Pollsters face another season of soul searching after 2020 presidential race

Polls aren't dead.  

But they're sure in trouble after pre-election surveys created an expectation of a blue Democratic wave and an emphatic Joe Biden victory over President Donald Trump.

Instead, multiple battlegrounds, including Wisconsin, went down to the wire, and big Democratic gains forecast in Congress didn't materialize.

It was a miss, a big one, and it comes on the heels of the polling failure of 2016 when Trump upended the polls and shocked the pundits to claim the White House.

Marquette University Law School poll director Charles Franklin said, "The patient is on the emergency room table."

"You notice I did not say we are on the autopsy table," Franklin added. "I think the patient can still be helped."

Wisconsin, ground-zero in presidential politics, also was a center for polling's crisis of confidence.

Four years ago, every public poll had Hillary Clinton beating Trump in Wisconsin. But they all missed Trump's late surge to win the state.

This year, the final pre-election polls showed Biden leading in the Badger State but by various margins.

At one end, Marquette showed a 5-point Biden lead among likely voters, while at the other end, an ABC News-Washington Post survey showed a 17-point Biden lead.

The Elections Research Center at the University of Wisconsin-Madison pegged Biden's lead at 9 points, while the New York Times-Siena College Poll showed Biden ahead by 11 points.

Unofficial results showed Biden winning the state by less than 1 percentage point. The Trump campaign said it will request a recount.

Just like what occurred after 2016, the polling industry is headed for a period of introspection.

The leading association of pollsters is warning against a "rush to judgment on the 2020 polls" and urging people to wait until final election numbers are fully analyzed.   

The American Association for Public Opinion Research is convening a task force of pollsters, statisticians and academic experts to evaluate the general election polls.  

'Not a magic ball'

"At the end of the day, a lot of pollsters will be doing some soul searching," said task force member D. Sunshine Hillygus, a political science professor and public policy director of Duke University's Initiative on Survey Methodology.  

She said pre-election polls provide useful information, "but they are not a magic ball."

"I personally think pre-election polls undermine the entire industry and we ultimately expect too much of them," she said. "We as an organization as survey researchers need to figure out how to do a better job of telling the public what exactly a pre-election poll can and cannot tell us.

"There is a lot of false precision in those pre-election polls. That's simply because we don't know who is going to show up to vote. Pre-election polling rests on a set of assumptions that ultimately are based either on the best guesses of the pollsters or on the history of previous elections."

Franklin, whose Marquette poll interviews voters by cellphones and landlines, has built a healthy local and national following. His poll releases are eagerly anticipated by the media and those who closely watch politics.

"Weather forecasting is something everyone complains about and has its misses," he said. "But everybody watches the weather forecast. There is an element of that to the polling industry. We are going to have errors and mistakes and the sunny days that are going to turn around and rain. It's still something people are concerned about and want to know about."

Franklin said that in 2016 and 2020 the Marquette survey provided an accurate snapshot of the Democratic vote but under-represented the Trump vote.

"Whether it's shy Trump voters or Trump voters that declined to be interviewed in the first place, or maybe something else dealing with turnout, it's striking that we just don't get many Trump voters as he's actually getting in the vote," Franklin said.

Franklin said while presidents complain about polling, "President Trump is somewhat unique in the lengths he goes to describe fake polling, suppression polling." That message, Franklin said, may make the president's followers suspicious of polling.

"The irony is what that does is, if it keeps a small percentage of his supporters from responding to polls, as a consequence it makes him look like he's doing a little bit worse from what he ends up doing," Franklin said. "The rhetoric may lead to a little bit greater reluctance to tell us what their opinions are and choices are in relation to Trump voters."

Franklin said pollsters have to figure out, "are we in the way we're sampling people missing Trump people. I think that's less likely but not completely off the table."

Barry Burden, founding director of the Elections Research Center at the University of Wisconsin-Madison, acknowledged the difficulties of trying to track Trump's level of support.

"There is something amiss in these two elections where Donald Trump is on the ballot," he said. "And it's not clear what that is." 

After the 2016 election, Burden said, some pollsters made "some tweaks to their procedures," including accounting for education.

Burden anticipated the environment in 2020 "would make the polling more accurate."

"Compared to 2016, there were fewer undecided voters, fewer third-party voters," he said. "There was an incumbent on the ballot who people had strong feelings about. And there were lots of people voting early, so as they were being interviewed they had already cast ballots."

Burden said the "miss" is bigger in some states than others.

"In Wisconsin the miss looks pretty big again as it was four years ago," he said. "Next door in Minnesota actually the polls got it about right. That adds to the mystery of what's happening."

The Elections Research Center poll is a panel survey, where the same voters are interviewed several times over the course of the election year.  

Some voters drop off and are replaced by others.

Burden said around two-thirds of the original 800 people participating in Wisconsin made it through the entire election season.

"Of the people who stayed in our panels, they were much more balanced in their vote division between Biden and Trump," he said. "Their votes are pretty close to the statewide results. The new respondents we added were much more likely to be Biden voters."

Burden said in the coming months, he and others involved in the poll will be "doing a lot of reverse-engineering" to see "what could have been done differently."

"Polls aren't going anywhere," he said. "They have a lot of value outside the election-prediction business for providing a look at the electorate, taking the temperature of the public on issues, their attitudes toward politics, the whole variety of things that aren't just about voting."