2020 polls persistently understated support for Trump and GOP, study finds
WASHINGTON (SBG) — Public opinion polls consistently misread support for former President Donald Trump and Republican candidates in the run-up to the 2020 election, a new report found, as pollsters encountered new challenges that resulted in some of the most inaccurate polls in decades, and experts are unsure what to do about it.
A task force convened by the American Association for Public Opinion Research found the 2020 election produced the widest polling error in the national popular vote in 40 years and the widest error in state-level polls in the 20 years for which performance has been tracked. The errors generally favored Democrats, overestimating support for President Joe Biden and underestimating support for Trump, compared to final election results.
In the last two weeks before the election, national polls overstated Biden’s lead by an average of 3.9 points and state-level polls were too favorable toward Biden by 4.3 points. For senatorial and gubernatorial races, support for Democratic candidates was overestimated by an average of 6 points.
“Whether the candidates were running for president, senator, or governor, poll margins overall suggested that Democratic candidates would do better and Republican candidates would do worse relative to the final certified vote,” the report stated.
As a result, Democrats lost several Senate races they expected to win, Republicans made unanticipated gains in the House, and Biden’s victory over Trump was narrower than many projected. The authors of the report were able to rule out a repeat of many of the mistakes pollsters made in 2016, but nailing down precisely what went wrong this time was more difficult.
A handful of national polls put Biden ahead of Trump by double-digits, but most showed his lead between 4 and 8 points, not far from the 4.5 points by which he ultimately won the popular vote. Some state polls were much further afield, including one that gave Biden a 17-point edge in Wisconsin, which he won by less than 1 point, and several that showed him ahead in Ohio, where Trump won by 8 points.
“The core of the problem is, Trump is not pollable accurately,” said Michael D. Cohen, CEO of Cohen Research Group and author of “Modern Political Campaigns.”
Some Senate polls were way off-base too, potentially signaling deeper issues. Republican Sen. Joni Ernst comfortably won reelection in Iowa after several polls suggested her race was a toss-up, and none of the public polls conducted in Maine all year showed GOP Sen. Susan Collins ahead in a race she went on to win by more than 8 points.
Although Trump’s unexpected victory in 2016 left many observers frustrated with public polls, the average of national surveys in the final weeks of the race similarly came pretty close to predicting Democratic nominee Secretary Hillary Clinton’s popular vote edge over Trump. State-level polls, which were conducted less frequently and used smaller samples, were much less reliable.
Polling firms responded to those missteps by weighting their samples by education, which appeared to have helped produce more accurate results in the 2018 midterms. In 2020, though, historically wide errors steadily occurred across all types of polling in both presidential and down-ballot races.
The report dismissed the possibility that some factors that contributed to inaccuracies in 2016 affected 2020 surveys, such as late-deciders breaking toward Trump or respondents being reluctant to admit they supported him. Exit polls indicated those who decided in the last week of the election were as likely to vote for Biden as Trump, and the error was higher for the state polls that did not involve Trump than for those that did.
The authors determined it was “impossible” to reach definitive conclusions based on the available data, but one hypothesis they presented was that Republican voters who generally respond to polls and Republican voters who do not respond behave differently. Republicans who are willing to talk to a mainstream pollster might be more amenable to voting for a Democrat than those who dismiss polls as “fake news.”
“The Trump base, which is highly distrustful of polling and institutions writ large, has kind of systematically different opinions and are not responding to polls,” said Lara Brown, director of the Graduate School of Political Management at George Washington University.
If that is the case, even a poll that includes a sufficient share of self-identified Republicans might not accurately reflect the views of Republicans who turn up to vote on Election Day. According to Ashley Koning, director of the Eagleton Center for Public Interest Polling at Rutgers University, the recent polarization surrounding polls has created new obstacles for compiling a representative sample of the electorate.
“An issue with 2020 seems to have been not missing or underestimating Republicans, overall - as seen in 2016, especially in regard to missing the non-college-educated white vote - but perhaps interviewing the ‘wrong’ kind of Republicans,” Koning said.
