For decades, pre-election polls have been essential to how journalists and the public understand the ebb and flow of presidential campaigns. Much the same can be said for this year’s election cycle.
Along with polls in abundance over the next few weeks, we’re likely to encounter some of the exaggerated and cockeyed tall tales about presidential campaign surveys — such as the notion that polls got it dramatically wrong in the 2016 election. That sense spread widely in the aftermath of President Donald Trump’s stunning victory over Hillary Rodham Clinton and persists to this day. It’s enough to make opinion researchers bristle.
In his presidential address a couple of years ago to the American Association for Public Opinion Research, Timothy Johnson said: “If you’re like me, seldom does a day pass when you are not obliged to correct the declaration of a friend, an acquaintance, or a university administrator that ‘the surveys got it wrong in 2016.’ This is going to be with us for a long time.”
Nationally, the polls were pretty close, in aggregate, thanks to Ms. Clinton’s wiping out Mr. Trump in California. She carried the Golden State by more than 4 million votes, a margin that erased Mr. Trump’s popular vote advantage in the rest of the country and gave Ms. Clinton an overall lead of 2.1 percentage points. According to the final pre-election compilation of Real Clear Politics in 2016, the national polls overall placed Ms. Clinton ahead by 3.3 points. So the national polls overall didn’t deviate greatly from the final result.
But there’s more to the story of the polls in 2016 than the misunderstood popular vote. In decisive states the polls clearly misfired, especially so in Wisconsin and Pennsylvania, and to a lesser extent Michigan. Mr. Trump defied the polls in those states and won them all narrowly. Had Ms. Clinton carried Wisconsin, Pennsylvania, and Michigan — as had been expected — she would have won the Electoral College and the presidency.
Anyone who takes more than passing interest in election surveys surely has heard the tale that pollsters were so confident Republican Thomas Dewey would win the presidency in 1948 that they stopped taking surveys weeks before Election Day. The blunder, it is said, caused them to miss a late surge of support for President Harry Truman.
It’s a plausible excuse for what was an epic polling failure. But it’s not entirely true.
George Gallup, who before the 1948 election was called the “Babe Ruth” of political polling by Time magazine, did complete nearly all his poll-taking 10 days or so before the election. But his principal rival, Elmo Roper, was quietly in the field in late October, wrapping up just days before the election. He found that while Mr. Truman had sliced into Dewey’s lead, the Republican was well ahead.
Roper had pledged not to release new poll results unless a political miracle intervened. The late-campaign poll showed no miracle in the making so Roper didn’t report it. He repurposed a September poll and predicted Dewey would win by 15 percentage points. Truman won by 4.5 percentage points.
It was a huge miss that recalled the Literary Digest and its descent to oblivion following a notoriously inaccurate poll in 1936.
The Literary Digest was a popular weekly magazine that in the 1920s and early 1930s had achieved what often was called an “uncanny” record of accurately forecasting presidential elections. It sent postcard-ballots to millions of people across the country, asking that they be filled in and returned. In 1932, the Literary Digest mass poll forecast Franklin D. Roosevelt’s victory by 1.4 percentage points.
But in 1936, the publication’s poll projected Republican Alf Landon of Kansas would unseat Roosevelt by near-landslide proportions. Instead, Roosevelt carried all but two states in the most one-sided contested presidential election to that time. The Literary Digest poll was off by 19.9 percentage points. It likely fell victim to non-response bias, in that Roosevelt’s foes were more motivated to participate than were his supporters, producing a lopsided sample favoring Landon.
The polling failure was so immense that supposedly it drove the Literary Digest out of business — a not inconceivable penalty for getting it so woefully wrong.
But that’s another exaggeration. The errant poll didn’t destroy the publication. The magazine had been in decline years before 1936 and its content seemed stale in comparison to the edgier reporting of Time. “It lost out,” the New York Times said soon after the Literary Digest was absorbed by Time in 1938, “because it couldn’t adapt its formula to a jazzier lot of young people, whose attention is usually there when one wants it here.”
Like other myths, these tall tales persist because they are easy-to-grasp parables. They supposedly tell us something about the imperfection of election polls. They also simplify the intricacy of polling, making it seem more accessible if not entirely understandable.
W. Joseph Campbell (email@example.com) is a professor of communication at American University in Washington, D.C., and author of the recently published book, “Lost in a Gallup: Polling Failure in U.S. Presidential Elections” (University of California Press).