With only hours left before election day in the United States, analysts and citizens are analyzing public opinion polls for hints of the outcome.
After months of campaigning and hundreds of polls conducted, the only clear takeaway has become the frustrating refrain echoed by election experts worldwide: “it’s too close to call.”
What’s driving this uncertainty? And how should we interpret the polls?
The Predictive Nature of Polling
Throughout the campaign, both parties have showcased various – and often conflicting – polling results. The issue is that it seems no one can agree on which polls are reliable.
Recently, the Des Moines Register released a poll led by well-regarded pollster Ann Selzer, indicating that Kamala Harris holds a surprising three-point lead over Donald Trump in Iowa, giving her campaign a much-needed boost.
However, shortly after, a “confidential” memo from the Trump campaign dismissed these findings, and Trump himself tweeted about favorable results from AtlasIntel, claiming leads across all seven swing states.
In recent elections, although there have been inaccuracies in key states – especially Wisconsin in 2016 and 2020 – polling averages have generally provided fairly accurate reflections of public sentiment.
Nevertheless, the current swing state margins, all within the margin of error, offer little insight that is not already known: the American electorate remains deeply divided over their options.
This brings us to an important point: polls are not inherently predictive. They reflect the public’s feelings at the moment the poll was conducted, which can offer educated guesses about potential election outcomes.
However, the margins of error (which are often greater than most realize), paired with closely contested final vote tallies in critical states and the winner-takes-all structure of the Electoral College, significantly curtail their predictive power.
The variability and direction of polling inaccuracies are unpredictable, especially since errors do not uniformly affect all regions and historically do not consistently favor one party over another.
The Impact of Small Errors
Accurate election polling in the U.S. is complicated by high non-response rates and the voluntary nature of voting (which necessitates adjustments based on projected voter turnout).
Misjudgments in these evaluations contributed to polling inaccuracies in both 2016 and 2020.
In 2016, polls notably underestimated Trump’s backing by neglecting to account for educational background in their sampling.
This oversight led to missing out on support from white, non-college-educated voters, who played a crucial role in turning the Midwest red.
While 2016 is often recalled as a disastrous failure of polling – for seemingly not anticipating a Trump win – the polling averages leading up to election day were, in fact, fairly accurate.
National polls were among the most precise in 80 years, only slightly overshooting Clinton’s popular vote margin by about one percentage point.
In the ten closest states of the 2016 election, Trump was underestimated by just 1.4% on average.
Significant misses in certain battleground states, such as Wisconsin, had a disproportionately large effect on the final outcome.
The polling inaccuracies were such that even a slight variation could swing traditionally Democratic states and lead to what many analysts deemed an unimaginable win for Trump.
Yet, it was the polling data – not analytical shortcomings or sensational media coverage – that received the brunt of the blame for failing to indicate that an unexpected result was, in fact, plausible, despite the clear margins of error suggesting otherwise.
Polling in 2020 was further off but faced less public criticism as it correctly predicted (though narrowly) a Biden win.
EPA/JIM LO SCALZO
Are Pollsters Addressing Former Flaws?
In examining polling averages for 2024, a uniform error favoring Trump of less than 0.8% across seven pivotal swing states could result in a decisive 312–226 Electoral College outcome for him.
Conversely, a similar shift of fewer than three percentage points in Harris’ favor could yield a solid 319–219 victory for the Democrats.
The pressing issue is whether pollsters have sufficiently corrected the flawed methodologies that previously underestimated Trump’s backing in 2016 and 2020.
Some polls have introduced changes such as increasing hybrid sampling techniques (blending both phone and online surveys) and better weightings based on prior voting behavior.
However, these adjustments might be overcompensating for earlier errors, potentially leading to an underrepresentation of Harris’ support due to ineffective sampling.
Until the ballots are counted and a winner is determined, certainty is elusive.
The Limitations of Polls
Polling can still be quite accurate, considering the challenge of capturing the views of millions. Yet, it is a mistake to view polling as inherently predictive or decisive.
The unpredictable nature of polling – and the corresponding complexities of the U.S. electoral system – makes it impractical to seek deeper insights beyond the general public sentiments that polling reflects.
Ultimately, the results could hinge on just a few votes or lead to a sizable Electoral College victory.
Regardless, remember not to hold the polls responsible.