A surprising look at polling accuracy, standout performances, and lingering challenges.
The 2024 presidential election revealed an unexpected trend: most public pollsters performed reasonably well in key swing states. This insight comes from WPA Intelligence, a polling and analytics firm led by Trevor Smith and Matt Knee. Known for its sharp critique of public polling methods, WPA’s November 19th newsletter acknowledged improvements in polling accuracy—though significant errors in Florida and Texas highlight lingering challenges.
In battleground states like Arizona, Georgia, Michigan, Nevada, North Carolina, Pennsylvania, and Wisconsin, polling misses were smaller than anticipated. According to WPA’s analysis, the average miss ranged from just 1.3% in Georgia to 2.6% in Arizona, all within the margin of error. This performance marked a significant improvement compared to prior election cycles.
However, most errors skewed in favor of Kamala Harris. Polls from firms such as CNN, Data for Progress, and The Hill/Emerson consistently overestimated her support. Exceptions to this pattern included Rasmussen Reports in Georgia and Wisconsin, as well as Echelon Insights in Pennsylvania, which slightly overestimated Trump’s performance.
Despite improvements in swing states, Florida and Texas stood out as significant outliers, with public pollsters underestimating Donald Trump’s support by wide margins. According to WPA Intelligence, the average polling miss in Florida was 6 points, while in Texas, the gap widened to a substantial 8 points. These represent some of the largest polling discrepancies in recent history for both states.
In Florida, the Cuban-American and Venezuelan-American voter blocs, which together comprise approximately 20% of the state’s Hispanic electorate, were pivotal in the election. Many public polls failed to account for Trump’s success in mobilizing these communities. His messaging on socialism and economic freedom resonated strongly, resulting in a 10% increase in voter turnout among these groups compared to 2020.
In Texas, Trump drew robust support from rural counties, where he won by a 30-point margin, and Hispanic voters in the Rio Grande Valley, where he improved his performance by 12 points compared to the last election. WPA Intelligence noted that private campaign polling captured these trends more accurately, identifying that nearly 60% of low-turnout voters in these regions favored Trump, defying traditional expectations.
These errors reflect broader challenges for public pollsters in high-turnout, demographically diverse states like Florida and Texas. Factors such as late-breaking shifts in voter preference, response biases, and underrepresentation of key demographics—such as rural and working-class voters—continue to impede accurate forecasting.
Marist earned WPA’s title of “Worst Pollster of the Year”. Their surveys in Michigan, Pennsylvania, and Wisconsin overstated Harris’s chances. For example, Marist predicted Harris winning Michigan by 3 points, but she lost by 2—a 5-point miss. Similarly, their Pennsylvania poll missed by four points, and their Wisconsin projection was off by three points. While these errors were within historical norms, they stood out in a cycle of generally improved accuracy.
Another surprise came from Ann Selzer, a pollster historically regarded as one of the nation’s most reliable. Her Iowa poll missed by an unprecedented 17 points. Selzer attributed the discrepancy to late-deciding voters and demographic shifts. However, WPA Intelligence criticized her reliance on outdated methodologies and failure to adopt modern techniques like education weighting, which have become industry standards.
The 2024 election underscored the need for continuous evolution in polling methodologies. According to WPA Intelligence, pollsters must adapt to shifting voter demographics, technological changes, and evolving turnout patterns. For example, Trump’s unexpected support among low-turnout and nontraditional voters reflected a broader trend that many public pollsters failed to capture.
WPA also highlighted persistent biases in survey design, particularly among firms like Morning Consult and The New York Times/Siena, which tended to overestimate Democratic support in key battleground states. This suggests a systemic issue with sampling techniques that disproportionately favor urban and suburban voters over rural ones.
To address these challenges, WPA recommends greater investment in voter contact methodologies, including multimode surveys that combine online, phone, and in-person sampling. In order to take into consideration late-breaking voter trends and demographic shifts that have the potential to drastically affect results, pollsters must also improve their models.
While public polling demonstrated significant improvement in swing states, glaring errors in Florida and Texas remind us of the industry’s ongoing challenges. WPA Intelligence’s analysis offers critical insights into what went right—and wrong—this cycle, reinforcing the need for modernized methodologies. For those seeking a deeper dive into the data, Trevor Smith, Matt Knee, and Amanda Iovino at WPA Intelligence remain open to discussion.