The election night party Joe Biden planned in Delaware was built for a blue tsunami, with front-row seats for VIPs and a giant flag hung between two cranes, as his campaign advisers ultimately could not resist planning as if the polls that suggested a quick and commanding defeat of President Donald Trump might be right.
Instead, Democrats endured the second polling shock in four years, with election returns frequently coming in more favorable to Trump than in public surveys, in some cases outside the margin of error.
The underestimation of Trump’s turnout and support in many places, after similar issues in 2016, has raised again questions about the reliance of campaigns, the press and the public on surveys to shape the race. That in turn has prompted new questioning of the bedrock principle among political strategists that campaigns can divine public opinion before the votes are counted with enough money and talent.
Adding to the prominence of polling issues is an emerging prognostication industry built on public surveys, with news organizations and universities investing heavily in polling, and aggregation websites promising election predictions that can offer a deceptive level of confidence.
Trump seized on the polling misses Thursday in an address from the White House, where he cherry-picked polls that proved wrong, including one of Wisconsin voters in October by The Washington Post and ABC News, to argue falsely that the surveys reflected an organized effort to discourage his voters with misinformation.
New technologies, growing political polarization and candidates with devoted followings like Trump have at the same time presented challenges to uncovering what is happening and, therefore, make it more perilous to pretend as if anyone does. The errors affect public perceptions of the campaign, media coverage, campaign tactics and even fundraising efforts.
The Biden campaign appeared to place big bets late in the campaign in the hopes that the public polling would be right, expanding its television buy to states Trump would win easily. In the final week, Biden made appearances in Ohio, where he stands eight points behind with 93% of votes counted, and Iowa, where he lost by eight points with 92% of votes counted.
“We were operating in a reality that wasn’t reality, and we were operating off numbers that just were clearly not reflecting what turnout would be,” said one Democratic consultant, who spoke on the condition of anonymity to speak frankly. The Biden campaign officials have maintained that their internal polling proved more accurate than the public numbers.
Although the degree of polling misses is not yet clear as states continue to count mail-in ballots, a review of polling in 10 key states with more than 85 percent of the vote counted finds that public polls underestimated Trump’s vote margin by about 4.5 percentage points on average, similar to the size of errors in key states four years ago.
Polling problems were again concentrated in the Upper Midwest, with averages of polls in Wisconsin and Michigan showing Biden with leads of seven to eight percentage points before Biden eked out one- and three-point victories, respectively. Pennsylvania remains a close race, though Biden is likely to fall short of his five-point advantage in preelection polls.
Polls told a more accurate story in other key states, showing close races in both Arizona and Georgia, which appear to have mirrored the outcomes. And although polls in Florida and North Carolina overstated Biden’s support, they largely showed a tight and competitive race.
An association representing polling professionals released a statement Thursday asking for critics to withhold judgment until all the votes are counted.
“It will take weeks for election officials to carefully count all early, absentee, in-person and provisional ballots,” the American Association for Public Opinion Research said in a statement. “As such, it is premature to make sweeping judgments on the polls’ overall performance before all the ballots are counted. Patience is necessary.”
The frustration, however, has been clear for months as both Democratic and Republican campaigns have had to focus their energy to push back on polling-driven narratives that are now in doubt, while also assessing the reliability of the internal polls that drive their strategies.
“As much as there needs to be a reckoning for campaigns and issue campaigns, it is also for me a reckoning for the media that obsessively covers public polls,” said Michael Halle, a longtime Democratic consultant who worked for former Secretary of State Hillary Clinton and former South Bend, Ind., Mayor Pete Buttigieg.
During the campaign, Trump complained repeatedly of “suppression polls” by the news media, which he argued were an effort to discourage his supporters from coming to the polls. His advisers warned privately and publicly that traditional polling methodologies were not correctly reading the president’s support by failing to capture their expected turnout.
On the trail, Trump would cherry-pick and read out polls that were more positive for him than public polling averages and internal data the campaign was using, which continued to show trouble in the northern states that are now likely to prove decisive if Biden wins. But as the campaign closed, his top advisers sought comfort in the fact that Trump had beaten the polling data four years earlier.
