Donald Trump outperformed expectations for his third straight presidential election, which will surely raise more questions about pollsters’ ability to gauge where elections stand.
Trump pulled off a sweep of the main battleground states over Vice President Harris in the election Tuesday, and appears set to win the popular vote even as polls showed a neck-and-neck race throughout much of the campaign. He also made considerable inroads in comfortably blue states, losing some of them by smaller margins that Republicans have previously.
The results are another drop in the bucket for how the Trump era has rocked faith in the polling industry.
Still, pollsters maintained that the outcome the election produced was within what the polls suggested was possible.
“The margin gets amplified because it’s an all-or-nothing Electoral College. If we had a proportional Electoral College, then it would probably reflect it,” said David Paleologos, the director of the Suffolk University Political Research Center.
“It would be closer, but because it’s an all-or-nothing Electoral College, a half point win in a big state with a lot of electoral votes is going to give the illusion of a much bigger margin,” he continued, a reference to Trump’s 312-226 win in the electoral vote.
Compared to the past two elections, the final polling averages in the key states weren’t too far off.
Polling seemed to be closest to accurate in the three “blue wall” states of Pennsylvania, Michigan and Wisconsin, states that if Harris had won would have given her the presidency. The final averages from Decision Desk HQ/The Hill had Trump ahead by a few tenths of a point in Pennsylvania and Wisconsin and Harris about the same amount ahead in Michigan.
Votes are still being tabulated, but Trump’s margins in these states were actually quite close. He is currently ahead in Pennsylvania by 2, Michigan by 1.5 and Wisconsin by 1.
While Georgia and North Carolina were called for Trump first among the battlegrounds, Trump ended up winning them narrowly. He leads in Georgia by about 2 points; DDHQ/The Hill had him ahead by 1.7 in the final average, and FiveThirtyEight had him leading by about 1.
Trump’s over-performance in North Carolina and Nevada was a bit more — but only slightly. He currently leads both by just over 3 points, just over the roughly 1.5-point margin from DDHQ in each, but still within a normal margin of error.
Arizona appears to be the biggest difference, with a 6-point Trump lead compared to the DDHQ average of 2.5 and FiveThirtyEight average of 2.1.
Jim Lee, the president and CEO of Susquehanna Polling & Research, said pollsters weren’t “perfect” but did a “pretty good job” capturing that the race was close with a slight edge to Trump. He pointed to RealClearPolitics showing Trump ahead in the average for five of the seven states.
Going into this cycle, pollsters were well aware of the challenge facing them after the last two elections. Trump’s political career had been marked by outperforming polling and even more so public expectations.
In 2016, it carried him to an upset win for the presidency over Hillary Clinton. In 2020, when the polling error was even larger than four years prior, he fell short of reelection but made the race much more of a nail-bitter than most observers expected.
Pollsters adjusted their methods, as they often do from cycle to cycle, to avoid their past misses. They began to weigh more widely on educational background, which has become increasingly an indicator of how people vote, and worked to better reach certain voters they were struggling to fully take into account in polls.
But they said in advance of Election Day that only waiting for the results would allow them to know if the changes they made were successful. At the same time, they emphasized that with the close race that the polls found, a small error could cause a more comfortable win for one candidate.
Lee said he was surprised to see Trump take all seven battlegrounds but believes pollsters did well correcting for the past issues. But even as analysts said either candidate could slightly outperform and pull off a sweep, Trump was the one who did so.
“Trump, being Trump, over-performed all the polling,” Lee said. “If you look at his actual Election Day margins in all the seven battlegrounds, it was bigger than the average lead he had in those states. So Trump did it again.”
Pollster Nate Silver called the results a “perfect demonstration of correlated polling error.” He said in a newsletter on Thursday that Trump’s sweep of the seven swing states was the most common simulation in his model, happening 20 percent of the time, because polling errors tend to be correlated and Trump was leading, albeit slightly, in five of the states.
A Harris sweep was the next-most common scenario in the simulations.
“When polls miss low on Trump in one key state, they probably also will in most or all of the others,” Silver wrote.
On the nationwide level, the polling average had closed to Trump and Harris being essentially tied by Election Day. Trump is currently winning the popular vote by about 2 points.
Still, polling saw some notable misses that if more accurate could have more directly pointed to Trump’s win.
The final Des Moines Register poll of Iowa from revered pollster J. Ann Selzer right before the election showed Harris ahead by 3 points. The poll was seen as a likely outlier at the time — but even so, its miss is notable: Trump ended up winning the state by 13 points.
Selzer said after the election that she’s reviewing the data to find out where the poll went wrong.
Pollsters said they were surprised by the major improvement Trump had among many key demographics, like Latino and young voters.
Some polls had shown Harris with double-digit leads among Latinos, but she only won the group by 8 points, according to exit polls. Some polls showed Harris with the traditional Democratic dominance among young voters, but she only won among 18-to-29-year-olds by barely 10 points.
John Cluverius, the assistant director of the Center for Public Opinion at the University of Massachusetts Lowell, said pollsters still have work to do to improve their methods, especially facing the constraints of the rising costs of conducting polls and the difficulty of getting people to participate.
He said pollsters particularly had trouble reaching young voters for polls.
“I think that is always a struggle for pollsters in being able to get a large enough sample in a narrow enough time and so there may be changes that opinion polls are just going to continue to struggle to make,” Cluverius said. “I just think that it’s too early to tell if there is a lesson that could be applied to future elections.”
Democratic pollster Celinda Lake pointed to an ongoing difficulty to determine what turnout is going to be in polls.
In general, pollsters can ask which candidate a voter prefers and if they plan to vote and determine how likely they are to vote based on their background and voting history, but they can’t guarantee if someone will vote.
“What we saw was a surge in turnout of Trump voters beyond what we had expected and less of a turnout on our side than we had expected,” Lake said.
Cluverius said as pollsters continue to analyze their methodologies, they need to be transparent and talk to people directly if they have questions.
“The more we are humble about the fact that we are pretty good but not perfect at measuring the attitudes of the public, and we are certainly better than every other method that’s been tested, that I think is the space we have to operate in,” he said.
Caroline Vakil contributed.
