By Pat Poblete, WisPolitics.com
Another presidential election, another set of polls that missed the mark.
“The polls were wrong, no doubt about it,” said pollster Paul Maslin, a partner at FM3 Research who conducts private polling for left-leaning candidates. “In some ways, they may be more wrong this time than they were in 2016.”
Charles Franklin, director of the Marquette University Law School Poll, agreed.
“If 2016 put (polling issues) on the examining table, they’re all on the emergency room table now after last night’s polling errors,” Franklin told WisPolitics.com on Wednesday.
With unofficial results from all of the state’s 3,689 precincts in, a host of media outlets have called the race for Joe Biden. The 1.63 million votes for the former vice president give him a slender edge over President Trump, whose 1.61 million votes bring him well within the margin for a recount. His campaign has indicated it will seek one “immediately.”
Should the unofficial margin hold through normal checks and a potential recount, the race will finish Biden 49.6 percent, Trump 48.9 percent.
That result may come as a surprise to those tracking the myriad of public polls in Wisconsin. In the closing days of the race, polls showed Biden headed for a comfortable victory with most having him above 50 percent. Surveys were less frequent in 2016, when Hillary Clinton also was seen to be on track for beating Trump.
Biden won. Clinton lost.
But Biden’s advantage of 0.6 percent is a far cry from the margin suggested by the many polls.
Franklin’s Marquette University Law School Poll consistently pegged Biden’s lead around five points. His last survey of the campaign cycle, which came out of the field eight days before the election and was released three days later, put Biden up 48-43 with a margin of error of plus or minus 4.3 percentage points.
Another Wisconsin-based poll released the week before the election gave Biden a 9-point edge. The poll from the Elections Research Center at UW-Madison, which was in the field Oct. 13-21 and was released Oct. 26, found 53 percent of likely voters backed Biden, compared to 44 percent who supported Trump.
The poll led by UW-Madison political scientist Barry Burden tracked the same group of people over the course of multiple surveys to detect changes over time. The survey found little change in the closing months of the presidential race.
Meanwhile, polling conducted by some national outlets missed by even more.
The last New York Times/Siena College poll, conducted between Oct. 26-30 and released two days before the election, gave Biden a 52-41 lead with a margin of error of plus or minus 3.2 percentage points.
The last Emerson College poll, released the same day after being in the field between Oct. 29-30, had Biden up 53-45 with a margin of error of plus or minus 3.5 percentage points.
CNN’s last poll of Wisconsin, conducted Oct. 23-30 and released Saturday, pegged Biden’s lead at 52-44 with a margin of error of plus or minus 3.9 percentage points.
Those polls came close to projecting Biden’s strength, but badly missed forecasting Trump’s support.
The largest outlier of the group, a survey from the Washington Post and ABC News, gave Biden a 17-point lead in a poll released Oct. 28. That poll, conducted by landline and cell phones between Oct. 20-25, found 57 percent of likely voters backed Biden, while 40 percent supported Trump. The sample of 906 registered voters included 809 likely voters with a margin of error of plus or minus 3.5 percentage points among registered voters and plus or minus 4 points among likely voters.
And it’s not just public polls that missed the mark.
Sarah Simmons, a Republican strategist and polling expert from Purple Strategies who used to work in Wisconsin, said the errors extended to pollsters working for campaigns on both sides of the aisle.
“I don’t think the private guys were right either,” she told WisPolitics.com. “I don’t think anybody probably got it right, either side.”
Maslin, who indicated he didn’t poll heavily in Wisconsin this cycle, said he got “burned” on surveys conducted for the unsuccessful reelection campaign of U.S. Sen. Doug Jones, D-Ala.
“We always understood how difficult it was to get Doug Jones elected but our data did not suggest it would ever be a 20-point margin,” he said
Franklin told WisPolitics.com he needs to comb through the official data before drawing conclusions as to what went wrong.
“I think it’s a little bit about fighting the last war; you never know what the next problem is going to be,” he said.
But while Franklin, Maslin and Burden said a full postmortem would need to be conducted, they offered some early hunches.
Maslin said he saw pollsters correct several of the errors diagnosed in their surveys from 2016: they added more non-college-educated white people into the mix so their samples would more accurately reflect Trump’s charged-up base; and they stayed in the field later in the race to try to capture late deciders who broke heavily for Trump four years ago but largely went undetected at the time.
