CHICAGO — On Election Day 2016, the polls showed Hillary Clinton with a nearly 4 percentage point lead in Michigan and a 6-point lead in Wisconsin, but Republican Donald Trump would go on to win both states by less than 1 point.
Of 28 Wisconsin polls conducted in 2016, all of them showed Clinton with the lead, according to polling data tracked by RealClearPolitics. In Michigan, 36 of 37 polls had Clinton ahead. And yet Trump’s victories in both states won him the presidency.
“None of those Wisconsin polls had Trump ahead, and four or five of them were mine. My final poll was plus 6 for Clinton,” said Charles Franklin, a political science scholar and director of polling at Marquette University Law School. “Nobody got it right.”
A little more than a week out from this year’s election, the polls are offering a similar narrative with Democratic presidential nominee Joe Biden ahead of Trump by an average of about 5 points in Wisconsin and nearly 8 points in Michigan. All told, just six of 104 polls conducted in the two states this year have shown Trump with a lead — and just two in the last two months.
But all of this raises a simple question: Are the polls actually right this time?
The president’s campaign and Republican leaders in both states contend the race is closer than the surveys show — especially the recent polls that have Biden opening up a double-digit lead following the first debate and Trump’s COVID-19 diagnosis. For his part, Trump repeatedly has drummed up enthusiasm among his supporters by emphasizing how wrong the polls were four years ago.
“How about the last election? I was down in all nine places that I had to win. That wasn’t a good feeling,” Trump said during a recent rally in Janesville, Wisconsin. “By the end of the evening, I won all nine places, right? Think of that. Other than that, they did a great job of polling.”
Democrats likewise are waving off the polls, with Biden campaign manager Jennifer O’Malley Dillon writing in a recent memo that “even the best polling can be wrong” as she called on supporters to “campaign like we’re trailing.”
“I don’t care about the polls. There were a whole bunch of polls last time, didn’t work out,” former President Barack Obama said at a drive-in rally for Biden last Wednesday in Philadelphia. “Because all those folks stayed at home and got lazy and complacent. Not this time. Not in this election.”
Pollsters, however, point to a variety of factors that give them more faith their numbers are more accurate this time — fewer undecided voters, less influence from third-party candidates, a better understanding of Trump’s voters and very little fluctuation among Americans in their support for — and opposition to — the president.
“This year, I think the polls are going to be more reflective of the actual vote,” said Barry Burden, a political science professor and director of the Elections Research Center at the University of Wisconsin at Madison. “It’s not because the polls and the technology that are used have been improved tremendously. It’s mostly because it’s a better environment for polling.”
After the polling mishaps four years ago, Burden was charged with heading up a new battleground states survey for the University of Wisconsin that has polled the three swing states that Trump won by less than 1 percentage point — Wisconsin, Michigan and Pennsylvania.
Burden described a “paucity of high quality polling” in those states, particularly in Michigan, where he said plenty of “fly-by-night operations” and automated robopolls helped contribute to the bad polling averages.
So did a large share of undecided voters.
At this time four years ago, polls in Michigan and Wisconsin had anywhere between 10% and 15% undecided voters. But this year, that number is closer to 5%, Burden said. That leaves less room for volatility late in the race.
In addition, undecided voters unexpectedly broke big for Trump. Wisconsin exit polls in 2016, for example, showed those late undecided voters backing Trump by a 2-to-1 margin, Franklin said.
Because of that unexpected late shift, Franklin said Marquette added a question to its polls this year pressing undecided voters on who they’d cast a ballot for if they had to vote that day. That has helped them to conduct a statistical analysis on how those few undecided voters might break this time. So far, Franklin said, they are splitting evenly between Trump and Biden, making it less likely for a last-minute dramatic shift in support.
Four years ago, Trump and Clinton were both historically unpopular candidates. That’s less of a factor this year, both pollsters agreed. Trump is more popular with the GOP base than he was in 2016, and Biden has a much higher favorability rating among voters at large than Clinton had.
In 2016, Wisconsin polls showed that 20% of voters disliked both Trump and Clinton. Of those dissatisfied voters, exit polls showed 60% backed Trump, 20% backed Clinton and 20% voted third party, Franklin said. This year, the number of voters who dislike both candidates is just 8%, he noted.
