Some Republicans assume former President Donald Trump’s historic polling underperformance spells a bloodbath for Democrats in November. But the assumption’s accuracy, according to four center-right analysts, is muddled by new changes in methodology.
These analysts offered some insight into how the numbers might unfold, but conceded there’ll be no certainty until after Election Day. For them, 2024’s polling accuracy boils down to three main factors: demographic overcorrection, response bias and something called “recalled vote weighting.”
Current battleground aggregates place the former president even or ahead of Vice President Kamala Harris, bolstering Republicans who argue Trump will once again triumph due to polling error. Furthermore, some have taken to comparing recent polling to numbers from the same time during the 2020 and 2016 cycles, which shows Trump polling significantly ahead of his previous runs for president.
It’s not really that close Nate. https://t.co/oBZxpiWvfR
— Chuck Armstrong (@ChuckArmstron11) October 29, 2024
Polling averages placed President Joe Biden, for example, ahead of Trump by 8% in Michigan on Oct. 25, 2020, according to FiveThirtyEight — and Biden only won the state by 2.8%.
Harris led Michigan by 0.7% on the same day in 2024. With the same level of polling error, Trump would win the state by 5.1%.
But is it reasonable to expect the same level of polling error in 2024 — and Trump to sweep battlegrounds by five points or more?
Trafalgar Group founder Robert Cahaly doubts such dramatic margins, but said he still expects Trump to outperform October polling in an interview with the Caller.
“Is that possible?” he said. “Yes, but I think the likelihood is that he’s just going to overperform by about two or three points… It’s different in 2024, but we see a significant hidden [Trump] vote. We think that the places Trump is most likely to significantly overperform a poll are probably going to be Arizona and Michigan and Georgia.”
Trump is slated to outperform polls in these states, Cahaly argues, because his firm has found that undecided voters are breaking for the former president there. He added that pollsters struggle to capture Trump voters because they need to be “cajoled” into answering polls and assured of their anonymity, fearing social backlash.
There is some reason to doubt the notion that Trump’s true standing is so much higher than current numbers suggest; pollsters have identified and rectified many of the sources of their past error, with FiveThirtyEight even touting the 2022 midterm as the most accurately predicted cycle since “at least 1998, with almost no bias toward either party.”
WATCH: DecisionDeskHQ analyst @MichaelPruser debunks the notion that polls suddenly “fixed” for the “hidden Trump vote” and are thus now “accurate.”
“I have NO confidence, absolutely ZERO confidence, that polling will be correct this year.”
“The response bias is educated and… pic.twitter.com/YPY6cbpr0B
— Eric Daugherty (@EricLDaugh) October 27, 2024
“The reason they were really off, in my judgment, in 2016 is that they undercounted non-college-educated whites,” author and political consultant Ryan Girdusky told the Caller. “They had an over-response bias of college-educated white liberals.”
Pollsters generally didn’t account for voters’ education levels in 2016, a data point which hadn’t previously cleaved so sharply along partisan lines, according to a report from the American Association for Public Opinion Research. (RELATED: Trump Leading Harris In Key Swing State Two Weeks Before Election Day, Poll Finds)
Though firms had corrected the education oversight by 2020, polls still significantly underestimated Trump’s support in that cycle. Girdusky blames this on “partisan response bias,” the tendency of certain demographics to answer polls at a higher rate, and said that it could distort predictions for 2024.
“In July, when everybody on the planet was calling on Joe Biden to quit the race because he was going to lose, polls still showed him winning seniors, but he didn’t win them [in 2020],” he said. “That’s how overly large the response bias is among liberal seniors — the people who live to watch Rachel Maddow over and over.”
Some analysts doubt that the accuracy of 2022 polling necessarily guarantees similarly reliable predictions for 2024.
“It’s about the expanse of the electorate,” Mark Harris, the co-founder of ColdSpark, a political consulting firm, told the Caller. “The electorate gets much larger in a presidential election year, and the people who come into it are harder to get on the phone.”
He echoed concerns about response bias, noting that non-college-educated whites, who he said comprise about half of Pennsylvania’s electorate, are “just a lot harder” to poll.
But adjusting methodology for greater accuracy isn’t an insurmountable challenge. Private and internal polling, Harris argues, are usually pretty reliable — it’s the public pollsters who bear the lion’s share of the blame for past inaccuracies.
Universities and other public pollsters often can’t afford to conduct statewide surveys with large sample sizes, he explained, which forces them to artificially adjust, or “weight,” responses from under or overrepresented demographics — leading to less accurate predictions. Public pollsters also use online surveys, which is less “rigorous” than tactics employed by private firms, he argues.
“You could easily have [in a public poll], ‘Oh, we did 20 interviews with Republican women, and Trump’s only getting 60% of them…’ Well, if you extrapolate them up to be 40% of the sample, some weird, weird stuff is going to happen,” Harris said. “Anyone who does a lot of weighting is really gambling.”
“It’s just money,” he continued. “I have nothing against them. I get why they [weight their polls] — it’s still corners being cut.”
18 days until Election Day! Here’s how the 2024 presidential race looks compared to the same date during the 2020 and 2016 elections, according to the RCP poll averages. pic.twitter.com/hUug00Bryi
— Matt Margolis (@mattmargolis) October 18, 2024
Though ongoing issues with response bias and overzealous demographic weighting suggest pollsters are still undercounting Trump’s performance, Republicans should temper their expectations. Harris emphasized that response bias diminishes in the weeks leading up to Election Day. He also cautioned against “recall ballot weighting,” where pollsters ask voters about their past votes and adjust their sample based on that information. This approach, he argues, overlooks the fact that “no two electorates are ever the same.”
Pollsters have increasingly employed this technique to ensure Trump supporters are adequately represented in their samples, aiming to correct their previous difficulty in fully capturing the former president’s support, according to The New York Times. Some critics say it could lead to an overrepresentation of Republicans in current polling.
Mark Mitchell, the head pollster of Rasmussen Reports, defended the practice in an interview with the Caller.
“There’s a ton of reasons in my state polling why Trump is going to outperform,” he said. “I’m leaning heavily to the recalled vote. That is going to make my numbers less sensitive to low-propensity and first-time voters. And everything we’ve seen tells us that those people are showing up and are raking right.”
Recall weighting, Mitchell explained, was actually diminishing what he believes to be the full breadth of Trump’s support in Rasmussen’s polling, adding that the former president would be “up four or five points in the national popular vote” without recall weighting.
“I’ve been showing Trump [ahead by] 2% in the national popular vote for months now,” Mitchell said. “That means he’ll probably sweep all of the battlegrounds, and nobody’s talking about that.”
Despite some advances in polling methodology, there remain several indicators that pollsters will again underestimate Trump’s support, though it seems unlikely that November will vindicate the same level of polling error the “bloodbath” narrative suggests — the unique dynamics of each election cycle continue to make precise predictions elusive.