Cohen noted the AAPOR report was written primarily by academics and journalists, with input from only a few people who conduct election polling professionally. Several polling firms got the final popular vote margin between Biden and Trump right, and they could have offered valuable insight.
“They didn’t talk to the people who actually nailed it,” he said.
One of the biggest unanswered questions left by this report is how much of the error was specifically driven by Trump, who regularly denounced polls that did not show him leading as illegitimate. Trump’s Save America PAC blasted out a link to a Newsmax story on the study Monday night, and his allies will likely see it as proof he was right to claim media surveys systematically underreported his support.
"If it's bad, I say it's fake. If it's good, I say that's the most accurate poll ever," the former president said at the Conservative Political Action Conference earlier this month.
Trump has been noncommittal about his 2024 plans, as he positions himself to play a major role in House and Senate races in 2022. If he runs for president again, it might not be fully clear for several years whether he has permanently altered the Republican electorate in ways polls cannot easily measure.
“We do know that the 2018 midterm polling - where Trump was not at the top of the ticket - was highly accurate,” Koning said. “Without Trump at the top of the ticket in 2022, it could be a cleaner test that enables pollsters to dig deeper into the challenges polling currently faces.”
Polling errors in presidential contests have not been uncommon historically, and pollsters have struggled to adapt over the last 20 years as Americans moved away from landline phones. This report suggests they still have considerable work to do to account for modern communication preferences and political polarization.
The persistent problems with candidate polling do not necessarily cast all polls into question. An analysis released earlier this year by the Pew Research Center found even a large overrepresentation of support for Biden did not significantly impact measures of opinions on issues.
Polls rarely offer the level of certainty that the media and the public tend to read into them, and factoring in the margins of error can result in a more accurate but less definitive understanding of a race. Among 2020 polls where the leading candidate’s advantage was greater than twice the margin of error, 98% of state-level presidential polls and 90% of senatorial and gubernatorial polls predicted the correct winner.
“It does what it says on the box,” Cohen said. “There is a margin of error, and those national numbers were not outside that margin of error.”
Democrats, who went into Election Day expecting a more decisive victory, have also been trying to figure out what went wrong with public and private polls last fall. A coalition of top Democratic polling firms conducted an autopsy to address “major errors” in their data, but they reached no consensus on rectifying them.
While Democrats adjusted demographic models to reflect high support for Trump among white, non-college-educated voters, they found their estimates were still low because turnout among low-propensity Republican voters in 2020 was four times higher than among Democrats. They also considered the possibility that people who stayed home due to COVID-19 last summer and fall—and were therefore more accessible for polling—were more likely to lean left.
The Democratic consultants ultimately reached a similar conclusion to the AAPOR task force, that the people who were not polled are very different from those who were. However, they cautioned against overcorrecting for that phenomenon because it might be specific to Trump.
“What we have settled on is the idea there is something systematically different about the people we reached, and the people we did not,” their memo stated. “This problem appears to have been amplified when Trump was on the ballot, and it is these particular voters who Trump activated that did not participate in polls.”
As Republicans and Democrats prepare for the 2022 midterms and prospective 2024 candidates weigh launching their campaigns, the reliability of polls will soon become a pressing issue for strategists, pundits, and journalists attempting to understand the views of voters. After the last two presidential races, experts say they might need to rely more on what they learn from people on the ground than on raw numbers.
“Polling didn’t exist prior to 1940 in any sort of meaningful way,” Brown said. “What we’re going to have to do to a certain degree is rely on some of the more traditional understandings.”
Despite the apparent shortcomings of public polling, it remains the only scientific way to analyze broad public sentiments regarding issues and candidates. If properly understood and interpreted, Koning said surveys can still offer valuable insights as pollsters work to improve their precision.
“We can’t throw the baby out with the bathwater just yet,” she said. “What we need to do going forward is recognize what polls should and should not be used for. They are snapshots in time and do a much better job of explaining what the public was thinking and feeling in hindsight than serving as a crystal ball.”