Fewer dissatisfied voters paired with the high-stakes nature of the race has meant far less support for third-party candidates, removing another factor that led to late swings in the race four years ago, Franklin and Burden agreed.
Yet another reason the pollsters say their final surveys are likely to be more accurate: early voting. As many as 50% to 60% of voters could bank their votes by Election Day, Burden said.
“That winnows down the wiggle room for the polls to be wrong,” he said.
Also different this time: Trump isn’t just known as a bombastic TV-reality-show-star-turned-politician who might be worth a roll of the electoral dice. Now, he is an incumbent who has been impeached by Congress and has had to govern amid a pandemic and subsequent economic recession.
“People have strong views about the guy, and everyone has an opinion,” Burden said. “This is a referendum on his presidency, and there tend to be fewer undecided voters in that environment.”
The opinion of Trump has been incredibly stable, Franklin noted, pointing to the president’s approval rating hovering around 40% for much of his tenure. That makes it unlikely voters will change their minds about Trump in the eleventh hour.
“When you look at job approval in the Gallup poll going back to Franklin Roosevelt, no president is anywhere close — ever — to how small the variation has been in Trump’s approval rating over the four years,” Franklin said.
The biggest miss among many pollsters four years ago was not accounting for the educational makeup of the electorate. That turned out to be a critical factor as Trump’s support skewed heavily toward white, non-college educated voters.
Typically, polls “weight” their results to give proper representation to a group that might have been under-sampled in a survey, compared to census data for that area. The practice is done by race, age, sex and other factors, but many state-level polls four years ago did not weight education.
“Less educated people are harder to get on the phone, and they’re harder to get to interview for a poll, but that has been true for decades,” Franklin said. “Twenty years ago, if you had too many highly educated people in your poll, it didn’t shift your vote estimate much but in recent elections, it does and that’s why it’s so important to weight it.”
While Franklin did weight voters by education in 2016, his final poll was still off by 6 points. If he hadn’t accounted for that factor, Franklin said it probably would have been off by about 9 points instead.
In his post-mortem analysis, Franklin placed most of the blame on missing the mark on the high number of undecided voters moving late to Trump — a dynamic that is unlikely this time because it’s a smaller pool of voters.
“My conclusion is our errors came because people were genuinely conflicted about Trump, and that was especially true among Republican and Republican-leaning voters. They really loathed Hillary Clinton and couldn’t imagine voting for her, which left them with a third-party candidate or sitting out the race,” Franklin said. “And in the end, a lot of them came back and voted for Trump.”
Both the Marquette poll and the University of Wisconsin’s battleground state poll have placed Biden’s lead in Wisconsin at about 5 points. The battleground poll has had that lead at about 6 points in Michigan.
Franklin and Burden use different methodologies. Marquette conducts random live telephone interviews with likely voters while the University of Wisconsin partners with YouGov to identify a group of 800 likely voters in each state and then surveys the same group of voters online periodically throughout the election cycle.
“One thing that has been reassuring is we’ve been using two different types of surveys but are producing very similar estimates of the breakdown of Biden and Trump,” Burden said of the two polls. “It’s comforting to see them be pretty similar.”
Other polls have shown a larger lead for Biden in recent days in both Wisconsin and Michigan, with the margin growing since the president’s confrontational first debate and his hospitalization with the coronavirus.
A recent New York Times/Siena poll put Biden’s lead in Wisconsin at 10 points, while a new Fox News survey has Trump down 12 points in Michigan. Franklin said he didn’t see as large of a bump in his polling following the debate, which also partially coincided with Trump’s initial time in the hospital.
Both noted how major news stories gave Clinton temporary advantages in 2016 polls that quickly disappeared. Burden pointed to Trump’s “Access Hollywood” tape scandal, in which he was recorded bragging about groping women, as an example of a polling sugar high that wore off.
Whether that’s the case with Trump’s debate performance and COVID-19 diagnosis remains to be seen, the two pollsters agreed. Both will release their final surveys this week.
“I think it’s impossible to look at my polling and everyone else’s polling in Wisconsin and not come to the conclusion that Biden has a reasonably solid lead here. We can debate how big it is, but there are no recent polls that have Trump ahead,” Franklin said. “Of course, that’s where the alarms go off because that was true in 2016, and that’s where legitimate caution comes in — but the fact remains we have a lot of reasons to feel more confident in the accuracy this time.”
