Primary Tuesday in New York, Kentucky, and Virginia was a quiet one. There were no crowded victory parties or subdued concession speeches. There weren’t even many winners or losers. Instead of exit polls and vote tallying, reporters told viewers that they would likely have to wait another week for results in big-name races because of an increase in mail-in voting.
This is election night in the age of COVID-19 and a likely sign of what’s to come on Nov. 3.
Presidential pollsters, meanwhile, are scrambling to recalibrate their methods in the wake of unprecedented campaign and new voting standards. A lot is riding on the accuracy of polling this election cycle, especially after the fumbles of 2016, and research institutions are feeling the pressure. They fear that misrepresenting the outcome in November could lead to a further decay in public trust in news institutions and research organizations.
Polling has also become big business: Political polls can easily cost $20,000 to $40,000 each and help advertise and draw in big business for companies like Gallup, which then make hundreds of millions of dollars from corporate partners.
But this election presents an unprecedented challenge in accurate reporting. Pollsters worry that long lines at a smaller number of polling stations and an increase in early voting and vote-by-mail ballots will skew the exit polls that ABC News, CBS News, CNN, and NBC News often use to predict and call elections for candidates.
Those major news institutions are members of a group called the National Election Pool, run by Edison Research. The group has been the only one to conduct national exit polls in the U.S. since 2004. Fox News and the Associated Press were part of the group until 2016; now the two organizations conduct polling online, but they don’t talk to people in person. Edison Research told Fortune that the company typically conducts 100,000 in-person interviews with people as they leave the polls as well as some in-person interviews at early polling sites and phone interviews with absentee voters.
But when voters are forced to wait in lines for hours to cast their votes, as we’ve now seen happen in primaries around the country, they’ll be disinclined to stick around and answer questions from exit pollsters, said Trent D. Buskirk, the Novak Family professor of data science and chair of applied statistics and operations research at Bowling Green State University. Combine that with large groups of people staying home and voting by mail, or voting early over a number of days, and exit polls could be “more varied in their estimation,” he said.
Republicans and supporters of President Donald Trump say that they are more likely to resume their usual activities in the next three months than Democrats, meaning they may be the ones turning out to the polls, skewing exit polls further to the right, noted Buskirk, who also served as a chair at the American Association for Public Opinion Research.
Election projections, which factor in exit polls and the demographic data of likely voters, may be further askew if they don’t revise their calculations to account for specific regions that have been hard hit by COVID-19, said Buskirk, especially in regions where voting by mail isn’t widely available or where applications for a mail ballot aren’t sent to every voter.
A middle-aged black Democrat who has voted in every election since he was 18 would typically be included in projection models as a very likely voter for Biden, for example, but if he lives in a county that has been locked down because of the pandemic, he may not go to the polls despite what the models predict.
“Some of these likely voter models may actually miss the market,” said Buskirk. “They don’t incorporate some of the more geographically specific information about how corona is actually impacting people’s lives.”
It’s possible, he said, that a state like Ohio could be projected on the news to go to Trump on election night only to have the results show that vice president and presumptive Democratic candidate Joe Biden is the winner two weeks later.
“The problem, really, is not who wins the election because obviously the person who wins the election is going to win the election by the proper count and not because Fox News calls it,” said Buskirk. “The problem is really the public perception of the polls.”
In order to get things right for the 2020 election, new models that incorporate more nuanced geographic health factors will have to be built, he said.
“I’ve been doing this since 1988, and by far this is the year with the most last-minute changes in terms of dates and procedures and how votes are being reported,” said Joe Lenski, executive vice president at Edison Research. “I expect a lot of change between now and Nov. 3. Many election officials frankly don’t even know what procedures they’re going to have in place yet.”
Lenski says that Edison will use June and July primaries as testing grounds to make sure that it has protections in place to conduct surveying during a pandemic and to readjust its current models. The company will likely end up increasing the number of people contacted by phone to ask about vote-by-mail results and the number of exit surveys it conducts in early voting while decreasing Election Day exit polls.
Exit polls, said Lenski, will still probably be used to reflect the composition and thoughts of the electorate, but the process of any projection is “just going to be more cautious this year.” Typically exit polling data is used only to project landslide elections anyway, he said.
