Digging deeper into voters' emotions will help prevent polling misses
Donald Trump’s shocking win last Tuesday brought back some memories in Alberta of Alison Redford’s surprise victory in 2012. The polls were wrong again, just like the Brexit vote and British Columbia’s last provincial election. So is polling broken? What’s going on?
There are three reasons election polls miss the mark: timing, shy voters and turnout.
In the week leading up to Alberta’s election in the spring of 2012, Danielle Smith’s Wildrose Party had a lead of seven to 10 points over Redford’s Progressive Conservatives. But when the votes were counted the PC’s had won by 10 points. The reason the polls were wrong is that people changed their minds in the last couple of days.
A survey after the election found that 36 per cent of Albertans made up their mind in the last three days, and those who did broke decisively to the PCs. This was backed up by the one poll that was conducted the last day before the election and it found the Wildrose lead had dropped to only two per cent. The reason why the polls missed in 2012 was because of timing — they had stopped collecting data too soon.
Another reason polls are wrong is “shy voters” — people who are willing to take a poll but are not willing to admit to voting for an unfashionable candidate or cause. This is rarely more than one or two per cent of the population, but in a close race that can make all the difference.
This effect can be really important in elections with a populist versus elite opinion dynamic. Trump’s election is a perfect example. With the media and elites from both parties criticizing him and saying his supporters were sexist or racist, is it any wonder that some of his supporters didn’t want to admit who they were voting for, even in an anonymous poll?
The most common and important reason that polls are off is that they misjudge how likely people are to actually vote. People who answer polls, either on the phone or online, don’t like to admit they aren’t likely to vote. You are supposed to vote, so they say they will.
In fact, in election polls I have conducted when respondents are asked how likely they are to vote on a scale from 0 to 10 — typically more than 80 per cent say “10” for “absolutely certain to vote.” Actual turnout in most elections in Canada is far lower — usually closer to 60 per cent.
To get around this pollsters build complex models where they try to predict how likely someone is to actually cast a ballot based not just on what they say, but on their age, gender, education and income.
These clearly failed in the U.S. last week, especially in the rust belt states that gave Trump victory. In Wisconsin, Michigan and Pennsylvania, the polls had Clinton ahead by anywhere from five to seven points. But Trump won them all.
The difference, which the polls didn’t take into account, was the level of enthusiasm that Trump voters had over Clinton voters. They were simply more motivated to go and vote in a way that could not be predicted by pollsters asking their simple likelihood of voting or using demographic models.
The solution is that pollsters need to look out for situations where problems like “shy voters” and turnout differences between groups are likely to occur and take steps to understand what is happening.
To better predict turnout, more questions should be asked, not just about the likelihood of voting, but the level of enthusiasm they feel for their candidate, or how passionate they are about the issues. Even identifying which emotions they associate with voting can be useful — people motivated by anger are a lot more likely to vote than ones motivated by respect or admiration.
Pollsters can and will develop new ways to measure turnout and ferret out shy voters. Polling misses, like last week, do not need to keep happening.
Hamish I. Marshall is president of Torch, a Canadian polling and advertising agency.