Opinion polls underestimated the level of Donald Trump’s support for the third straight presidential election, predicting a neck-and-neck race this year with Democrat Vice President Kamala Harris when, in the end, Trump held sway over her in battleground states.
Trump’s victory involved surging support in a number of demographics and regions, but analysts said pollsters failed to accurately predict races in states where the results differed significantly from the 2020 election, which Trump lost to Joe Biden.
“They did fine in battlegrounds, but … they failed to provide the essential information that Trump was surging across the board,” said Michael Bailey, a professor of political science at Georgetown University.
More than 90% of U.S. counties voted in higher numbers for Trump than they did in 2020, according to The New York Times.
Overall, the polls predicted razor-thin margins in races in the seven battleground states that decide close elections. As of Wednesday, Trump was projected to win five of those states by between 1 and 3 percentage points.
The former president was well on his way to sweeping all seven states, according to those projections.
“Trump may have been mildly underestimated but I think the polls ended up doing pretty well, collectively – this was not a huge miss,” said Kyle Kondik, a political analyst at the University of Virginia. “The polls suggested Trump had a decent chance to win, and he won.”
The pollsters’ performance was under the microscope this year, after two big misses in succession: they failed to anticipate Trump’s victory in 2016 and overestimated the margin by which Biden won against him in 2020.
“Trump was underestimated by about 2 points this time around” in key states, said Pedro Azevedo, head of U.S. polling at AtlasIntel.
In Pennsylvania, the latest polling average from RealClearPolitics put the Republican in the lead by 0.4 percentage points. As of Wednesday, he was ahead by 2 points.
In North Carolina, polls predicted a 1.2-point margin for Trump, and he won by 3 points over Harris.
In Wisconsin, Harris was given a 0.4-point lead, but the projected results showed Trump leading by 0.9 points.
The main problem has not changed since Trump’s arrival on the political scene about a decade ago: a fringe of his electorate refuses to take part in opinion polls, and firms have failed to be able to accurately gauge their impact.
In the most recent polls conducted by The New York Times with Siena College, “white Democrats were 16 percent likelier to respond than white Republicans,” data analyst and polling guru Nate Cohn wrote two days before the election.
That disparity had grown over the course of the 2024 campaign, he added.
Although pollsters like The New York Times/Siena tried to compensate for these flaws with statistical adjustments, it was clearly not enough.
“It is apparent that polls significantly underestimated Trump’s growth among Hispanic voters,” said Azevedo, pointing to Trump’s larger-than-expected victories in Nevada and Florida.
“This is also the case among white voters,” he said, adding that although most polls expected Harris to “improve her margins” in this demographic, Trump outperformed the polling and ran up his numbers in rural areas.
Iowa was a prime example of this, with a poll three days before Election Day giving Harris a 3-point victory in the solidly Republican state. In the end, Trump won it comfortably by 13 points, Azevedo said.
J. Ann Selzer, the author of that inaccurate Iowa poll, said the difference could have been made by late-deciding voters.
“The late deciders could have opted for Trump in the final days of the campaign after interviewing was complete,” she told the Des Moines Register. “The people who had already voted but opted not to tell our interviewers for whom they voted could have given Trump an edge.”
