Election watchers may have been surprised that former Vice President Joe Biden won the presidency with narrow margins in a handful of battleground states, after national polls indicated the Democratic nominee held a comfortable advantage over incumbent President Donald Trump for much of this year.
But with the polls appearing to have been off by at least 3 percent, for the second straight presidential election, national polling organizations have begun diagnosing the latest inaccuracies and are even confronting questions about the value of public polling itself.
New Hampshire pollsters, for their part, say they are confident in the industry’s future, despite a host of new challenges, and pointed to their own success forecasting the 2020 race.
In a poll published Oct. 29 by the Saint Anselm College Survey Center, part of the N.H. Institute of Politics in Manchester, Biden led Trump in New Hampshire by 8 percentage points, 52 percent to 44 percent. The same poll also indicated that incumbent Gov. Chris Sununu, a Republican, and U.S. Sen. Jeanne Shaheen, a Democrat, held 25-percent and 15-percent advantages, respectively, over their opposite-party challengers.
And an Oct. 29 poll by the UNH Survey Center in Durham showed Biden leading Trump by 8 percent among Granite State voters, 53 percent to 45 percent, Sununu ahead of state Sen. Dan Feltes, D-Concord, by 24 percent and Shaheen up on Bryant “Corky” Messner, a Wolfeboro Republican, by 11 percent.
Those results were largely borne out in the Nov. 3 election, with Biden defeating Trump in New Hampshire by more than 7 percentage points, according to results from the N.H. Secretary of State’s Office. Sununu and Shaheen won re-election with 32-percent and 16-percent margins, respectively.
Neil Levesque, executive director of the N.H. Institute of Politics, praised the center’s Oct. 29 poll for its accuracy and criticized some national media organizations for excluding NHIOP polling from their pre-election polling averages. Several outlets required certain weighting criteria — pollsters’ adjustments meant to reduce the effect of known inaccuracies — that the NHIOP believed would produce unreliable survey results, Levesque explained.
“As a result, they are putting their own bias into these polls, and they’re getting it wrong,” he said. “[The problem] is not the data that’s collected — it’s how you weigh the poll.”
Levesque argued that one error by the national media was to overestimate participation by young voters, which he said favored Democratic candidates in the polls due to liberal candidates’ popularity with that demographic.
National turnout among eligible voters ages 18 to 29 is estimated to have risen by approximately 10 percent this year from 2016 and comprised 17 percent of the electorate, up slightly from four years ago, according to the Center for Information and Research on Civic Learning and Engagement at Tufts University in Medford, Mass.
Granite State pollsters incorporated lessons from the 2016 presidential election — when many polls, especially at the state level, showed little chance of a Trump victory — into their methods this year.
Andy Smith, director of the UNH Survey Center, attributed the past inaccuracies to many Trump voters’ opting not to participate in telephone surveys, which he called “selective non-response.” That behavior, in addition to misleading responses from Trump voters who did participate, was a product of their aversion to associate publicly with a candidate widely perceived as unpopular, he explained.
In 2016, selective non-response caused polling organizations like the UNH Survey Center to overestimate support for former Secretary of State Hillary Clinton, the Democratic nominee, according to Smith.
“If you don’t have people who are willing to be honest or are willing to talk with you, it doesn’t matter how you weight the sample [or] how you word the questions,” he said.
Smith explained that the center’s decision during the recent election cycle to conduct polls via online form, rather than telephone, helped eliminate non-response bias because respondents are less prone to social pressures in that format. Several national pollsters, including Pew Research Center, have also adopted web-based polling in recent years.
Levesque, the N.H. Institute of Politics executive director, argued the polling error in 2016 was largely due to pollsters’ failure to identify likely voters, in addition to some voters’ decision to not participate in surveys. He said pollsters often did not contact many eventual Trump voters because those voters had not regularly cast a ballot in prior elections.
“Donald Trump communicated with people that normally didn’t … vote and who were motivated by his message,” he said. “They went out and voted, and they were missed by pollsters.”
The NHIOP improved its survey accuracy in the 2020 election cycle by canvassing more of that group, using a state database to identify people who voted in 2016 after having been infrequent voters or non-voters, Levesque said.
He added that NHIOP polls were conducted via text, rather than telephone call, beginning in late March as part of its own efforts to improve polling accuracy. Levesque said the organization may continue to use that method because it generated larger, and thus more accurate, samples.
Despite the recent improvements, Smith and Levesque identified several challenges that continue to plague public polling.
Telephone surveys have become increasingly expensive as their response rates have dropped significantly, forcing pollsters to spend more time gathering a sufficiently large sample, according to Smith. He explained that a statewide poll in New Hampshire now costs more than $75,000 after being about one-third of that price four years ago — another reason why the UNH Survey Center adopted web-based polling for the latest election.
Smith, a member of the American Association for Public Opinion Research’s election polling task force, said the group plans to study the results of 2020 polls and identify ways to make future surveys more accurate.
Levesque expressed concern that future political candidates and campaigns will instruct their supporters to decline to participate in surveys or to lie to polling organizations. He also said that like media outlets, pollsters may be forced to confront a polarized electorate that dismisses as fraudulent any survey results seen as unflattering for their preferred candidate.
Polling inaccuracies, which many experts claimed had been addressed after the 2016 election, in addition to those continued challenges, have led some to ask whether public polling is a worthy undertaking at all.
New York Times pollster Nate Cohn acknowledged those concerns on the Times’ podcast, “The Daily,” last week, arguing that polling informs people about the beliefs among voters outside of their own social circles. Still, Cohn acknowledged that continued inaccuracy would take “a toll on people’s trust in institutions” and could undermine the credibility of polling organizations’ other political coverage.
However, Levesque rejected the possibility that inaccuracies in recent elections and possible credibility issues will reduce the appetite for public polling.
“I think polling is going to be around for a long time,” he said. “… I think that there’s an urge by the public to know who they think is going to win.”