This article is part of the Debatable newsletter. You can sign up here to receive it on Tuesdays and Thursdays.
For weeks now, gleeful Republicans and anxious liberals alike have been warning that Joe Biden’s edge over President Trump is on the verge of collapsing. But eight weeks away from the election, it hasn’t: Mr. Biden continues to lead Mr. Trump by seven to 10 percentage points in the latest polls, as he has for several months, and has roughly a 71 percent chance of winning the election, according to FiveThirtyEight.
If those odds seem familiar, it’s because they’re almost exactly the same ones that were given to Hillary Clinton in the run-up to the 2016 presidential election. Four years after what was widely seen (rightly or wrongly) as a failure of data analysis, how has the inexact science of polling changed, and is it worth paying attention to? Here’s what people are saying.
‘The idea that polls are useless because of what happened in 2016 is just wrong.’
The prevailing narrative about 2016 polling is wrong, or at least incomplete, writes Quartz’s data editor, Dan Kopf. It’s true that major vote forecasters heavily favored Mrs. Clinton to win, putting her chances in the range of 70 percent to 99 percent, but her loss hardly invalidated the profession: All in all, the presidential election polls in 2016 were about as accurate as they have been on average since 1972, according to FiveThirtyEight’s founder Nate Silver. Both FiveThirtyEight and The New York Times projected Mrs. Clinton would win the national popular vote by three percentage points, and she ultimately won by about two.
Of course, American elections are decided by key swing states in the Electoral College, not the popular vote, which is where prognosticators did indeed err. What went wrong? According to Nate Cohn, a correspondent for The Upshot, there were at least three unforeseen factors at play: Undecided voters broke for Mr. Trump in the last days of the race; more Trump supporters turned out than expected; and state polls underestimated Mr. Trump’s support in the Rust Belt, in part because they did not correct for the overrepresentation of college graduates.
Some of those issues have been remedied or have become less relevant. As Li Zhou writes in Vox, political allegiances have hardened since 2016, and there may well be fewer undecided voters this time around. Whereas polls from the last election overestimated Democratic turnout because they used Obama-era models, this year’s polls use 2016 as a reference point and so may even underestimate Democratic turnout. And perhaps most important, many pollsters now weight their sample to properly represent voters without a college degree.
“That is a huge change — an elixir against being deceived again,” writes Stanley Greenberg, a political strategist and polling adviser, in The Atlantic. “The pain of Trump’s victory and disastrous presidency has concentrated the minds of campaign staff and the polling profession in ways that give me confidence that Biden’s lead in the polls is real.”
Ultimately, many pollsters believe that whatever decline in trust their profession has suffered since 2016 owes less to a failure in polling itself than to a general misapprehension of what it can and cannot tell us. While most public opinion surveys use demographic benchmarks to ensure a highly accurate representation of the American population, election polls require educated estimates about which subset of the population will end up voting. For that reason, the American Association for Public Opinion Research reported in 2016, it’s “a mistake to observe errors in an election such as 2016 that featured late movement and a somewhat unusual turnout pattern, and conclude that all polls are broken.”
The 2018 midterms were a case in point: Of 506 congressional races, FiveThirtyEight called the correct winner in 490, or about 97 percent. “Polls were never as good as the media assumed they were before 2016 — and they aren’t nearly as bad as the media seems to assume they are now,” Nate Silver wrote that year. “In reality, not that much has changed.”
‘Nobody can predict this election.’
By almost any measure, 2020 will be unlike any other election, writes David Byler, a data analyst at The Washington Post. He cites seven factors that make the election extremely difficult to forecast, chief among them a pandemic unlike any in recent memory, a correspondingly unusual economic recession and an anticipated but unpredictable expansion of mail-in voting.
And while polls do tend to get more accurate as an election nears, unforeseeable, game-changing events — like the letter about Mrs. Clinton’s emails that the former F.B.I. director James Comey sent to Congress just over a week before the 2016 election — are always a possibility. Who’s to say a second pandemic won’t have darkened the globe come November? “If ever there has been a year where unforeseeable things can happen, it’s 2020,” J. Ann Selzer, widely regarded as one of the best pollsters in the country, told Axios. “Anybody that’s not wary this time is kind of kidding themselves.”
Concerns about polling’s usefulness for presidential elections also extend beyond the particulars of this one. As Courtney Kennedy, the director of survey research at Pew Research Center, notes, polling margins of error are often bigger than advertised, on average about double the three percentage points typically reported. One reason for that is the declining rate of response to telephone surveys, a problem that has been looming over the industry for several years now. The problem hasn’t been solved by the rise of online polls, which tend to lean to the left of telephone surveys and are generally, though certainly not always, less reliable.
The upshot, according to Dr. Kennedy, is that “while polls remain useful in showing whether the public tends to favor or oppose key policies, this hidden error underscores the fact that polls are not precise enough to call the winner in a close election.” Presidential elections often are close, which calls into question whether polling ought to play such a prominent role in them.
Some critics argue that the problem with polls is not so much their accuracy as their effects on the voting process. As Molly Olmstead notes in Slate, election forecasting can increase certainty about the outcome, sow confusion and even decrease turnout, according to a study published in The Journal of Politics. “Even though they make for dramatic headlines,” one of the study authors told her, “the research shows it is nearly impossible to convey probabilities in a way that does not generate confusion.”
Such unanticipated harms are one reason the Harvard historian Jill Lepore has gone so far as to suggest that polling is “completely inimical to representative democracy”: By allowing polls to consume so much of our civic and journalistic attention, we resign ourselves to an anxious vision of democracy in which voters are less agents than spectators. “Tweeting is to talking what polling is to voting,” she writes in The New Yorker. “Democracy requires participation, deliberation, representation, and leadership — the actual things, not their simulation.”
It’s possible we’re close enough to Nov. 3 that simulations of the election aren’t very far from the actual thing. Either way, we’ll find out in eight weeks.
Do you have a point of view we missed? Email us at email@example.com. Please note your name, age and location in your response, which may be included in the next newsletter.