After what happened in 2016, should you believe 2020 polls?

If all those polls in 2016 had been right, Hillary Clinton would now be running hard for a second presidential term.

Democrat Clinton did win the national popular vote that year, as most of the polls said she would. And her 2.1% margin was close to the 3% forecast.

But in presidential elections, it’s the Electoral College that matters. And that’s where many polls got it wrong. Especially in three Rust Belt states — Pennsylvania, Michigan and Wisconsin — that had long been part of the Democrats’ so-called “Blue Wall.” Republican Donald Trump turned all three red, though just barely, and he won the White House.

In North Carolina, a perennial battleground state, the polls that came out closest to Election Day 2016 were all over the map: The New York Times poll saw a tie. The CBS News and Quinnipiac polls had Clinton up by 3%. And Raleigh TV station WRAL said Trump was ahead by 7%. When the real votes came in, it was Trump by 3.6%.

“The 2016 presidential election was a jarring event for polling in the United States” was how the American Association for Public Opinion Research began its post-mortem evaluation of what happened. “Pre-election polls fueled high-profile predictions that Hillary Clinton’s likelihood of winning the presidency was about 90 percent.”

Fast-forward to 2020. Another presidential election year, another steady parade of polls. But given their track record in 2016 and the real complications of holding an election during a pandemic, why should people believe the polls this year?

‘Weighting’ education

One reason — according to pollsters, professors and political analysts interviewed by the Observer — is that more polls this year are asking some questions they should have been asking in 2016. Especially this one: whether the voter being interviewed has a college degree.

“People with college degrees are more likely to answer their phones and talk to pollsters,” said Jason Husser, director of the Elon University Poll in North Carolina. “Education wasn’t always correlated with party as much as it is now.”

Translation: College-educated voters tend to skew Democratic and that meant Clinton supporters were over-represented in pollsters’ samples. That was especially a problem in those Rust Belt states, where non-college-educated voters make up a bigger portion of the electorate than in many states.

“A lot of lower-educated voters showed up because of the populism of Trump,” said Brock McCleary, a Republican pollster who conducts the Civitas Poll for the conservative-leaning Civitas Institute in Raleigh.

Amy Walter, national editor of the Cook Political Report, said this issue of college-educated vs. non-college-educated has turned into “one of the biggest dividing lines, especially among white voters.”

Pollsters are trying to do a better job this year of including the right percentage of working-class voters by giving proper “weight” to education, said Peter Francia, a professor of political science at East Carolina University who directs the ECU Poll — the newest survey in the state, having debuted this Election Year.

He said pollsters can get it right either by sampling more from less-educated voters or by following a mathematical formula that will make sure the percentage of non-college-educated voters reflects the percentage in the electorate.

Many pollsters are making that adjustment in 2020. But not all, Courtney Kennedy, director of survey research at Pew Research Center, told Axios. “Alarmingly,” she said, some polls done this year in Florida, Michigan, and Wisconsin, have not improved.

“The structural challenges we had in 2016 are still with us,” she told the news site.

Fewer undecideds

Tom Jensen, who directs Democratic-leaning Public Policy Polling in Raleigh, points to another difference this year: undecided voters. There seem to be far fewer of them as Trump battles Democratic Joe Biden. And this time, third-party candidates appear to be less of an attraction to voters than the Libertarian and Green Party candidates in 2016.

In September 2016, Jensen’s firm, polling in Michigan, had Clinton ahead of Trump, 44% to 40%. This September, he has Biden up, 50% to 46%. Same margin, but many more voters have made up their minds.

“There was a lot more room in 2016 for undecideds to break strongly and late for one of the candidates,” Jensen said. “Most of the undecideds (in Michigan) were Republican-leaning, but didn’t like Trump. But in the end, they decided they didn’t like Clinton more.”

Added Walter: “One thing pollsters didn’t do in 2016 ... is they weren’t polling in the week going into the election. And that’s when a lot of people decided. The race had been stable, stable, stable. Then four or five days before the election it broke.”

Jensen learned another lesson in 2016: Don’t assume that people who have not voted in the last few elections won’t vote in this one.

