The Polling Problem in This Year's Presidential Election

Polling in this year’s presidential election has proven to be unhelpful in predicting the winner with any accuracy.  In part this has to do with the unusual closeness of the contest for most of the year.  (Only in the immediate aftermath of the presidential debate between Joe Biden and Donald Trump was there a significant gap between the two major-party candidates.)  The limited value of the polls this year also stems from the decline in polling itself as a reliable guide to public opinion about the election.

There was a time when polling in presidential campaigns could be trusted to be reasonably accurate.  In the mid 1930s, Archibald Crossley, George Gallup, and Elmo Roper were among the first to develop representative samples of the electorate as a whole, and to combine telephone surveys with in-person interviewing so as to make certain polling response rates were high.  By the 1940s, public opinion research had entered a new, more reliable era.  Mistakes sometimes were made, as in 1948 when pollsters stopped surveying voters before the campaign ended because Republican challenger Thomas Dewey had consistently led by a substantial margin over the embattled incumbent, Harry Truman.  As a result, pollsters missed a late swing toward Truman (who ultimately won), but that sort of goof became rare by the 1950s.  What dawned then was a period in which response rates were very high, population samples were representative, and polling was usually very accurate.  Much of the reason for that had to do with the increasing ubiquity of land lines in American homes due to broad post-World War II prosperity and low rates for local telephone usage.  The pattern of a phone in every home made Americans easier to reach, as did the very high likelihood then that at least one adult was there during the daytime and so could answer pollsters’ questions.

That polling heyday lasted into the 1980s, which created an expectation among presidential election watchers that they could know with real certainty who was ahead during the campaign and thus likely to prevail on Election Day.  The big exceptions to that situation were contests in which the major-party candidates were neck and neck for most of the campaign, or near its end, but those were unusual.  1960 was the best example of the first situation, with both Kennedy and Nixon more or less even for most of the campaign.  Examples of the second, where the polls indicated (correctly) that the race had tightened so much by the final week as to become too close to call, were 1968 and 1976.  (In both of those elections, the candidate trying to come from behind fell short, but only just barely.)

What undermined polling accuracy by the start of the 1990s were four things: ever more households with no adult at home during the daytime; a growing tendency to screen calls as telephone solicitations became annoyingly common; and, in some ways most important, the related decline of land lines and rise of cell phones (the former had numbers easily accessed via telephone directories, unlike the latter).  With those changes, pollsters began to have much more trouble reaching adult Americans, and polling response rates began to decline.  Pollsters have tried to cope with that situation by coming up with ever-better sampling techniques, so that even with low response rates polling could show the state of the presidential contest accurately, but that was hard to do.  Greater volatility in turnout since the 1990s compounded that problem, as figuring out which kinds of people were most likely to vote became more difficult, which complicated the effort to produce representative samples of the electorate.

 As a result, polling in recent years has become less reliable even though many older Americans (who came of age during the era of accurate polls) tend to expect such surveys to provide a correct reading of public opinion.  Instead, a new era dawned in the 2000s characterized by unreliable polls and Election Day surprises.  Polling by then had become an international industry, with the USA and United Kingdom in particular using the same firms to predict national election results.  Among the most notable polling goofs in recent years have been the outcome of the June 2016 Brexit referendum in Britain and the Clinton-Trump presidential contest later that year, as well as the 2020 and 2022 national elections in the U.S.  In 2020, Trump did much better than the polls indicated that he would (although he still lost).  In 2022, it was the Democrats who over-performed (when compared with the polls) in the off-year congressional elections.  Pollsters claim to have learned from these experiences and corrected their techniques to avoid repeating them, but even the pollsters cannot be sure they have succeeded in doing so.

All of these factors have converged during this year’s presidential election campaign.  The unreliability of polling in general combined with what seems to be a persistently close race (according to most polls) means that no one knows for sure who is ahead or will likely win.  The unappealing nature of the choice between Kamala Harris and Donald Trump for many swing voters has made estimating who will actually vote even harder than usual.  And while early voting has already begun and Election Day (November 5^th) is fast approaching, there is still time for an unexpected development to determine the outcome, if the race is as close as it seems to be.  And in such important swing states as Pennsylvania and Wisconsin, rules against counting early votes before Election Day mean slow tabulation there, which could prolong the uncertainty for days after voting ends.  What this situation demands from all concerned is a kind of patience that is unusual for a fast-paced society like the United States.  How well Americans cope with that challenge remains to be seen.