Republican strategists working on key Senate races have also argued for months that inaccurate public polling was having a damaging effect on their campaigns. Public polls in North Carolina and Iowa underestimated Republicans’ vote margin by about five percentage points on average compared with the current vote count, which is about 95 percent complete. In North Carolina, polls showed a small advantage for Democrat Cal Cunningham, which flipped to an apparently small win for Sen. Thom Tillis, a Republican.
Democrat Sara Gideon led at least numerically in all 14 public polls tracked by the website RealClearPolitics in the U.S. Senate race in Maine, including seven polls that showed her leading by five points or more. Sen. Susan Collins, a Republican, ultimately won reelection in that contest, notching a seven-point lead, with 88 percent of votes counted.
Jesse Hunt, the communications director for the National Republican Senatorial Committee, said the public polls were consistently different from the numbers Republicans were getting internally for that race and became a burden for the Collins effort.
“These polls can have a major effect on candidate fundraising and the trajectory of the race,” Hunt said. “And it is irresponsible for reporters to write stories and horserace coverage off of the public polling.”
When polling missed in the 2016 race, the polling industry and Democratic Party insiders launched separate investigations of the errors. They concluded that there were multiple causes of the failure, including a sharp divide in the Upper Midwest between college- and non-college-educated voters that many polls did not account for when calibrating their samples. Methodologies were adjusted by some pollsters this year to weight by education, but it is not clear that the adjustments improved the results.
The 2016 race also featured a clear last-minute shift in Trump’s direction, as late deciders shifted and others changed their minds. Public polls this cycle showed a far more modest shift. And as in cycles past, there remained notable outlier polls, including the Post-ABC News poll in Wisconsin that showed Biden winning by 17 points in late October, a result that scared Democrats in the state nearly as much as Republicans, because they feared the numbers would be taken at face value by voters, depressing Democratic turnout.
The Post story noted that the poll represented a shift from September, when Biden held a six-point edge. Slightly more registered voters in the October poll reported having voted for Clinton than Trump in 2016, whereas the previous survey showed voters essentially split.
“We are taking time after the election to review all our polling, including why our October Wisconsin poll overestimated Biden’s support, and will use the findings to strengthen future polling,” Molly Gannon Conway, communications manager at The Post, wrote in an emailed statement. She added that, when other Post-ABC surveys are taken into consideration, the “balance of Post-ABC polling this cycle contributed to our reputation for rigor.”
Since the results have started to come in, Democrats have been divided about how badly their internal numbers missed, with several Democratic pollsters declining to discuss the issue until the final vote count is known. Biden campaign manager Jen O’Malley Dillon made clear publicly weeks before the election that the public margins for Biden were greater than the campaign’s internal polling.
“We have been abundantly clear that we thought this would be a very close race in most of our states,” O’Malley Dillon said in a briefing to reporters Wednesday. “We worked hard to make very transparent the fact that we did not believe that the race was in the wide margin that we were seeing as the national poll numbers.”
The Democratic consultant who spoke on the condition of anonymity said that pollsters, both inside and outside of campaign efforts, have been forced to increasingly make estimations because of an inability to get consistent responses across all voter groups.
“Where polling is fundamentally broken now is these guys have to guess at what the electorate will look like and sometimes guess at how the electorate feels,” this person said. “So we are piling guesses upon guesses upon guesses.”
One of the few bright spots this cycle was J. Ann Selzer, whose firm Selzer & Co. does polling for The Des Moines Register and found Trump leading in Iowa by seven points just days before the election. Her poll looked like an outlier at the time. It was the only public poll in the state since June to get within two points of the final result.
Part of her success was timing, as polling did suggest a late shift in support toward Trump, following her September survey that showed the race tied. But she credits her accuracy to her knowledge of the state and the assumptions she was able to make about how different groups would vote.
Unlike other polls that weighted their polls by educational attainment, she weighted her poll by age and sex, and isolated the different parts of the state by congressional district to get a better read on the urban and rural divide.
“It is a science of estimation,” she said. “The key word being estimate.”
— — –
The Washington Post’s Annie Linskey in Wilmington, Del., and Scott Clement contributed to this report.