The last three general elections held in the U.S. have felt pivotal. In 2016, current president Donald Trump won the electoral college but not the popular vote. In 2020 he lost both to Joe Biden. Most recently, in 2024, he reclaimed the presidency with both electoral and popular majorities. Were these results expected by the professional pollsters of the United States? Did the country have an accurate idea of the likely outcome of the election? Was the population satisfied — if not with the president-elect himself, at the very least with the predictions of pollsters?
No, not in the slightest.
Something is going wrong within the U.S. polling system. While polls were correct regarding Hillary Clinton’s victory in the popular vote in 2016, they were wrong pretty much across the board in critically important swing states. In fact, many political scientists were not even looking at the candidates specifically; rather, they were looking at opinions on things like the economy, historical voting, and partisan breakdown of the electorate. They failed to account for the drastic rise in identity politics generated by Trump’s campaign. They did not account for the drastic decrease in voter turnout that year — Wisconsin was considered a safe state for Clinton, but with the lowest turnout in 16 years, Trump won it by 27,000 votes. The people did not like Donald Trump, but those who voted for him disliked Clinton even more.
It seems that identity politics and dissatisfaction with presidential candidates presents just as much of a problem for pollsters as their research methods. The Pew Research Center believes that the 2016 and 2020 polls were skewed, as the recruitment cohorts for polls have had fewer and fewer registered Republicans. This is due to potential unwillingness to participate in polls and changes in research methods and the technology associated with them.
Despite general accuracy in the 2018 midterm polls, the public was, and remains, mistrustful of polls. This proved truer among conservatives than liberals, a trend seen often on the right — conservative former president Nixon often referred to the “silent majority” left unvoiced in polls. Even Trump leans into this, telling an interviewer at CNN: “Everybody loves me.” Although the polls did correctly predict the headlines in 2020 — Biden victories in the popular vote and electoral college, and a Democratic Senate — they overestimated many figures, from Biden’s nationwide popular vote to his performance in swing states and statewide elections. To combat this, pollsters have been expanding the use of weighting, a practice in which pollsters deliberately overvalue the opinions of groups their poll may have underrepresented.
Pollsters hold a range of opinions about the downsides of their own research methods. Many say that reliance on now-outdated phone calls hinders research and accuracy, while others dislike the use of online surveys, which are often argued to provide lower quality data. In fact, it seems that the reliance on telephone calls has caused inaccuracies in polling that then affect the nation at large. The New York Times’ own phone polls only yielded a full interview 0.4% of the time in 2022, down from an already-small 1.6% in 2018. This low number is reduced even further due to necessary screenings that ensure the sample of respondents is representative of the American people. Given these unrepresentative numbers, is it possible to paint an accurate picture of public opinion?
Another question that we need to ask of these polls is whether they themselves ask the right questions for the current state of American politics. Identity has become a “lightning rod,” but still there was not much research as of 2021 into how identity affects political choices. This oversight likely had an effect on the inaccuracies of 2016 and 2020 predictions. However, this seems not to account for the monumental overshot from the usually reliable poll published by The Des Moines Register in 2024. Trump had the victory in Iowa with a 13-point margin, 16 points away from what J. Ann Selzer had predicted. However, her poll does not weight identity nearly as much as other polls. This means we are left with the question of how the 2024 polls incorrectly predicted a Harris win, and how the Iowa poll missed the mark by so much, when it has a usually accurate history.
It is likely that we won’t get an answer to these questions for a few years, or until pollsters have had ample time to extensively investigate their research methods. However, it is likely that the power of identity politics, and how the candidates for the 2024 election stirred that pot, played into the mistakes made by polls.
Furthermore, it seems that this issue is now here to stay. Future polls will need to take more than just economic prospects into account; they will have to focus on how much the American population like, trust, and relate to the identity of the candidate running, rather than their party affiliation.
