinterpret even accurate polls.
These problems will be compounded if Web sites aggregating polls don’t fully capture any late shifts. On the day of the Canadian election, Loewen and Rubenson collected (pdf) every pollster’s vote estimate, every aggregation, and every seat forecast. Pollsters who captured changes in the final days predicted the outcome more accurately than poll aggregators or the seat forecasters.
What does this have to do with Trump? A great deal of attention is being paid to his standing in the polls. But we are still months from the first caucuses and primaries, and there is great potential for strategic voting because there are 15 Republican candidates running.
In other words, all of the factors that led members of the British and Canadian media to misread the polls are present in America right now. This is not to say that Trump’s support is illusory, but it is to say that it is important for commentators to be very cautious in placing much weight on today’s polls.
Peter J. Loewen is a political scientist at the University of Toronto.
Arthur Lupia is a political scientist at the University of Michigan and author of “Uninformed: Why Voters Know So Little about Politics and What We Can Do about It.”
Daniel Rubenson is a political scientist at Ryerson University.