But a lot is riding on the accuracy of these polls. If exit polls are wrong, public faith in news intuitions, polling, and elections will erode, said Buskirk. It also makes it difficult for polling to remain accurate, as untrusting Americans will be less receptive to picking up the phone and answering survey questions. When a candidate is predicted to be winning by a landslide, voter turnout is also often lower; for that reason, some candidates even prefer to appear to be in tight races.
Generally speaking, the polling industry can be slow to adapt, said Chris Jackson, who heads up public polling at multinational market research company Ipsos.
“We tend to be conservative and do things that are tested and work a certain way,” he said. An environment that’s volatile and chaotic makes it particularly hard to take a current poll and forecast it forward to accurately predict what will happen in November.
“I think anyone saying they know what’s going to happen in November is really misleading themselves or their peers,” he told Fortune. “And we’re very much aware and concerned about how different voting rules or laws may actually impact who shows up.”
There are certain patterns of voting that have held up for decades and which have been used to inform predictions, said Jackson. Older people tend to vote more often than younger people; people who are more educated and affluent show up to vote more than those who are less so; women vote more than men; white Americans show up to vote more than black Americans, who show up to vote more than Hispanic-Americans.
But do social movements or widespread vote-by-mail change that?
“We have no idea,” said Jackson. “There are some examples we can look at in Colorado or Washington but on a national scale, no idea.”
The confusion comes as polling companies work to repair damaged reputations that came from the 2016 presidential elections, where survey results clearly missed their mark. At the end of the day, polls cost tens of thousands of dollars to create and are big business. Another disastrous election cycle could upend the industry.
Jackson says Ipsos is looking at a number of ways to make sure its surveys are more accurate. The company is changing how it weighs education, building sensitivity analysis into data, and making sure all subpopulations are being represented in correct proportions. Ipsos is also implementing control models to look at unemployment levels and GDP to see if they match what poll results indicate.
But Ipsos hasn’t yet implemented any changes around how voting will change this year.
“We’re thinking a lot about that,” said Jackson. “We haven’t changed the way we’re doing our polling yet, but we’re very much aware and concerned about how different voting rules or laws may actually impact who shows up.”
Eran Ben-Porath, executive vice president of Public Opinion Research at SSRS, a research and polling company, told Fortune that experts are still trying to figure out how to adapt exit polls to this new reality.
“There’s a lot of discussion about that, and there’s going to be a lot of conditioning to the idea that it’s reasonable that we won’t know the election results in certain tight states for a while,” he said. “People tend to be restless on election night and want to know the results, but there’s a more than decent chance that election night will leave us with more than a few questions.” Predicting elections before a significant number of votes have been counted is bad business, he said.
In terms of predictive polling, Jackson warns that Joe Biden’s 10-point lead over Trump may not indicate much either. “It’s really impossible to forecast any of what we’re seeing right now forward because people are moving so quickly on a lot of these issues,” he said. He also called attention to undecided voters, who tended to vote in favor of Trump in 2016.
The latest Quinnipiac poll, for example, put Biden eight points ahead of Trump, but if the voters who are still undecided are factored in as voting for Trump, that lead narrows to three points, which is within the margin of error. If voters who say they’ll be voting for another candidate are also swayed in Trump’s direction, then Biden and Trump are in a dead heat. In volatile times, Jackson noted, it’s more difficult for pollsters to predict where undecided voters may end up.
Biden’s lagging decision on a vice presidential pick also makes it difficult for pollsters to project, said Buskirk. He expects to see a shift in numbers once a VP announcement is made, likely before August.
Still, Americans have been stuck at home, and that typically translates into good news for pollsters. Buskirk says that since the pandemic reached the United States, he’s seen a 150% to 200% increase in survey response rates. This change increases accuracy and also gives those conducting polls a better return on investment.
“We’ve had an easier time doing interviews since the shutdown started because people are home. They’re not going to work or going out,” said Jackson. “They can answer the phone and have free time and are willing to talk to us. It’s made it easier to do our jobs.”
People are bored and want to connect with others, said Ben-Porath. But he noted that they’re also excited to talk about what they see happening in their country. “There’s interest in participating right now because these are interesting times,” he said.
