A girls colours an electoral map of the United States in either red or blue as returns are announced for the U.S. general election at Republican Governor Pat McCrory's election-night party in Raleigh, North Carolina on Nov. 8, 2016.Jonathan Drake/Reuters
With the 2020 presidential election looming, David Burrell knew his polling data was wrong – he just couldn’t figure out why.
His company, Wick, had emerged in 2008 as a tech-based challenger to old-school market-research companies, yet his polling showed the same unsatisfying result that he was seeing across the U.S. political landscape: an easy Biden victory verging on a landslide.
“I didn’t think it was right,” he said from the company’s Atlanta headquarters. “I talked to our customers and they didn’t think it was right either.”
He hit the books, poring over academic journals and making methodological adjustments on the fly that made Wick one of the more accurate pollsters this cycle. His efforts distanced the company from a broader industry misfire that prominent Republican pollster Frank Luntz said could spell the end of political polling.
With vote counting still under way, it’s far too early to say how the 2020 election cycle will be regarded within the checkered annals of U.S. pre-election polling. It’s not the epic failure of 1948, when pollsters predicted a comfortable Thomas Dewey victory over Harry Truman. Nor is it 2012, when a field of statistical whiz-kids led by FiveThirtyEight.com’s Nate Silver came up with a stunningly accurate state-by-state forecast of the Obama victory.
But it’s clear that many pollsters got it wrong. On election eve, an NBC News poll tracker put the average Biden lead at seven percentage points. FiveThirtyEight put his advantage at 8.4 points. Most showed that the Democrats would flip the Senate and earn a larger majority in the House.
In reality, Biden’s lead is razor-thin, the Senate is poised to remain in Republican hands and Democrats have given up seats in the House.
“It’s rather embarrassing some poll results were so far off target,” said W. Joseph Campbell, author of Lost in a Gallup: Polling Failures in Presidential Elections. “Where was the blue wave that was anticipated by polls but never developed?”
Pollsters faced a similar predicament in 2016, when Hillary Clinton was heavily favoured to win the presidency, based on extensive pre-election polling.
In the wake of Donald Trump’s surprise victory that year, the American Association for Public Opinion Research (AAPOR), the country’s largest organization of survey researchers, found that pollsters had failed to adjust for an overrepresentation of college grads, who tend to lean toward Democrats.
Some companies adjusted accordingly this time around. AAPOR is warning that it’s too early to draw any conclusions about polling accuracy for 2020.
But some miscues are too glaring to avoid. No public poll suggested that Republican Senator Susan Collins had a hope of re-election in Maine (she’s up by nine points). One put Biden up by 17 points in Wisconsin (he’s ahead by 0.6). Most showed Biden was a lock in Florida (Trump took it by three points).
The venerable polling centre at Quinnipiac University showed an 11-point lead for Biden nationally as well as advantages in Florida and Ohio.
“A full examination of what went wrong with polls this year is going to take a while,” said the poll’s director, Doug Schwartz. “At the moment, I still need to see the final election results and final exit-poll results, and without those I’m not able to make even preliminary hypotheses about what exactly the issues are.”
Another theory hatched after the 2016 election held that social-desirability bias made Trump supporters hesitant to declare their voting intentions to pollsters. This became known as the “shy Trump theory.” Researchers can only draw simple conclusions about the elusiveness of Trump voters this time around.
“His supporters may be particularly unlikely to participate in polls, at least in some parts of the country,” said Courtney Kennedy, director of survey research at the Pew Research Center.
But those who did participate were far more enthusiastic than Biden supporters, something the IBD/TIPP poll weighed to reach one of the most accurate predictions of popular support for the candidates. “There were these 10-point leads for Biden in polls and I wasn’t seeing any enthusiasm for him,” said Raghavan Mayur, president of TechnoMetrica, which runs the IBD/TIPP poll. “Common sense told me something was off.”
At Wick, Mr. Burrell said he found that Trump supporters simply weren’t participating in polls. They weren’t shy, he says, they were disillusioned.
“It’s clear there’s a subset of the population who feel underrepresented and disenfranchised who don’t want to take polls any more,” he said. “They don’t trust the democratic process, they don’t trust the media. That’s a difficult thing to account for.”
In recent weeks, Mr. Burrell dropped $2,000 on scholarly articles looking at polling methods in the Soviet Union, wartime Germany, China and other non-democratic societies where a lack of institutional trust led to disengaged polling subjects.
He made an array of methodological changes. He found that Wick had an overrepresentation of voters with postgraduate degrees and early voters – two categories that lean toward Mr. Biden – and an underrepresentation of white working-class voters, who favour Mr. Trump.
After accounting for these issues, Wick polls correctly picked winners in most of the battleground states. However, he also had Mr. Trump winning by a hair.
“I’ve heard people say this will kill polling," he said of industry errors. “No it won’t. The demand for accurate public-opinion research is higher than ever. This will get corrected, people will figure it out.”
Our Morning Update and Evening Update newsletters are written by Globe editors, giving you a concise summary of the day’s most important headlines. Sign up today.