Throughout the 2024 election cycle, the consensus among pollsters and analysts was that the race was neck-and-neck. Donald Trump and Kamala Harris were projected to be in a tight competition, with polls indicating a narrow margin that could tip either way. But when results rolled in, it became clear that Trump had achieved a commanding victory. His win included several battleground states, with notable gains even in regions where he had previously struggled. This outcome raised questions about the reliability of pre-election polling, especially as Trump is now positioned to become the first Republican to secure the popular vote in two decades.
The polls, it seems, had again misjudged the extent of Trump’s support, especially in some key areas. This is now the third consecutive election where polling data has underestimated his reach. Nationally, polling firms had portrayed the race as close, with a slight advantage for Harris in several states. Yet, in many of these battlegrounds, Trump either met or exceeded projections. Pennsylvania, for instance, saw results within a few points of polling expectations, though Trump ultimately secured the state more comfortably than many analysts anticipated.
Polling experts observed that while average polling errors in battlegrounds weren’t massive, even small inaccuracies proved impactful. As many noted prior to election day, a narrow polling margin could mean an apparent landslide win for either side once votes were counted, largely because of built-in error margins. The closeness of pre-election predictions contrasted sharply with what became a substantial win for Trump.
In states that received less polling attention, Trump’s support was underestimated more noticeably. These “blind spots” revealed a significant disconnect between polling practices and on-the-ground trends. Michael Bailey, a professor of political science, noted that, at first glance, polling appeared to be fairly accurate in battleground states, yet a deeper dive showed a less promising picture for pollsters.
Taking Florida as an example: polling averages in the weeks leading up to election night showed Trump with a slim five-point lead. Ultimately, he won by a substantial 13-point margin. Similarly, in New Jersey, polls had given Harris nearly a 20-point lead, but she won by only about half that margin. “If this shift had been recognized a month earlier, it likely wouldn’t have altered the final outcome of the election, but it definitely would have influenced public perception of the race,” Bailey commented.
Several analysts pointed to the reliance of pollsters on assumptions rooted in the 2020 election. The prevailing belief that voter behavior would largely mirror 2020 proved faulty, especially with certain demographic shifts favoring Trump. Notably, polling missed increased support among Latino and younger voters, who swung more toward Trump than anticipated. Polling models that heavily assumed similar voting behavior as in past cycles can struggle when there’s a significant shift.
Signs of an unexpected Trump performance were indeed present in some less-publicized polls, but they were not widely discussed. Some polls, for instance, showed Trump making significant gains in traditionally Democratic strongholds, an early indication of his growing support. Analysts have since pointed out that these signs were visible in certain data but were overlooked or underreported.
As discussions around polling errors continue, attention has also turned to alternative forecasting tools. Betting markets, which have been promoted by several public figures, including Trump supporters, predicted a Trump win more accurately than most traditional polls. While betting markets are not without their own risks, they have started gaining recognition for capturing electoral outcomes, perhaps because they incorporate a wider range of factors than standard polling.
Polling in the digital age faces numerous obstacles. With fewer people responding to phone calls from unknown numbers, pollsters struggle to maintain representative samples. Furthermore, mistrust in media and government institutions is particularly pronounced among Trump’s supporters, possibly leading to their under-representation in surveys. In Iowa, a well-known poll indicated a slim Harris lead, but Trump carried the state decisively. Such outcomes reflect the growing unpredictability of voter behavior and the challenge of accounting for these shifts in a traditional polling model.
Many high-profile polls now rely on modeling to account for gaps in response rates, weighing responses based on demographics, turnout assumptions, and other factors. However, this approach still has critics who argue that these models can add complexities without necessarily improving accuracy. Some polling organizations have shifted to online surveys, but these too have their drawbacks. Voters who participate in online polls tend to skew Democratic, representing younger, more engaged demographics, which can produce a distorted picture of the electorate.
The future of polling may well involve a combination of traditional methods and more sophisticated models that are rigorously tested. Pollsters may need to work on refining sampling techniques to better capture demographic shifts, as well as to adapt to the changing political landscape. Experts suggest that while modeling is essential in today’s environment, pollsters must balance these innovations with a solid foundation in data gathering and analysis.
Jon Krosnick, a professor at Stanford University, advocates for a return to basics, emphasizing that polls must prioritize accuracy over complex modeling. Without truly representative sampling, he argues, polls will continue to struggle with reliability. Krosnick believes pollsters are “trying to be too clever” and suggests that an investment in high-quality data collection could lead to more dependable results.
The 2024 election underscores the need for ongoing adaptation within the polling industry. For polling to maintain its role as a trusted source of insight, experts say, it must find new ways to meet the challenges of changing voter behavior, technological barriers, and declining trust.