Ballot shock?
Repeat of 2016 polling miscue that predicted Trump loss is possible, experts say
Michael Moore is definitely no fan of Donald Trump’s, but his comments last week would have made the president smile.
The polls showing Joe Biden enjoying a comfortable lead over Trump in Moore’s native Michigan, a key battleground state in Tuesday’s U. S. election, are undoubtedly overstating the Democrat’s popularity, the filmmaker asserted.
“The Trump vote is always being undercounted … The Trump voter’s very suspicious of the ‘ Deep State’ calling them and asking them who they’re voting for,” Moore told The Hill. “Whatever they’re saying the Biden lead is, cut it in half, right now, in your head.”
His assessment was not exactly scientific, but it had some credibility. The Roger and Me director made the same observation before the 2016 vote, when polls in such states suggested that Hillary Clinton would ease to victory.
Indeed, the survey misfire of four years ago continues to haunt U. S. politics, and raises a nagging question: could polls showing a consistent lead for Biden nationally and in swing states once again prove off the mark?
At least some pollsters have tried to fix problems identified in the wake of the 2016 debacle, while the Democratic candidate’s advantage is stable, barely changing over the last several months. But Trump’s unexpected election underlined how difficult it can be to get polling right, and not all experts are convinced this year will be different.
“Maybe I’m just an anxious person, but it is possible that we would have another surprise on the order of 2016,” said Courtney Kennedy, director of survey research at the respected Pew Research Center. “Donald Trump, he’s a very unique candidate, and the type of voters he has turned out have proven tricky for pollsters to represent.”
A scholar who has studied a long history of opinion- tracking missteps says it’s unlikely, but he would also not rule out a repeat of four years ago.
“Maybe lightning will strike twice,” said W. Joseph Campbell, a communications professor at American University in Washington, D.C., and author of Lost in a Gallup: Polling Failure in U.S. Presidential Elections.
The first time lightning struck, it was by no means an across- the- board screwup, despite public perception.
“In 2016, polls were about as accurate as they have ever been,” Doug Schwartz, director of the well- regarded Quinnipiac University Poll, insisted by email.
That’s true of the national surveys, certainly. As the campaign ended, they showed Clinton ahead by an average of about three percentage points and she wound up taking the popular vote by just over two points.
The problem was in individual battleground states, especially the upper midwest ones that had been part of the Democratic “blue wall.”
In Michigan, Wisconsin and Pennsylvania, pollsters reported an edge of up to about 6.5 percentage points for the Democratic candidate. Trump ended up winning all three states, albeit by the narrowest of margins, about 77,000 votes out of 14 million cast. It was enough to capture a majority of electoral college votes, and the White House.
Trump’s surprise win was a “jarring” event for the industry, shocking even his own pollsters, says a postmortem report by the American Association of Public Opinion Research. What went wrong? One major flaw centred around Trump’s famous success with white people lacking college degrees, the association concluded. Opinion surveys tend to have better luck connecting with voters who have more formal education, meaning pollsters should “weight” their surveys for the non- college types — adjust the results to reflect that under- representation. Many did not do that.
Also a factor, the association found, was a lastminute shift of undecided voters to Trump. That break to the Republican seems to have come too late to have been picked up by the last polls of the 2016 election.
Trump has often accused pollsters of being biased against him, but the industry group found no evidence of partisan prejudice, noting that polling underestimated Democratic support in 2000 and 2012.
Since the association’s report, research uncovered another problem: a surge in voting by rural Americans — benefiting Trump — that was different from behaviour in previous elections and not accounted for by polls, said Kennedy.
The question now, at least in part, is whether four years later those errors have been corrected. The answer is somewhat murky.
A late- arriving tide of Trump support seems not so likely in 2020. While undecideds were as high as 15 per cent toward the end of the 2016 campaign, they are a third of that or less now, said Kennedy. And the deluge of advance voting would tend to solidify current polling trends.
Schwartz said his firm has always adjusted for education and believes colleagues who didn’t in 2016 are doing it now.
“I am confident that our weighting is not missing any major factors,” said the
Quinnipac pollster.
His organization’s most recent poll, released last Thursday, showed a tight race in Florida and Iowa, with Biden ahead in Pennsylvania and Ohio.
Daron Shaw, a University of Texas professor who runs the Fox News poll, said the so- called “shy-trump voter” issue — that people responding to polls are reluctant to admit they support the controversial president — could rear its head, but noted there are ways of avoiding that pitfall.
“I haven’t seen much in the 2020 data indicating that this is an issue,” Shaw said by email. “But we’re still very cognizant of it.”
The last Fox poll showed Biden with an eight- point lead nationally.
Meanwhile, prognosticators at The New York Times and Fivethirtyeight. com argue it would take a much bigger survey blunder in 2020 for Trump to win, given Biden’s wider lead.
Then again, some pollsters have said they did weight for education last time, and still failed to accurately predict Trump’s breakthrough, said American University’s Campbell.
History suggests polling miscues, when they occur, each have their own unique twists, he said. In other words, there could be some other factor masking Trump’s support from pollsters — or Biden’s — that’s yet to be uncovered.
“These miscalls or flubs or fiascos, whatever you want to call them, don’t happen every election and don’t happen quite the same way every time,” said Campbell.
Kennedy said she’s not sure that all pollsters, especially smaller outfits producing the state- level surveys, have fixed the weighting issue. And she foresees another potential problem among firms that are using cellphone texting and other online methods as opposed to the traditional phone calls, to contact people.
Those communications mediums are favoured by urban, liberal people, and using them could skew results toward the Democrats, she said.
Pew itself, well aware of their limitations, has eschewed horse-race polls altogether, while still conducting other election- related surveys.
“We do not think that polls predict the future, and we don’t really want to give our audience that impression,” said Kennedy. “They’re a snapshot in time … It’s kind of concerning the extent to which people want polls to be crystal balls, and they’re just not.”