With less than a month until Election Day, the nation is mining for any sliver of information that could provide insight into who will come out victorious in November.
For many, that may mean poring over scores of polls.
Surveys over the last few weeks have generally shown former President Donald Trump and Vice President Kamala Harris neck and neck in national and battleground states. Polling analysis website FiveThirtyEight’s average of national surveys currently has Harris with a slight advantage, just more than 2.5 percentage-points, over Trump.
In the first few days of October in 2020, on the other hand, President Joe Biden’s lead over Trump in 538’s national average ranged from 7.4 percentage points to as high as 8.2 %. A CNBC/Change Research poll released on Oct. 6, 2020, had Biden with solid leads outside the margins of error in six key swing states at the time. That survey showed Biden winning in Florida, which is now considered to be a reliably red state, and North Carolina, both of which he lost to Trump.
In 2016, FiveThirtyEight had the Democratic nominee that year, former Secretary of State Hillary Clinton, with a 3-4% edge at the start of October, while an NBC News poll released on Oct. 4, 2016 had her up by six points nationally.
In a presidential contest that has been marked by a flurry of unprecedented events – including an incumbent president dropping out less than four months from the election and two apparent assassination attempts against a candidate – one constant of the last two cycles remains: Trump as the GOP nominee.
And while it may be tempting to look at where Harris' polls numbers stand against the former president compared to the previous two Democratic candidates at this point in 2016 and 2020, experts urge caution in doing so -- in part because pollsters have changed the way they conduct surveys in recent years.
“The public and the news media at large should be treating the polls with some wariness because, although they are signaling a close race and have been signaling a close race for quite some time, there is some variation among the polls and who does them and what methodologies they use,” W. Joseph Campbell, a professor emeritus at American University said.
Campbell, who is also the author of Lost in a Gallup, a book about polling failure in presidential elections, noted there is still time for developments to impact the trajectory of the race – particularly in a cycle that has been, as he put it, “extraordinarily unpredictable.”
Not to mention, he added, the nation has been burned by polls before.
“The polls in 2016 and 2020 were among the least accurate in many, many years,” said Scott Keeter, senior survey advisor at Pew Research Center, a nonpartisan organization that, among other things, conducts public opinion polling. “In both years, support for Donald Trump was underestimated.”
In the days before the election in 2020, national polls had Biden ahead by as many as 10 and 12 percentage points against Trump. The average of national polls by 538 had Biden with an 8.4 percentage point advantage on the day before the election.
He ended up winning the popular vote by about 4.5 percentage points and the presidency with a 306 to 232 electoral college victory, notching wins in Pennsylvania, Michigan, Wisconsin, Arizona, Nevada and Georgia.
“It’s a clear victory in the popular vote but nowhere near the landslide that some well-known, high-profile pollsters were anticipating,” Campbell said of Biden’s 2020 numbers
As for 2016, NBC News’ final poll of the election cycle had Clinton maintaining a six-point lead over Trump. Two separate polls, one by CBS News and another by ABC News/Washington Post, in the final days each put Clinton four points ahead of Trump.
In the end, the former secretary of state and first lady won the popular vote by about 2.1%, but lost the White House with an Electoral College defeat. Trump received 306 electoral college votes to Clinton’s 232.
Keeter noted that while the midterm election polls in 2018 and 2022 – when Trump wasn’t on the ballot – turned out to be more accurate, it has not eliminated concerns that surveys this year could be off.
“Because the pattern appears to be when Donald Trump is on the ballot, polls seem to have a hard time getting the appropriate share of his supporters,” he said.
Campbell referred to the question of whether the polls will once again underestimate support for Trump as the “major known unknown of this campaign.”
“Pollsters have a lot of trouble getting access or getting to, getting responses from Trump supporters,” he said. “In 2020, they thought they had made the corrections and adjustments that threw them off in 2016 and that proved not to be the case.”
Keeter noted the “demographic profile” of many of those who back the former president – such as non-college educated voters who may also be more likely to have lower levels of trust in institutions and thus be less likely to answer polls by major news organizations – is what has made reaching Trump supporters difficult.
“Now pollsters are aware of this and they’re trying new things to see if they can reach a more complete set of voters than they did in 2016 and 2020,” he said. “In particular, our research found that a majority of polling organizations that polled in 2016, were using different methods in 2022, in the midterm elections, than they had in 2016 or 2020.”
He noted that polling by telephone is much less prevalent, with pollsters now conducting surveys mainly only and using methods such as text messages.
Keeter said Pew Research Center’s own methods have changed “pretty substantially” compared to three cycles ago. In 2012, he said, all of their polling was done by telephone. Now, their survey panel responds mainly online, although he noted Pew also has a telephone to ensure it is not missing people who are not comfortable with technology.
Despite the lessons and changes, how effective it all turns out be this year, Keeter said, is “really anybody’s guess.”
“The dynamics of every presidential election are different, they’re never the same and the same holds for polling” Campbell added.