Why the Polls Missed Again in 2020

by Caroline Sutton

Illustration by Weronika Ziarek

While the result of the American presidential contest couldn’t be more different from four years ago, one phenomenon has political junkies feeling a terrible sense of whiplash. Once again, pollsters told a story to public and campaign strategists alike that right-wing populists would be swept away by a “blue wave,” and once again forecasters' assurances seem out to sea—at Democrats’ expense. While Joe Biden will be America’s next president, he did not sail to victory by double digit margins in key states as polls led us to believe he would, and the party lost vital state legislatures just before redistricting.

A quick postmortem indicates expectations of Democratic domination were rather naive. Of course, a head-of-ticket who mobilizes moderate Republicans will hurt down-ballot candidates in a country where ticket splitting—when a voter selects candidates from different political parties when multiple offices are being decided—is common. The Biden team’s principled decision to campaign mainly online in order to avoid in-person interactions due to COVID-19 likely also hurt their chances, as research suggests digital activism favors conservative causes. But why weren’t Democrats warned of these trendlines beforehand? How can these polling inaccuracies, so maligned four years ago, appear again this cycle in nearly the exact same form?

The answer lies in the “blind spots” of surveys—and in how the world of US politics has changed from polling’s heyday in the 1960s and 70s.

First, bear with me for a quick technical rundown: polls do not predict voter turnout. That is to say, likely voter models work best when turnout is “steady”—the same people who vote in the last presidential election vote in this one, with a predictable number and kind of voter being added to the electoral pool each election. This stability lets pollsters have a clear sense of the people they need to contact, and then discover if those people prefer one candidate or another.

Campaigns in the 1960s and 70s mirrored the predilections of polls. Strategists and candidates saw the way to win races as convincing the median ideological voter to support their run, thus appealing to the greatest portion of the electorate. Richard Nixon epitomized this approach. In his first successful run for president in 1968, he was the “moderate” candidate that the Republican Party chose after Barry Goldwater’s far right-wing candidacy in 1964 alienated many voters; he lost in a landslide. Nixon, perhaps not coincidentally, was obsessed with polls and likely used them more than any candidate had previously. He utilized not just electoral polls, but paid to conduct private, personalized polls that he used to shape his governing process and public policy agenda. Nixon requested information on the public’s reaction to Vietnam, inflation, busing, taxes, and China’s admission into the United Nations. He demanded that his private polls be accompanied by reports that “tell him what’s really important and where we need to make changes.” 233 such polls were taken in Nixon’s first term. The Nixon years marked the era when polling became a consistent, clear shaper of American politics.

And why wouldn’t Nixon be preoccupied with polling? Conventional wisdom at the time advised that the path to victory was through courting the median ideological voter. Rational choice theory, the influential political science approach that endorsed a centrist approach to politics, had only been published in the late 1950s, and recent elections seemed to prove the theory correct time and again. As mentioned, Barry Goldwater, whose campaign rallies in the Deep South reporting from The New Yorker at the time referred to as “great carnivals of white supremacy,” was a polarizing figure who carried only six states on election day and allowed his opponent to be reelected by the largest popular vote margin in US history.

Nixon gives his trademark "victory" sign as he campaigns in Pennsylvania, July 1968

In the election of 1972, Nixon would confront avowed liberal George McGovern, who ran on an anti-Vietnam War platform that included granting amnesty to draft evaders. Again, Nixon would see his centrist approach to politics rewarded immensely when he would go on to win every state but Massachusetts and beat McGovern by a popular vote margin of 23.2 points. Success in politics, it seemed, was easy: appeal to the middle and reap the rewards. Polling was the key to it all, with its ability to let a politician check in regularly with voters and ascertain the consensus viewpoint on any given position.

Polls that help forecast election results are best suited to environments where campaigns focus on persuading middle voters rather than turning out new or partisan voters, as this helps pollsters be more certain about how to construct their samples. There were other factors that helped make the 1960s and 70s so suited to polling, however. This was an era when nearly every household had a landline, and pollsters could reliably contact a representative sample of the American electorate. Now, in the days of cell phones, response rates have plummeted. One pollster, speaking to The Boston Globe, recalled that response rates have fallen from 80% to 8% over the course of his career. The further American society moves away from these “ideal’ conditions,” the more difficulties may appear in forecasting election results.

These days, we see more “extreme” candidates like Donald Trump and Bernie Sanders who campaign, not by appealing to the ideological middle, but by turning out voters who do not reliably participate in electoral politics. Polling, which came into maturity in a very different political environment, cannot effectively grapple with this “new” style of political campaigns and as such is not reliable in this environment, as we have witnessed in issues for both Sanders and Trump.

The issues that cropped up this cycle were especially worrying. It is normal for errors in polling to occur but they are relatively random, i.e. they err in favor of both parties. This time, the errors almost uniformly showed Trump and other Republicans trailing what would be their actual vote margin on election day, which suggests a very serious problem in how pollsters determine the correct makeup of the electorate. To make matters worse, the errors in 2016 were similar, but the problem was mainly confined to the Rust Belt. Now, they are present in a variety of swing states. Because of the way polls function, it will be very difficult for experts to solve this problem. In the meantime, it appears that polls, which guide countless strategic decisions, are consistently unreliable when politicians campaign based on voter turnout rather than persuasion. With all this in mind, pollsters, and those who make decisions based on their work, should be cognizant of just how much has changed since polls first became a key part of electoral success in the 1960s and 70s, and build this into their calculations.

Caroline Sutton writes on on political communication, technology, and public diplomacy. She currently resides in Wilmington, North Carolina and is a graduate of the MSc Program in Politics and Communication at the London School of Economics.