But Burden said weighting polls for education is not a “cure-all.” While both he and Franklin weighted their polls for education, he pointed to the Des Moines Register/Mediacom Iowa Poll conducted by pollster J. Ann Selzer, which did not.
“She had Trump leading by seven, which a lot of people dismissed, but turned out to be pretty close to the result,” Burden said. “So there’s a pollster who’s intentionally not accounting for education and yet got it pretty much right, and the pollsters who did account for education are mostly off the mark.”
Instead, Burden highlighted a different metric he said might do a better job of capturing the willingness of the president’s supporters: institutional trust.
“One theory that had been going around in 2016 is that there is a group of people who were sort of distrustful of institutions of media and of pollsters, and they were reluctant to participate in surveys,” he said.” They were private and protective and distrustful, and that correlates with being a Trump voter in 2016.”
Maslin said he is “absolutely becoming convinced” of the same thing.
“It’s not that Trump supporters are coy and shy — they’re defiant,” he said. “They don’t trust elites, academics, scientists, media, pollsters, any of us.
“It’s not that they do a survey and don’t tell us, I think they’re perfectly willing to tell anybody that they support Donald Trump. … [T]hey’re simply not participating.”
Maslin said his polling model has a profile of a Trump supporter, but he worries the supporters of the president he does reach don’t fully match that profile.
“The people I actually get in my sample are 60/40 for Trump of that profile, and I think I’ve weighted them correctly,” he said. “But what if in reality, that group that didn’t participate was more 80/20 or 90/10 Trump? Well then all the weighting in the world can’t correct for that, because they never even got into the sample.”
Franklin said he also worried about missing Trump supporters.
“One thing that we warned about all year, but did not find evidence for was that Trump supporters simply systematically declined to talk to pollsters,” he said. “It’s not for lack of looking for these things. It’s possible that that’s still there.”
In searching for solutions, all four offered a range of potential remedies — from shortening surveys to changing the way they reach voters to incorporating questions on institutional trust to gauging how many Trump supporters are being captured in samples.
“All of these are things that are being actively looked at, but I don’t think we are ready to say that any of them are magic bullets that guarantee success,” Franklin said.
But when asked about reaching Trump supporters they worried they missed, several of the pollsters referenced polling done by the Trafalgar Group, one of the few that correctly projected Trump’s victory in 2016. That survey, led by pollster Robert Cahaly, sought to identify “hidden” Trump backers through a proprietary methodology.
Because the Trafalgar poll didn’t disclose its methods, WisPolitics.com didn’t report on its findings. Its final Wisconsin poll conducted Oct. 24-25 gave Biden a 0.7 percent edge with a margin of error of plus or minus 2.89 percent.
The little that is disclosed about the poll shows it tries to overcome so-called “social desirability bias” — the tendency for respondents to say what they believe interviewers want to hear rather than what they actually believe. The poll uses a combination of short phone calls, text messages and emails to reach what Cahaly believes is an accurate representation of the electorate.
“This violates polling principle, by the way,” Maslin said. “It’s like the whole public principle is everybody has the equal chance of being interviewed.”
Still, Maslin said Trafalgar “may be onto something.”
“I think you’ve got to go with a hybrid approach, and this is what the Trafalgar guy is saying, and he may be right,” Maslin said. “You may have to treat parts of the population entirely differently.”
Simmons adds what she’s heard from insiders on Trafalgar’s methodology made sense in the current political climate.
“He’s saying, ‘Actually, I use this proprietary method where I look at social media and I look at other data sources’ — to me, that’s part of the answer,” she said.
Franklin was more skeptical.
“I think Trafalgar did come closer in a number of states, but you’ve also got to pay attention to where they were further off, not closer,” he said. “As with all of us, our successes and our failures, not just cherry-picking the best results.”
He highlighted largely accurate polling in the 2018 midterms and the Dem presidential primary early this year, noting the polling errors appear to be “so specific to Trump elections.”
“Maybe it does have something to do with Trump voters — having more difficulty reaching them, possibly having them turn out at different rates, maybe more energized by Trump than what we traditionally see in a pool of likely voters,” he said.
Quizzed on if that trend would continue in the next presidential election if Trump isn’t on the ballot, Maslin punted.
“We don’t know how much impact he’s going to have and what legacy to leave — he’s not going away as a political force,” he said.
“There are a bunch of things that none of us ever thought about before that we better think about because somehow the tried-and-true methods … maybe even they have flaws.”