Political polling pitfalls
Election forecasting is far from being an exact science
On the day after the presidential election 70 years ago, error born of certainty figured prominently in American journalism.
Drew Pearson’s “Washington MerryGo-Round” syndicated column of Nov. 3, 1948, mused about the men likely to fill senior positions in Thomas E. Dewey’s White House. They represented “an exciting, hard-working, close-knit clique who function with almost too much perfection,” Pearson said in the column, which was written before the election and distributed for publication afterward.
The same day, an early edition of the Washington Post published an admiring profile, that spoke of Dewey’s “persistence” — a trait the newspaper said defined the man just “elected to become the… President of the United States.”
More famously, an early edition of the Chicago Tribune declared in bold type across its front page, “Dewey Defeats Truman.” The lead story in the Journal of Commerce carried the headline, “Dewey Victory Seen as Mandate to Open New Era Of Government-Business Harmony, Public Confidence.”
Hubris, overconfidence and getting it wrong at election time are nothing new, as those anecdotes from President Harry S. Truman’s surprise re-election in 1948 suggest. Two years ago, of course, many journalists were likewise sure Hillary Clinton would be elected president. New York magazine’s pre-election issue featured the image of a snarling Donald Trump on the cover, across which was emblazoned the epithet “Loser.”
Natalie Jackson, then the senior polling editor for Huffington Post, wrote the day before the election that her poll-based model gave Ms. Clinton “a 98.2 percent chance of winning the presidency” and that Mr. Trump had “essentially no path to an Electoral College victory. Clinton’s win will be substantial, but not overwhelming.” Ms. Clinton, she added, “should fairly easily hold onto Michigan, Wisconsin and Pennsylvania” — all of which Mr. Trump carried, if narrowly, in winning the presidency.
While they offered little hint that Mr. Trump would prevail, national polls in 2016 broadly signaled Ms. Clinton’s narrow popular-vote victory. More significant were polls in swing states — notably those conducted in Michigan, Wisconsin and Pennsylvania — that consistently showed Ms. Clinton ahead. Additionally, pollbased statistical models, such as Ms. Jackson’s and others, contributed powerfully to the presumption that Ms. Clinton’s victory was inevitable.
Polls and poll-based forecasts helped set the election narrative for journalists and pundits in 2016, much as they had in 1948. It was almost unthinkable that Mr. Trump, or Truman, would win. So confident were the national pollsters in 1948 that they stopped sampling voters a week or more before the election. One of them, Elmo Roper, released no new polling data after Sept. 9, 1948, saying Dewey was destined to prevail “by a heavy margin.” Given Dewey’s seemingly insurmountable lead, Roper said he was going to devote his “time and effort to other things” that fall.
A few days before the vote in 1948, George Gallup, the polling evangelist of his time, made the remarkable claim that the election would be an opportunity for the public “to see down to the last percentage point how good we [pollsters] are.” As it turned out, Gallup underestimated Truman’s popular vote by five percentage points, and overstated Dewey’s vote by more than four points. Gallup later said the strain of miscalling the 1948 race was such that he went sleepless for days.
Journalists in 1948 certainly understood the agenda-setting effect that the polls exerted. Arthur Krock of The New York Times wrote they “had considerable influence in persuading the political writers of the press that they saw what was never in the election picture.” Richard L. Strout, a prominent national correspondent for the Christian Science Monitor, said the “authoritative verdict of the polls” seemed “overpowering” in advance of the election.
Krock and Strout may have been blame-shifting, a bit. But they were accurate in saying the polls had fortified conventional wisdom: Almost no one gave Truman any chance of winning in 1948, just as almost no one anticipated Mr. Trump’s victory in 2016.
Several weeks after Truman’s surprise win, pollsters were admonished by a committee of the Social Science Research Council for having “attempted the spectacular feat of predicting the winner without qualification. The presentation of the results gave the impression of certainty as to the outcome.”
It was a cautionary note that resonates today. Analogous criticism was directed at poll-based forecasters who declared Ms. Clinton’s election a near-certainty. A report prepared for the American Association for Public Opinion Research and released in 2017 said such forecasts “helped crystallize the belief that Clinton was a shoo-in for president.”
While no election or polling-based miscall is quite the same, the run-up to the 70th anniversary of the “Dewey defeats Truman” election offers a timely opportunity to revisit what was a polling and journalistic fiasco, and to ruminate about pitfalls that vex election forecasting.
Among those pitfalls are hubris and a disinclination to consider the unthinkable. Such temptations are not easily sidestepped, as the elections of 1948 — and 2016 — amply demonstrated. W. Joseph Campbell is a professor at American University in Washington, D.C., and the author of six books, including “Getting It Wrong: Debunking the Greatest Myths in American Journalism” (University of California Press, second ed., 2017). His Twitter handle is @wjosephcampbell.