“A lot of people came out to vote who may have been registered, but who hadn’t voted for a decade. They were not considered ‘likely’ voters,” he said. “Now we call everybody. It’s not based on voting history. (In 2016), we didn’t call everybody. And it turned out that Trump had this unique appeal with voters who came out to vote who had not voted in a long time.”

But with weeks to go before Election Day 2020, intense support for Trump on one side and burning antipathy toward him on the other have led many more to decide early how they’ll cast their ballot.

In the most recent ECU poll, 96% of those North Carolinians who intend to vote for Trump or Biden said they’d made up their minds. Only 4% said they were open to changing them.

Proceed cautiously

While many pollsters adjust their methodology and adapt to new circumstances, others — voters, journalists and political forecasters —should keep some things in mind as this election season enters the homestretch.

Walter, host of a podcast called “The Takeaway: Politics with Amy Walter,” advises looking for trends in polls “rather than cherry-picking numbers.”

And speaking of numbers, Walter said, appreciate the margin of error in any poll. “If it’s a 2 or 3 point race,” she said, “it could be 6, it could be zero.”

Elon University’s Husser said consumers of polls should remember that each poll is “a measurement, not a prediction.” With possible surprises in October or before, “things could easily change.”

In the last days of the 2016 campaign, for example, then-FBI director James Comey announced that the bureau was re-opening the private-email case against Clinton. A few days after that, he re-closed it. But, for Democrats, the damage was done. Clinton’s poll numbers took a late dip.

Then there were those forecasts in 2016, with some media outlets announcing that, based on their analysis of the polls, Clinton had an 85% or greater chance of being elected the 45th president.

“What got people off the scent was the forecasting that overstated Clinton’s likelihood of winning,” said Francia.

Instead of using percentages to explain the likelihood of something happening, Francia said he uses a football analogy. The analogy can show how unlikely things do happen sometimes.

Take The Huffington Post’s forecast that Clinton had a 98% change of winning, for example. That’s “like kicking the extra point — they make it almost every time.”

But, he said, the 71% forecast from Nate Silver, the statistics wunderkind who edits FiveThirtyEight, “is more like a 40-yard field goal. It should go through, but if you’ve watched enough football games, you’ve seen some kickers miss.”

Added David McLennan, director of the Meredith College poll in Raleigh: “Polling is based on probability, not certainty.”

Big unknown

Finally, the country is holding a presidential election in the middle of the worst public health crisis in more than 100 years. And COVID-19 presents difficulties and advantages for pollsters trying to gauge where voters stand.

For starters, more people are at home because of the coronavirus. “More people are answering than usual,” said Jensen. “There’s a better response rate.”

But in other ways, polling during a pandemic can be a challenge.

McLennan wonders: How will colleges releasing students back to their home counties — or other states — affect voter turnout?

Republican pollster McCleary said he’ll be asking parents of elementary- and high school-age children whether frustrations related to this “experiment with learning” will perhaps drive more of them to the polls this year.

And political analyst Walter said one of the biggest unknowns is “what’s happening in a place like North Carolina: a record number of (people) voting by mail. How much bigger is that going to make the pool of voters this election? How does that change the makeup of the electorate? Is that going to mean that somebody who might have never voted before is going to be engaged?”

And will President Trump’s unsubstantiated claims that so-many mail-in ballots will lead to “fraud” have any effects on polling?

Such questions can’t be answered yet, said Francia of ECU. “We can’t say: ‘This has happened before. Here’s what to expect.’ There’s been nothing comparable in recent times.”

Going back to the 1918 election, when Americans voted during the Spanish flu outbreak, isn’t much help, Francia said. Back then, women couldn’t even vote.

The encouraging news is that, by the mid-term elections of 2018, the polling mistakes of 2016 were leading to some changes. Amid a record turnout for an off-year election, Walter said, the polls “basically got it right.”

In 2016, PPP’s last Michigan poll, which did not take into account respondents’ education levels, had Clinton up by 6 comfortable points. Days later, she ended up losing the state by 0.3%.

But in 2018 in Michigan, PPP “nailed” the governor and U.S. Senate races, both of which went to Democrats, Jensen said.

“After that rough year in 2016,” he said, “2018 was one of the best years for pollsters. We had a pretty good handle on it — after making adjustments.”

This story was originally published September 12, 2020 at 9:56 AM.