With 16 months remaining until the 2020 presidential election and surveys weighing the favorites in a large and field of candidates released each week, some remain wary of the poll results in the aftermath of the 2016 cycle.
Despite the public scrutiny of surveys and pollsters following President Donald Trump’s victory and the recommendations for altering approaches to polling that emerged in the election’s wake, national reports show that most haven’t changed their operations or aren’t planning to this time around.
And in one major area — weighting responses by education — two pollsters in a pair of key Midwestern states aren’t straying from their competing 2016 practices either.
Overall, national polls were found to be generally correct during the 2016 presidential race, while state-level surveys showed a competitive contest. Trump’s support was under-estimated in the Upper Midwest, according to an election postmortem from the American Association for Public Opinion Research.
One contributing factor, the May 2017 report found, was that many state-level pollsters failed to correct for education levels among respondents. That approach would have better represented voters without a college degree, who tended to break for Trump.
But in Wisconsin’s very own Marquette University Law School poll, director Charles Franklin has always weighted his data by education and continues to do so. And yet his surveys in the run-up to the election found Democratic candidate Hillary Clinton leading Trump among likely voters.
“It’s not a magic fix,” Franklin said in a recent interview.
Polls already tend to overrepresent those with higher education levels, as they’re the individuals who are more likely to take surveys, University of Wisconsin-Madison professor Barry Burden said.
Prior to 2016, he added, that wasn’t much of a problem. But the division that placed college graduates in Clinton’s camp and those with lower education levels in Trump’s was “the surprise piece” that largely caught pollsters and the public off-guard.
Just as Franklin is continuing to weight by education in his surveys, in Michigan, one pollster said his firm is planning to embrace its longstanding practice of not doing so.
Bernie Porn, a pollster at EPIC-MRA in Lansing, which has a host of media clients including the Detroit Free Press, said while the firm has weighted for age or partisanship in the past, it doesn’t do so for education.
“I’m aware of the fact that some pollsters think that you need to,” he said. “We have not found that to be necessary.”
Porn pointed to EPIC-MRA’s fall 2016 polling, which he called “pretty accurate” and showed days before the election that Clinton had a 4 percentage-point lead over Trump, tighter than the 7-point lead she had at the end of October.
If there had been another of his firm’s polls in the field the weekend before Election Day, Porn said he was sure it would have shown the race even. Trump ended up winning the state by 11,000 votes.
Asked if he thought weighting by education would have changed the surveys’ results, he responded: “No, not at all.”
While neither poll is changing its approach to the education aspect, Franklin is tweaking one element of his surveying in order to get at another observation in the AAPOR report, which noted that some Trump supporters who participated in polls prior to the election didn’t reveal their allegiances.
The report noted that could be because the voters decided late to vote for Trump, or they misreported their stances in the polls, the so-called “Shy Trump” effect.
Franklin said he’s looking to get at that by pushing respondents to select a candidate in a given race. For example, after initially asking respondents who they’d cast their ballots for, a pollster would follow that up with the question: “If you had to decide today, how would you vote?”
“That (question) is explicitly there to try to capture those undecideds who may not really be ready to say who they’re voting for,” he said.
Some 13 percent of voters in Wisconsin, Florida and Pennsylvania decided who they’d vote for in the final week of the campaign, and those voters broke by almost 30 points for Trump in Wisconsin, per the report.
Porn said his firm’s pollsters also push undecided voters to express their opinion by asking respondents who they’d vote for if the election were today and they had to decide now. He added pollsters also seek to determine the intensity of respondents’ views, whether they’d definitely vote for a candidate versus probably, for example.
Pew Research Center senior research methodologist Andrew Mercer cautioned the approach of pushing voters to select a candidate isn’t infallible, as surveys aim to give a snapshot of respondents’ answers at a given point in time and who’s undecided in an election could change moment to moment.
“Even if you push them and even if you do get them to tell you who they’re leaning toward, there’s no guarantee that’ll still be true in a week, two weeks,” he said, adding: “People can always change their mind about who they’re going to support.”
Going forward, pollsters and survey experts said consumers should keep in mind the limits of survey results and generally look for averages of many polls, rather than drawing conclusions based on one.
Mark Blumenthal, a freelance survey research consultant, noted that looking at multiple surveys “means you’re relying on the wisdom of many” pollsters, a practice he said is generally a safer bet than just a single one.
Still, he cautioned it’s possible many polls can “all be wrong together” and demonstrate systematic errors.
Mercer, of the Pew Research Center, said it’s important for members of the public to have proper expectations when reviewing results. For example, he said, surveys can tell readers they’re looking at a close election but “can’t necessarily go much further than that” to actually go on and predict the winner.
“If you’re realistic about what you’re trying to take away from polls, I think you can have a lot of faith in them,” he said.