Biden’s campaign has told supporters not to believe it has a double-digit lead in any swing state, with O’Malley Dillon emphasizing that “every indication we have shows that this thing is going to come down to the wire.” Wisconsin Republican Party chairman Andrew Hitt also dismissed the polls showing Trump trailing by double digits in the state and lauded the Marquette poll, which has shown a closer margin, as the “gold standard.”
“Our internal numbers and modeling tell us this is a very close race, razor tight,” Hitt said. “All of this feels extremely similar to 2016, where the public polling says the president is dead, but our internal polls say it’s very close and we can see this underbelly of growing support going through the roof.”
While the fundamentals of the 2020 race have been far more stable than 2016, there is one potential wild card — turnout.
Records already have been set for the number of early votes cast and some projections suggest turnout could hit the highest rate in more than a century. If record turnout leads to a particular demographic or voting bloc turning out in higher-than-expected numbers, it could affect the accuracy of polls.
“Every election there is some new challenge for polling that’s not really anticipated or where there isn’t a solution in advance,” Burden said. “The likely high turnout rate and the different patterns of turnout are the biggest challenges this year. We just don’t know.”
New voter registrations also could affect turnout. Overall registration numbers have been down in many areas as fewer registration drives have been conducted amid the pandemic.
In some states where voters are registered by party, including Pennsylvania and Florida, data has shown Republicans narrowing the gap in registrations with Democrats, largely because the GOP continued knocking on doors and holding event throughout the summer.
In Michigan and Wisconsin, voters do not register by party so it’s harder to get a feel for whether one side has an advantage, the pollsters and party leaders all agreed. Still, Hitt, the GOP chair in Wisconsin, said his party has put a greater emphasis on it this time, and some analyses have shown voter registrations dropping off at a slower rate in counties won by Trump four years ago.
Complicating matters, however, is the fact that Michigan and Wisconsin both have same-day voter registration, a tool Democrats most frequently push.
“It’s going to come down to how you turn out your voters, and that’s where I think we have an advantage. The Democrats don’t have a ground game. They’re not knocking on doors,” Hitt said. “They’re up in the cloud, and we’re down in the ground turning people out.”
In Wisconsin, Democrats have refrained from knocking on doors because of high COVID-19 cases, but they have dropped off campaign flyers in some areas, local officials said. In Michigan, Democrats did start canvassing door-to-door a few weeks ago, but have pulled back from some areas as COVID-19 cases have started to surge again, said the state’s Democratic Party chairwoman Lavora Barnes.
Barnes said she tells anyone who will listen to ignore the polls, but acknowledges Democrats are in a much better spot than four years ago. Biden and his running mate California Sen. Kamala Harris are campaigning much more in Michigan and the state party has a larger infrastructure and has been far better organized, she said.
“In 2016, those of us on the ground in Michigan knew it was going to be very close and we were very nervous that the numbers would not be there and it turned out that we were right,” Barnes said. “It’s just night and day this time around with having this operation on the ground, having the Biden campaign actively campaigning here, having the candidates come here. It’s completely different.”
Part of how Burden has tried to gauge potential turnout in his battleground polling has been to ask respondents whether they have been contacted by a campaign. In Michigan, he said, voters were more likely to hear from Democrats while Wisconsinites were more likely to hear from the Trump campaign.
That’s what has Tracy Thompson worried about the polls. She chairs the Rock County Democratic Party in Wisconsin, where Trump had a large rally earlier this month in Janesville and said she’s not convinced Biden has a big lead like some recent surveys have suggested.
“I’ve driven around the state recently and there are pockets where you just see Trump signs everywhere — big ones, multiple ones — and if I was to go off those, I would say Biden has no chance,” said Thompson, whose historically Democratic county Clinton won by 10 points in 2016. “Trump has been focused on Wisconsin for a long time, he’s visiting here and I’m getting at least three or four text messages a week from his campaign or allies urging me to vote for him, and I’ve never seen that before. They’re intense.”
While Democrats may be engaging in virtual organizing that isn’t as visible, Thompson said she’s still worried.
“It’s hard to say what’s going to happen, but I just have to be optimistic,” she said. “It’s going to come down to last few counted votes, I do believe. The race feels that close.”
©2020 Chicago Tribune