4 minPolitics
Pollsters repeatedly miss election outcomes by fighting the last war
Opinion writer traces a long record of polling failures, from the 1936 Literary Digest miss to recent primary errors, and argues that human modeling choices, not technology, drive most errors.
Polling has repeatedly failed to predict American election outcomes, and the errors often stem from human decisions about how to measure the electorate rather than from the tools themselves. That is the central argument of an opinion piece tracing the field's record from the early twentieth century to the present.
The author points to the 1936 presidential race as a cautionary starting point. The Literary Digest, which had correctly forecast winners since 1916, predicted that Alf Landon would beat Franklin Delano Roosevelt. Roosevelt won handily. The magazine had surveyed about 10 million people by postcard, drawing names from telephone books and car registration lists. Nearly 2.4 million responded, but the sample skewed toward voters who were more affluent than the broader electorate, and the Digest folded shortly afterward. Gallup began tracking presidential approval ratings in 1938, but the science remained imperfect. In 1948, the Chicago Daily Tribune's front-page headline «Dewey Defeats Truman» rested in part on earlier polls that did not capture the late shift on Election Day.
Campaigns gradually came to depend on pollsters. John F. Kennedy was the first president strongly associated with a polling firm, using Lou Harris and his organization in the 1960 race. In the 1970s a generation of so-called celebrity pollsters emerged, including Pat Caddell for Jimmy Carter, Richard Wirthlin for Ronald Reagan, and Stan Greenberg and Dick Morris for Bill Clinton. The author argues these figures were successful but not infallible. Caddell was largely responsible for Carter's 1979 «Crisis of Confidence» speech, which misjudged the public mood and became widely criticized as a damaging address.
The piece describes a series of later misses. In 2004, exit polls predicted a win for John Kerry over George W. Bush, and Bush won a second term. In 2012, Mitt Romney's campaign believed it was headed for victory after its pollster made overly optimistic weighting assumptions, including an incorrect expectation that Barack Obama would not repeat his 2008 levels of youth and minority turnout. In 2016, pollsters confidently predicted that Hillary Clinton would defeat Donald Trump, and she did not win the election. The author notes that current midterm primaries have produced a new kind of error, with partisan polls overstating progressive support and, in some cases, causing candidates to drop out based on inaccurate data.
The author contends that modern polling faces harder conditions than in 1936. Nearly everyone has a phone, but many people screen calls from unfamiliar numbers, shrinking response rates and enlarging what the author calls the «silent majority.» In that environment, the piece argues, human modeling decisions matter more, not less. Errors arise from imperfect judgments about how to deploy new technologies, and pollsters can overcorrect for what went wrong in the previous cycle, much like generals fighting the last war.
The author's conclusion is that polls are more reliable as snapshots of opinion than as predictors of behavior, particularly of who will actually cast a ballot. Readers, the piece suggests, should treat poll results with caution and focus on turnout, which ultimately decides elections.
7
