Campaign

Pollsters: Don’t be so sure Trump will outperform our surveys

by Jared Gans - 09/30/24 6:00 AM ET

Pollsters caution against assuming that the polls this cycle are susceptible to the same errors as previous ones that underestimated support for former President Trump, arguing that every election is different and that this year’s polls are an accurate reflection of the competitiveness of the race.

Polls now show Vice President Harris leading Trump by about 4 points, according to the average from The Hill/Decision Desk HQ. But the race in the roughly half dozen battleground states is even closer, and a polling error like the ones in the past could mean Trump is in a stronger position to prevail than the data says.

But polling analysts say it’s not that simple.

“We don’t always see the misses in the same direction,” said Chris Jackson, the senior vice president of public affairs for Ipsos. “I can tell you that the polling industry has done substantial changes to how we do our surveys to try to account for what we think was driving those errors in 2020. So while there undoubtedly will be errors in the future, they’re probably going to be driven by different things and go in different directions.”

Pollsters have had a rough couple of presidential cycles in the Trump era, and it’s led to widespread skepticism of just how accurate their measurements are, even as news story after news story details the latest polling findings.

In both 2016 and 2020, Trump was the underdog, first to Hillary Clinton and then to Joe Biden. In both cases, he outperformed most of his polls.

The first time, it was enough to win the Electoral College. The second time, Biden won, but it was a very tight race in a number of battleground states.

Pollsters acknowledge Trump’s rise has posed a new challenge for the industry trying to accurately track voters’ preferences, but they say methodologies have adjusted.

Jackson said the polling industry in 2020 and before looked to “reliable benchmarks” for sampling and weighting surveys, usually based on census data, to ensure pollsters had a representative sample. But pollsters now realize that trends were happening that demographics were not accounting for.

This led to a significant adoption of other factors such as party registration and past voting history for added political criteria for weighing results.

“There’s a bunch of different ways of doing it that are currently being used in the field, but that has been a relatively widespread shift in the last four years,” Jackson said.

“I think the methodological changes are significant enough that we shouldn’t necessarily take the 2016 and 2020 results as certain,” he added.

The trouble with turnout

Turnout has been a problem for pollsters when it comes to Trump.

Democratic pollster Celinda Lake noted that pollsters have tried “a number of techniques” to avoid the inaccuracies of measuring Trump’s support in the key states, but calculating turnout remains difficult.

“Figuring out the turnout is the hardest thing out there. And this cycle there may be a surprise Trump vote and a surprise Harris vote. You may have some younger, some people of color who don’t look like they’re turning out either,” she said.

Lake said while Democrats are gaining an edge with newly registered voters, many of whom are young women concerned about access to reproductive health care, more voters who are already registered and planning to vote for the first time this election favor Trump.

“What we’re doing in our polling at this close [to the election] date is we’re often giving [multiple] turnout models. This is where we’re at. If it’s like this, this is where we’re at,” she said, explaining that different turnout models for Election Day can produce significantly different outcomes.

Perception vs. reality

Some pollsters also argued that the public’s interpretations of the polls and forecast models are more responsible for the perception that Trump outperforms the polls — rather than the numbers themselves.

Lee Miringoff, the director of the Marist College Institute for Public Opinion, noted that the final national polling for the 2016 election was close to the result, with Clinton winning the popular vote by about 2 points. FiveThirtyEight’s average had her up 4 points, while RealClearPolitics had her ahead by about 2 points.

The average in the key states that ultimately gave Trump victory were a bit more favorable to Clinton, but the forecast models still had her as the overwhelming favorite.

“The forecasting and polling doesn’t necessarily get separated for the different measurements that they really are,” Miringoff said.

He also argued that the polls for the states did not have time to capture last-minute developments in the race, like former FBI Director James Comey’s letter reopening the investigation into Clinton’s email server in 2016.

The shy Trump voter

Experts also rejected the idea of the reason for past polling difficulties being the “shy Trump voter,” the idea that certain voters were uncomfortable with telling pollsters in the lead-up to the 2016 and 2020 elections that they planned to vote for Trump.

“I’ve talked to Trump voters. They don’t seem shy to me,” Jackson said.

The American Association for Public Opinion Research released a report following the 2020 election evaluating preelection polling, which overstated Biden’s support compared to Trump by an even larger margin than in 2016.

Jackson, who was involved in crafting the report, said it found “no real data” to support the idea of shy Trump voters. Instead, it found some evidence, among other conclusions, that the issue was a struggle in reaching these voters who were more likely backing Trump.

“It’s not that they answer the phone and lie, it’s just that they don’t answer the phone,” he said, but he added that pollsters have been working to address this as fewer people use landlines.

Miringoff said most pollsters have changed their methodologies since 2016 because reaching people by landline is “no longer feasible.” Some have adopted hybrid models to reach voters through different methods.

“Statistically, it’s not enough to warrant the conclusions that things are really broken because Donald Trump is on the ballot,” he said. “I think we can have problems as a polling community with and without [him].”