The Memo: Democrats fear Trump will outperform polls again

Democrats who are in high spirits over Vice President Harris’s debate performance and buoyed by her rise in the polls have one doubt that nags at them above all others.

It’s the knowledge that former President Trump has consistently outperformed his poll numbers in the past — and the fear that he might do so again.

Given that many polls already have the battleground states balanced on a knife-edge, the idea of history repeating itself is a Democratic nightmare.

“I think we still have to worry about a Trump surge,” Celinda Lake, one of two leading pollsters for President Biden’s 2020 campaign, told this column via email.

One of the central difficulties for any pollster is how to model turnout. In most cases, that involves an educated guess about how many people from which demographics will actually cast ballots.

That’s one reason for Lake’s concern.

“Trump is winning men who have not voted” previously, she stated. “Most pollsters are adjusting. Our firm looks at two turnout estimates now. One the average and one looking at [a] Trump surge.”

Harris doesn’t have any leeway if such a surge took place.

In the polling averages maintained by The Hill and Decision Desk HQ (DDHQ), Harris leads nationally by 3.4 points, but the races are much closer in most of the key states. 

In the three “blue wall” states of Pennsylvania, Michigan and Wisconsin — all of which are vital to Harris’s hopes — she leads by 0.4 points, 0.8 points and 3.0 points respectively.

DDHQ currently gives Harris a 54 percent chance of prevailing in November.

The organization’s director of data science, Scott Tranter, warned Democrats against taking any confidence from that number. 

In probability terms, a 54 percent chance of winning basically means that in an imaginary scenario where the election could be run 20 times, Harris would win 11 times and Trump 9 times.

“If you feel comfortable with a 54 percent chance, then you probably have to understand probabilities a little better,” Tranter said. “This is a coin flip. Nobody should be surprised if Kamala Harris wins or if Donald Trump wins, any more than you would be surprised if you flipped a coin and it came up tails.”

As the public focuses on the tightness of this year’s race, the specific ways in which the polls went awry in Trump’s previous two White House runs bears emphasizing.

Back in 2016, national polls measuring Trump’s performance against Democratic nominee Hillary Clinton were not off by a massive amount.

The final RealClearPolitics (RCP) average before Election Day had Clinton besting Trump by 3.2 points. In the actual results, she beat him by 2.1 points in the popular vote — while, of course, losing the Electoral College.

The result was such a shock partly because the polls missed so badly in the decisive states. Across the blue wall, pollsters measured Clinton’s showing with startling accuracy but markedly underestimated Trump’s vote share.

In the worst case of all, Wisconsin, the final RCP average predicted Clinton’s vote share within 0.3 percentage points of what she actually received. 

But it underestimated Trump’s by almost 7 points. He edged Clinton out by winning 47.2 percent of the vote, way ahead of the 40.3 percent the final RCP average had projected.

In the wake of the 2016 shocker, there were various explanations offered about why the polls might have underestimated Trump’s support. 

Suggestions included that the pollsters’ models of turnout had been off, underestimating rural and exurban turnout for Trump; that Trump supporters, perhaps more distrustful than their Democratic counterparts of polls and the media, were more reluctant to respond to surveys; or that some respondents had simply lied, ultimately backing Trump in the privacy of the voting booth having declined to tell a pollster they were going to do so.

The 2016 outcome sparked a lot of introspection on the part of pollsters, but the industry’s overall performance in the 2020 race wasn’t much better.

In fact, in 2020 the national polls were further off than in 2016. The final RCP average predicted a 7-point victory for Biden, though he prevailed in the real popular vote by 4.5 points. 

Another factor remained the same: Pollsters were excellent at estimating the Democratic nominee’s level of support but poor at predicting Trump’s. The final RCP average came within one-fifth of a percentage point of Biden’s actual vote. But it underestimated Trump’s by almost 3 points.

In the three blue wall states, Wisconsin once again produced the biggest miss, with the final RCP average underestimating Trump’s actual vote share by almost 5 points.

Misfires on that scale put the current polls in this year’s race into stark perspective. As of Thursday evening, The Hill/DDHQ polling averages of seven battleground states showed only one in which either Trump or Harris was leading by more than a single percentage point. 

The state in question, where Harris is up 3 points, was Wisconsin.

There have not been enough polls since Tuesday’s debate to get a firm picture of whether the clash changed the race in any fundamental way.

Harris was widely seen as the victor and, on Thursday, Trump asserted in a social media post that there would be no more debates.

Also Thursday, Trump pollsters Tony Fabrizio and Travis Tunis issued a memo — made public by the campaign — that asserted that its own polls showed a 2-point bump for Trump after the debate, while Harris’s support had “remained flat.”

However, a new Morning Consult poll, also taken after the debate and released Thursday, showed Harris with a 5-point national lead — her largest edge in any poll so far from the organization.

Harris’s campaign continues to insist she is the underdog — though there is a question mark over whether her aides sincerely believe this or are using the claim to keep the motivation of her supporters sky-high.

One thing is sure: The race is very close indeed, and Democratic anxieties are going to remain sharp all the way until the results come in.

The Memo is a reported column by Niall Stanage.