The approach of the midterm elections is renewing doubts about the accuracy of polling, after the predictions of strong Democratic Party support before the last two presidential contests turned out to be wrong.
Most polls failed to predict former President Donald Trump’s win in 2016 but did better with the 2018 midterms. Then in 2020, they underestimated Mr. Trump’s portion of the vote by an average of more than 4 percentage points and overestimated how well Democrats would do in House and Senate races.
“The standard assumption that people generally tell the truth is what enables polls to be mostly right, but undercounting of certain groups, including those skeptical of pollsters, a group that skews conservative, can lead to flawed results and conclusions,” said Renee Van Vechten, a political science professor at the University of Redlands in California.
Experts say conservative voters’ mistrust of polls could again skew election polling leftward before November. Since Republicans lead Democrats in generic congressional ballot surveys by about 4 percentage points, even a slight undercounting of GOP votes could mean a bigger-than-expected drubbing for President Biden’s party at the ballot box. Historically, the political party in the White House loses seats in Congress during the midterm elections.
Ms. Van Vechten expects national pollsters will work harder this year to reach potential nonparticipants, using weighting and other statistical techniques “to compensate for their recent inability to achieve a truly representative sample — and will also try to salvage their reputations.”
“The answer is that it all depends on what kinds of polls are in question — such as state or national, and how wide the margin of error is — and whether the pollsters weight the results to account for an undercount of Republicans,” she said.
But Robert Cahaly, who founded the Trafalgar Group in 2016, says many pollsters have yet to realize that Trump voters lie to them as easily as they lie to their doctors. He says that causes pollsters to overestimate turnout from Democrat-leaning voting blocs, underestimate Republican turnout or “just pick the results they want.”
“It’s much more of an honor system than I’m comfortable with,” Mr. Cahaly said. “And when they all get it wrong, everyone seems to think that’s OK.”
Nate Silver’s website, FiveThirtyEight, awarded Trafalgar an A-minus rating for accuracy, noting it had the second-lowest error rate (2.6 percentage points) of any polling firm in its 2020 state-by-state surveys. Trafalgar puts more emphasis on anonymous text-message surveys that take up to 3 minutes than on conventional landline telephone calls, longer surveys and questions requiring detailed personal information.
Mr. Cahaly, who accurately predicted the 2016 presidential election, says he prefers to get predictions right rather than tell his side — the Republican Party — what it wants to hear. To that end, he believes long robocalls and panel focus groups are a waste of time.
“Who is going to answer 35 or 100 questions at 6 o’clock on a Tuesday night? The most common question people have is ‘how long will it take’ and they hang up if it’s longer than 5 minutes,” he said. “It makes me laugh that some pollsters still think they can call a home line and get anyone under 50 on the phone.”
James Carville, a longtime Democratic Party political strategist, said more pollsters will rely on cellphones this year because “people don’t answer their home phones as much.”
“It all used to be landline calls,” Mr. Carville said. “Clearly in 2016 and 2020, they had trouble getting noncollege Whites in some places, and some good polls missed Wisconsin by a mile. So the polling industry got a black eye.
“But the turnout in 2020 was breathtaking, and the most important two things to watch are trends and averages, so I wouldn’t take the polls as gospel.”
Most polls got 2020’s final presidential and congressional votes wrong by even bigger margins than in 2016, despite working to improve.
The nonpartisan American Association for Public Research Opinion reported on July 19 that Trump voters’ distrust for polls drove the margin of error in the 2020 general elections to be “the highest in 40 years for the national popular vote and the highest in at least 20 years for state-level estimates of the vote in presidential, senatorial and gubernatorial contests.”
In the last two weeks of the 2020 campaign, presidential polls inflated Mr. Biden’s support by 3.9 points in the national polls and 4.3 points in statewide polls above the final popular vote. He beat Mr. Trump 51.3% to 49.6%.
“The polling error was much more likely to favor Biden over Trump,” a summary of the report stated.
In Wisconsin, where Mr. Biden won the state’s electoral vote by 0.63% margin over Mr. Trump, the polls were especially inaccurate. One ABC News/Washington Post survey conducted on Oct. 20-25 gave Mr. Biden a 17-point lead over President Trump in Wisconsin, though most gave Mr. Biden a lead in the high single digits.
Ideological factors also keep many Republican voters from responding to election surveys, according to some conservative media watchdogs.
L. Brent Bozell, founder of the conservative Media Research Center, says some polls make conservatives feel guarded by inserting “woke language” about topics such as gender and race.
“The stilted phrasing pollsters use can make it difficult for conservatives to respond,” Mr. Bozell said. “They are going to be asked about ‘undocumented immigrants’ and not ‘illegal immigrants.’”
Adam Guillette, president of Accuracy in Media, says some conservatives may conceal their true views from a hostile-sounding survey to give the polling firms “data that reinforces their worldview.”
“Polling companies have become the political equivalent of Lucy snatching the football away from Charlie Brown,” Mr. Guillette said. “Each and every election we learn how wrong they are, how biased they are. Then two years later, they expect us to take them seriously again.”
But it’s also possible the polls will be more accurate without Mr. Trump in the race this fall since they were only half-a-point more favorable to Democrats than the final vote tally of 2018’s midterm elections. Without Mr. Trump in the race, conservatives may be more honest about who they will vote for.
But even though the polling might be skewed, the surveys are mostly accurate about people’s opinions on social issues, according to the Pew Research Institute.
In March, Pew reported in an analysis of its survey questions from 2020 that a hypothetical poll overestimating Mr. Biden’s support by 12 percentage points did not change Americans’ positions on key social issues by more than 1 percentage point. Skewing the poll toward Mr. Biden failed to significantly change how people responded to topics ranging from Black Lives Matter to immigration.
“Opinions on issues and government policies are strongly, but not perfectly, correlated with partisanship and candidate preference,” Pew said in a summary. “A minority of people who support each candidate do not hold views that are consistent with what their candidate or party favors.”