An Examination of US Election Poll Accuracy

For a significant portion of the 2024 US presidential campaign, both polls and political commentators deemed the contest too close to predict. Subsequently, Donald Trump secured a decisive victory over Kamala Harris, winning at least five battleground states and performing unexpectedly strongly in other areas. He is now set to become the first Republican in two decades to win the popular vote, and could assume office with a Republican-controlled House and Senate supporting him. This raises the question: were the polls inaccurate in portraying a tight competition? Nationally, polls certainly appeared to underestimate Trump for the third election in a row. However, in key battleground states such as Pennsylvania, which were the focus of most polling efforts, Trump’s winning margin was typically quite close to his predicted performance in the polls, even if the forecasts were slightly lower than the final outcome. The average polling error observed in these swing states was not particularly large. Nevertheless, in closely contested campaigns, minor differences can have a substantial impact. Prior to the results, media organizations, including the BBC, cautioned that despite the close figures in the polls, the election could ultimately appear as a landslide for either candidate, given the inherent margin of error. In some less scrutinized parts of the US, polls underestimated Trump’s support more considerably—a symptom of certain blind spots, according to Michael Bailey, a professor at Georgetown University and author of the book Polling at a Crossroads. “At a glance, in the battleground states, polls ran a little hot for Harris but really not so bad, but when you dig deeper, it’s all a little less impressive,” he said. In Florida, for instance, polls tracked by RealClearPolitics in the final weeks of the election indicated Trump leading by approximately five points. He ultimately won by a larger margin of 13. In New Jersey, Harris was projected to win by nearly 20 percentage points, based on the two most recent polls tracked by the site. Her actual winning margin was narrower, closer to 10. “Just imagine if we knew that or had a better sense of that a month ago. I don’t know that it would change the election but it would certainly change our expectations,” Prof Bailey said. He suggested that pollsters this year might have relied too heavily on assumptions that people would behave roughly as they did in 2020, failing to anticipate the extent of the shift among Latino and young voters toward Trump. “These models that assume so heavily that we’re going to get a repeat… they’re a disaster when there’s a big change,” he said. Conversely, pollster Nate Silver, founder of the 538 polling analysis site, noted that warning signs were present for those who were attentive. Writing about the election results, he pointed to a poll from New York City last month, which indicated Trump making major inroads in the traditional Democratic stronghold. “This is a problem the party should have been prepared for, because there was plenty of evidence for it in polls and election data,” he wrote. Discussion surrounding the polls is expected to persist in the coming months. This is particularly relevant in a year where figures like Trump and billionaire supporter Elon Musk promoted betting markets—many of which did forecast a decisive Trump victory—as a more accurate alternative. Experts stated that the polling profession indeed faces challenges. Response rates to surveys have plummeted, as it has become easier for individuals to screen calls from unknown numbers. This decline has also coincided with increasing distrust of media and institutions—a characteristic particularly pronounced among Trump supporters that some argue has led to their under-representation. Prof Bailey mentioned that the significant error in the much-discussed poll of Iowa by Ann Selzer—which was released days ahead of the election and indicated a three-point lead for Harris in the state—highlighted the risks associated with the traditional approach. To compensate for such issues, many of the most high-profile polls now function more like models, with firms weighting responses from different groups and making other assumptions about factors such as turnout. Many pollsters have also transitioned to using online surveys, but experts cautioned that these are known for being unreliable. This year, voters who were inclined to complete online polls were more likely to be Democrats, James Johnson of London-based polling firm JL Partners told the Times of London newspaper. They were “more likely to be young, they’re more likely to be highly-engaged, they’re more likely to be working from home,” he said. Prof Bailey argued that pollsters needed to “move on” from random samples and become comfortable with modelling, while simultaneously improving their testing and explanation of assumptions. However, Prof Jon Krosnick of Stanford University contended that without truly random samples, surveys would remain susceptible to error. Pollsters are “trying hard but they keep trying to be too clever,” he said. “What we need to do is go back to basics and spend the time and money that it takes to do polls accurately.” North America correspondent Anthony Zurcher provides analysis of the race for the White House in his twice-weekly US Election Unspun newsletter. Readers in the UK can sign up here. Those outside the UK can sign up here. Copyright 2024 BBC. All rights reserved. The BBC is not responsible for the content of external sites. Read about our approach to external linking.