When looking back at the 2016 election polls, one of the most prominent things that come to mind is how wrong they were – significantly off the mark. However, it’s worth noting that while pollsters underestimated Donald J. Trump’s support in crucial swing states that year, the national polls were fairly accurate in reflecting the popular vote, which Hillary Clinton won. The 2020 national polls were even less accurate, but they correctly predicted Joseph R. Biden Jr.’s lead and ultimate victory, making the discrepancy less memorable.
An analysis of polling averages over the years reveals that the polling in the last two election cycles was not unprecedented. However, with another tight presidential race this year, even a minor polling error could significantly impact the outcome.
National polling inaccuracies in presidential elections
Note: Arrows indicate the weighted national polling average margin compared to the final national popular vote margin.
Each election cycle sees some deviation between the polls and the actual results. This is inevitable as pollsters can only estimate who will turn out to vote, some voters make their decisions at the last minute, and unexpected events can occur late in the race.
Data spanning the past four decades demonstrates that polling biases do not consistently favor one party over another, and past performance is not indicative of future polling accuracy. For instance, the polls for the 2022 midterms were remarkably accurate.
To assess the reliability of presidential polls, the charts in this article depict averages that consolidate multiple polls into a single estimate for each election. Between 1988 and 2020, the final national polling average was off by an average of 2.3 percentage points.
Some years fared better than others: in 2008, the national polling average missed Barack Obama’s final result by less than one percentage point on average, while in 1996, it overestimated support for Bill Clinton by nearly four points.
State-level polls have not been as accurate. Since 2000, polls in closely contested states have been off by an average of 3.1 points. In both 2016 and 2020, the majority of state-level polling averages underestimated support for Mr. Trump, sometimes significantly.
State polling inaccuracies in presidential elections featuring Trump
Wisconsin experienced one of the most significant polling inaccuracies in recent election cycles, overestimating Mr. Biden’s support by 9 points.
Several states that ultimately favored Mr. Trump had polling averages indicating Mrs. Clinton in the lead.
The polling errors at the state level have been magnified in the last two presidential elections, both of which were closely contested races that underscored the significance of the Electoral College. Nonetheless, the polling performance in those years was not entirely anomalous. An examination of state polling inaccuracies since 2000 reveals that polls from previous elections performed comparably to today’s polls.
State polling inaccuracies in presidential elections
State polls underestimated Mr. Obama, except in South Carolina where Mitt Romney emerged victorious.
Polls indicated John McCain leading in Indiana, but Mr. Obama secured the state.
The polls presented a mixed picture, underestimating George W. Bush in some cases and John Kerry in others.
Al Gore won the popular vote but lost the election. Polls favored Mr. Bush in Florida, although the results were extremely close.
State polls have exhibited inaccuracies in both directions over the years. Given that pollsters underestimated Mr. Trump’s support in the last two elections, the question arises whether they will repeat this pattern in the current year. Some observers suggest mentally adjusting poll numbers to factor in a boost for Mr. Trump based on past experiences.
Pollsters believe they have identified the main reasons for the polling inaccuracies in 2016. A key factor was the failure to consider voters’ educational backgrounds, as highlighted in a report from a professional organization of public pollsters. The state-level polls in that year, in particular, overrepresented college-educated respondents and underrepresented those without a college degree.
This issue was less significant in previous elections when voting patterns were not as closely aligned with educational attainment. However, in 2016 and onwards, non-college-educated voters, who largely supported Republicans, especially Mr. Trump, were not accurately represented in the polls.
By 2020, most pollsters had adjusted their methodologies to account for education levels. Despite this, the polls still underestimated Mr. Trump’s support. One theory, proposed in a report evaluating the 2020 polls, suggests that Trump supporters were less likely to participate in surveys, leading to a bias in the sample. Chris Jackson, who leads U.S. public opinion research at Ipsos, noted that even after controlling for demographics, survey respondents were more Democratic-leaning.
Another challenge in 2020 was the unprecedented voter turnout. Approximately a quarter of voters in 2020 did not cast ballots in 2016, according to estimates from the Pew Research Center. Initially, polls suggested that new voters in 2020 would predominantly support Biden, but the actual split between Biden and Trump among these new voters was more evenly divided.
Looking ahead to this year, polling in seven swing states is incredibly tight, with Mr. Trump and Kamala Harris virtually neck and neck in most of these states. While the polls underestimated Mr. Trump’s support in the previous two cycles, historical data shows a mixed track record for these states, with inaccuracies on both ends of the political spectrum and varying levels of accuracy in different years.
Previous polling inaccuracies in battleground states during presidential elections
Bold years indicate instances where the polling leader lost in that state.
Note: Averages may not be available for all states in every year.
Most pollsters have made adjustments to their methodologies, partly in response to the polling errors in 2020, and the accurate 2022 midterm polls have instilled some confidence among pollsters.
One significant difference between the midterms and the last two presidential elections is the absence of Mr. Trump on the ballot. It is possible that there is something unique about Mr. Trump’s supporters that makes them particularly challenging to gauge in polls.
“There’s a real possibility that having Trump on the ballot makes it hard to poll for whatever reason,” stated Berwood Yost, the director of the Franklin & Marshall College Poll of Pennsylvania.
“But the other possibility is that we’ve made all these adjustments to correct the error. And perhaps we’ve overcorrected,” he added. “I don’t know which it is.”
