While the latest twist in Hillary Clinton’s email controversy may have shaken her support base, Democratic voters have been feeling rather confident of late. But lessons from the other side of the pond suggest that polls are no reason for complacency. Pollsters in the UK who were surprised by the outcomes of the 2015 General Election and Brexit referendum say the same issues could distort predictions in the US presidential election.
First, a recap of what went wrong in the UK. Britain choosing to leave the European Union sent shockwaves around the world. Betting markets, politicians, and pundits were confident that Remain would win. Though the race was narrow and polling analysts collectively suggested it was too close to call, four of the six surveys released the day before the vote showed a win for Remain. But when the results came in, the Leave vote tipped ahead.
The results were enough of a surprise that British Polling Council (BPC) president John Curtice is holding a conference in December to investigate how and why the polls went wrong. And the misreading was all the more striking following on from the UK’s May 2015 General Election, where the polls were wildly off and an inquiry into the inaccuracies found “systemic” errors in survey methods.
Could the polls be wrong in the US? Clinton’s lead has narrowed to just 4.4%, while 15% of voters are currently undecided (compared to 5% at this stage in 2012.) A surprise Trump win looks unlikely, but it’s certainly not impossible.
Republican nominee Donald Trump recently suggested that many of his fans are keeping their support secret, so the polls aren’t an accurate reflection of his chances. The notion of “shy voters” was first described by the 20th century German political scientist Elisabeth Noelle-Neumann, who argued that voters who thought their choice was unpopular would be more likely to fall into a “spiral of silence” and keep quiet about their voting choice. “If you perceive that a particular voting choice is unfashionable, perhaps socially unacceptable, you may be reluctant to declare your views,” says Curtice.
Jonathan Mellon, sociology research fellow at Oxford University with a focus on polling data, says there’s some evidence that Republicans are less likely to respond to pollsters after a bad news day for their party, which fits in with “spiral of silence” theory. But, in fact, “shy voters” were not a significant source of error in the 2015 General Election or Referendum vote. In general, Trump voters seem proud of their candidate, and there’s no significant evidence that there are hordes of secret Trump fans keeping quiet.
But there are bigger potential sources of error at play. BPC president Curtice explains that in order to get a representative sample, pollsters must rely on knowledge from past elections about how much each demographic tends to vote. This was the key to errors in polling for the 2015 UK General Election, which ultimately overestimated the number of young people who would cast their vote, and was also a likely source of misunderstanding in predicting the Brexit vote.
“The question of whether different people may turn out to vote is a much trickier issue and one that has affected polls in a lot of countries in the past,” says Mellon. “If that question becomes miscalibrated, then that can certainly affect the results because you’re systematically excluding people at a higher rate than you should within certain groups, or you’re under-excluding them.”
The current US election is quite different from those in previous years, given the polarizing effect of both Clinton and Trump, and Trump’s disconnect from his own party. So there’s potential for pollsters to misjudge how each demographic is likely to vote.
“We’re talking about an election where the demographics of the vote is somewhat different from what you’d normally expect at a Democratic-Republican contest,” explains Curtice. “Given that the demographics of the Trump campaign has some similarities with the demographics of the Leave campaign, that’s a more interesting question [than “shy voters”]. Will it be the case that older white working class voters will come out in higher numbers than the polls anticipate because these folk are energized by Trump?” Meanwhile, “the Democrats are getting the suburban, white, college-educated vote who you’d normally expect to lead towards the Republicans on average.”
There’s also the question of how traditional Republican supporters are likely to vote. In general, most voters tend to stick with the same party. “The thing above all that best predicts how you’ll vote in the next election is how you voted in the last one,” says Curtice.
Michael Hanmer, politics professor at the University of Maryland says it will be “interesting” to see how Republicans behave in this election. He notes that 90% of those who strongly identify as Democrat say they’ll vote for Clinton, compared to just 80% of those who strongly identify as Republican and currently say they’ll vote for Trump.
“I think those who say they’re undecided now are genuinely conflicted,” he says. But when it comes to Election Day, they may skip the vote, go for a third party, or, “when they’re staring at the ballot, select the Republican ticket.”
It’s also extremely difficult to judge the level of turnout. Hanmer expects many Americans to stay at home—some because they dislike the candidates; others because they have a false sense of complacency from the polls. “It’s a tough situation to predict,” he adds. Most polls are predicting results based on “likely voters”, but “we don’t have a single agreed-upon definition of what that is in political science or among the media,” he says. Hanmer adds:
“Trying to get your head around who’s going to show up is particularly challenging this time. Voter models, like all models, are imperfect in the most predictable circumstances and this is probably one of the least predictable elections we’ve seen.”
US polls do have more diverse methodology than those in the UK, and it’s unlikely that the pollsters will be drastically wrong. “Even if you’re really skeptical about the polls, at best Trump is hanging in there by his fingernails,” says Curtice. Yet all the polling experts I spoke to strongly warned against certainty.
Polls are “a reasonable best guess of what will happen,” says Mellon. It would take a “convergence of several forces of bad luck” for them to be wildly wrong, but an inaccurate prediction is certainly possible. After all, the candidates in this election are quite unlike those in previous years. “Because there are so many possible issues,” he says, “if they all converge in the same direction on Election Day, it could be enough to push it over.”