Frank Luntz is a noted Republican political pollster. In the days before the Nov. 3 election, he said if the 2020 pre-election polls were wrong again, “my profession is done.”
In the days after the election, Luntz said the results are “devastating for my industry.”
“The polls were really bad this time and really misleading, and I'm not sure why,” said Elaine Kamarck, a senior fellow at the Brookings Institution.
In 2016, the pre-election polls of individual states were off quite a bit, showing Democrat Hillary Clinton ahead of Republican Donald Trump in key states that Trump eventually won. Pollsters failed to adequately account for education levels and underestimated Trump’s support among whites without a college education.
Undercounting Trump’s support
Pollsters acknowledged the shortcomings and said they adjusted how they weigh education levels for the 2020 election. But there appears to have been a similar undercount of Trump supporters during the run-up to the latest election.
“We are missing some small set of Trump supporters, who are not necessarily Republicans, but who are declining to take part in the survey systematically," said Charles Franklin, director of the Marquette Law School Poll in Wisconsin.
In 2016, the Marquette Poll had Clinton up 6 percentage points in Wisconsin. Trump won by 1 percentage point. “This time we have the right winner, and we're off by 3 or 4 depending on turnout," Franklin explained.
Biden won Wisconsin by 20,000 votes, a 0.63 percentage point lead over Trump. The Marquette Poll showed Biden 5 percentage points up on Trump a week before the election.
Key states that were off the pre-election polling averages by 5 or more percentage points were Wisconsin (7 percentage points), Iowa (7), Florida (6), Michigan (5), Ohio (5) and Texas (5). Trump won four of those states – all but Michigan.
Nate Silver, editor in chief of FiveThirtyEight, a website that analyzes polling data, pushes back on the notion that the polls were wrong, writing that polls accurately called 48 of the 50 states along with the winner of the Electoral College and popular vote, Joe Biden.
There is an underlying possibility of error in any poll. Most of the polls used in the presidential polling averages allowed for a polling error of 3 to 4 percentage points either way. Silver says polling’s 3 to 4 percentage point underestimate of Trump’s vote is within historical standards, and the public has to adjust expectations that “demand(ed) an unrealistic level of precision” of polling.
“The polls set expectations too high for Democrats,” said Bill Schneider, professor at the Schar School of Policy and Government at George Mason University. Schneider, a longtime political analyst for CNN, said Democrats “ended up doing worse than expected.”
Many experts point to declining response rates to telephone polling as a problem in getting it right. They are also finding that Democrats are likelier to respond to polling inquiries than Republicans.
For campaigns that depend on polling to make decisions about resource allocation, “they have to go back to an old-fashioned way before we had polling,” said Kamarck, who is also a member of the Democratic National Committee. She said having people on the ground to canvass “precinct by precinct, county by county, state by state” may be the way voter research is conducted in the future.
“The state-by-state polls that they did this time in order to make up for the problems they had four years ago really have been wildly off,” Kamarck said.
And regarding Frank Luntz’s prediction about his profession if there was a repeat of 2016? “Frankie may be out of business,” Kamarck remarked.