Given the close results in each of the seven presidential battlegrounds, several pollsters and polling aggregators are already stating that their pre-election surveys were accurate as they perform their postmortems of the 2024 presidential election.
However, one important aspect that cannot be overlooked is that, despite the numerous adjustments pollsters made following their 2020 and 2016 misses, the polls once again underestimated the level of support for President-elect Donald Trump.
According to an NBC News Decision Desk analysis, the results in the swing states were close enough to fall inside the margin of error for a significant number of polls, and overall, this year’s polls appear to have fallen short by less than they did four years ago.
Although we arguably shouldn’t have been surprised, several people believed that the conclusion was unexpected because pre-election polling averages did not reflect Trump’s sweep of the swing states due to the absence of Trump supporters in public polls.
NBC News contrasted the proportion of votes that Trump received in the state and national elections with his support in likely voter polls that were conducted in October and November. Pre-election polls from the last two presidential elections showed a similar pattern: Nearly everywhere, the average poll underestimated Trump’s support; in the seven swing states, the shortfall was frequently between two and three percentage points.
The fact that polling averages in state-level presidential contests performed marginally better than they did in 2020 may give pollsters some solace, possibly indicating that polling tweaks helped reduce the overall polling error. Pre-election polls conducted in the final two weeks of the 2020 election undervalued Trump’s support by an average of 3.3 percentage points when compared to the final results, according to a study of all publicly reported polls during that time. The latest two weeks’ polls in 2024 understated his popularity by an average of 2.4 points.
The number of polls taken in October and November, as well as the discrepancy between the polls and the national results, are displayed in the chart below.
This underestimation of Trump’s support was pervasive in polls from swing states (Nevada, 2.9 percent too low), solid Republican states (Wyoming, 5.8 points too low), states that were allegedly trending purple in Democrats’ dreams (Texas, 4.4 points too low), and solid Democratic states (New York, 4.6 points too low). This is in spite of pollsters’ efforts to take into consideration the challenge of obtaining Trump supporters to participate in their surveys.
Depending on where you look, the polls may or may not have accurately predicted the race’s mood. Numerous surveys in Arizona, Georgia, North Carolina, Michigan, Nevada, Pennsylvania, and Wisconsin—all crucial swing states—suggested a race that was deadlocked or had a 1-point margin. Considering the sample error margins and additional polling error sources, such figures provided plenty of leeway for the final results to fluctuate. However, others were taken aback when Trump won all seven states due of the sharply divided polling.
Trump was properly predicted by the majority of polls to win Arizona, Georgia, and North Carolina, but not Michigan, Nevada, Pennsylvania, or Wisconsin. Furthermore, a significant portion of surveys underestimated Trump’s support by more than the margin of error, even when swing state polls accurately showed him as the front-runner.
For instance, 36% of Arizona surveys undervalued Trump’s lead by more than the margin of error, despite the fact that he led in 85% of the state polls.
Reasons for the divergence
What then took place? There are probably two culprits.
The first possibility is that the surveys were unable to include enough new voters or people who switched from supporting Biden in 2020 to supporting Trump in 2024. As we witnessed in 2020, Trump may have once again galvanized voters who were prepared to cast ballots but unwilling to speak with pollsters.
It is difficult, if not impossible, to gauge public opinion if poll respondents have different opinions from those who do not, particularly in ways or to a degree that pollsters do not expect. Voters may decide to completely ignore journalists or pollsters if they feel mistreated or misunderstood when they express their opinions.
Second, pollsters may have simply made incorrect assumptions when attempting to predict the demographics of the 2024 electorate, which could have easily led to a polling error similar to the one we just witnessed. This exemplifies a challenge specific to pre-election polling: the requirement for pollsters to modify their data to reflect their expectations of the public, without knowing if those adjustments are accurate.
However, the changes might not have been sufficient if the 2024 electorate had evolved in ways that pollsters had not anticipated.
For instance, several pollsters began using weighting in 2024 to ensure that respondents’ self-reported previous vote—that is, whether they stated that they had voted for Trump or Biden in the most recent election—matched the results of 2020. Pollsters utilized this cycle’s statistical correction to make up for the prior undercount of Trump supporters. Pollsters assumed that 2020 Biden and Trump voters would vote at the same rate in 2024 when comparing polls to match the popular vote from four years ago. However, if Trump voters were slightly more likely to vote and Biden voters were slightly less likely to vote, that could easily result in a two-point understatement in Trump’s support.
Where do we go from here?
Recently, Nate Silver expressed the opinion that polling is not the issue.We concur. The way polls are presented and understood is the issue, not the polling itself. Indeed, it is astounding that a pollster can speak with the 800 respondents and arrive with a conclusion that is only a few points off the results of an election that received close to 150 million votes.
Issues arise because people expect polls to do more than they can, such determining who is ahead in a close race or spotting slight shifts in a candidate’s support. Because polling findings are frequently discussed and presented in ways that imply they are more accurate than they actually are—often in visuals that show candidates separated by only a few decimal places—people believe polls can do this. The to-the-tenth-of-a-point graphics and averages provide the impression of accuracy.
Even while 2024 saw improvements in poll journalism, much of the media discourse still created the sense that polling was a surgically exact tool for analyzing political events and campaigns, with more coverage mentioning the margin of error alongside poll results. In actuality, polling is more like a butter knife than a scalpel; you can get quite close, but even with skill, accuracy requires some luck.
The fact that the majority of pollsters do not reveal how they get and modify the data is another issue that makes pre-election polling hard to understand. It becomes extremely difficult to assess or compare the results if one does not know the source of the data, particularly how pollsters have weighted and altered their data in an attempt to forecast the characteristics of the voter. It is impossible to determine how much the results reflect the choices made by pollsters, voters, or both when sensible weighting choices might shift a poll’s margin by up to 8 points.
Should we abandon pre-election polling (like the Pew Research Center and Gallup have done) in light of these worries and the polling failures that we witnessed once more in 2024? Despite the temptation, this is not the best course of action. When properly interpreted, pre-election polls can contribute significantly to democracy by giving an indication of the outcomes that appear likely. The likelihood that either candidate may win was underscored by the fact that the majority of polls indicated a close election in 2024, which may have contributed to the public’s increased acceptance of the outcome.
But change is necessary. As industry groups like the American Association of Public Opinion Researchers have advocated for greater transparency, it’s also critical to have a more modest viewpoint of the lessons that may be drawn from surveys. Although polls can be used to determine which problems voters find more or less important, they will never be able to predict the outcome of contests with a margin of 1 to 2. And those are frequently the contests we have in our deeply divided country.
Additionally, pollsters should be more open about how their decisions impact the results that are published. It appears wise to demonstrate how alternative plausible options matter, even if pollsters wish to draw attention to their assessment of the best estimate given their expertise and experience. It is crucial to determine whether various, acceptable judgments result in significantly different estimations because it is impossible to determine which choices are optimal until after the fact.A range of possible outcomes could be more effectively communicated by seeing how the results alter under different conceivable scenarios, such as high Republican and low Democratic turnout, or vice versa.
Polling before an election is challenging. That is a fact and not an excuse. Treating pre-election polls as revealing deep, knowable truthswithoutacknowledging the uncertainty inherent in those polls risks mistaken interpretations, media cycles driven by specious numbers, and the loss of public faith in the judgement and expertise of those involved in polling and analysis.
We are still expecting too much from an instrument that is too blunt, even though pollsters can legitimately claim that their surveys were in the ballpark of the outcome and that the polls performed better in 2024 than in 2020. Rather, we should think about how we may use pre-election surveys to properly characterize the uncertainty involved and communicate the electoral possibilities at stake.
Note: Every piece of content is rigorously reviewed by our team of experienced writers and editors to ensure its accuracy. Our writers use credible sources and adhere to strict fact-checking protocols to verify all claims and data before publication. If an error is identified, we promptly correct it and strive for transparency in all updates, feel free to reach out to us via email. We appreciate your trust and support!