For the third election in a row, polling companies have underestimated support for Trump. Despite efforts to correct past wrongdoings, significant blind spots remain.
Just a day ago, the outlook seemed straightforward: the race for the White House would be downright slim, a quick result Wednesday morning seemed unlikely, and a days-long deadlock, mirroring the stalemate of four years earlier, seemed entirely possible. Switzerland NZZ also described it as a “nail biter”.
This forecast was backed up by data from trusted poll aggregators. While models from The Economistthe website for political analysis FiveThirtyEight and pollster Nate Silver showed Kamala Harris with a small national lead, they still labeled the race a “toss up.”
But on election night, a distinct trend favoring Republican Donald Trump emerged unexpectedly early.
Within the margin, but missed the target
Of the seven key states that would decide the election, Trump secured early victories in North Carolina and Georgia. In the latest polls, he was expected to have small leads of 1.2 and 1.3 percentage points over Kamala Harris. With 99% of the vote counted, Trump beat those projections, winning North Carolina by 3.4 points and Georgia by 2.3 points — returning Georgia to red after its 2020 endorsement of Joe Biden.
Then the “blue wall” began to collapse. First, Pennsylvania, which the Democrats had captured in 2020, went to Trump. Before the election, the state was considered the crown jewel of the swing states, with polls predicting Trump to beat Harris by a slim margin of 0.4 points. Instead, his lead grew to 2.2 points when almost all the votes were counted.
In Wisconsin and Michigan, pollsters had expected a narrow Harris victory, but the former president prevailed there as well. Both states are now firmly in Trump’s column. With votes still being counted in Nevada and Arizona, Trump has commanding leads in both. If these results hold, Trump will have swept every contested state, each time by a larger margin than the polls had predicted.
Still, the pollsters are likely to exhale. By nature, polls are inherently imprecise, based as they are on samples that represent the larger voting population. This sample introduces some uncertainty, as the results are extrapolated across a diverse electorate. Polls before election day had an average margin of error of ±3 percentage points. With the exception of Nevada, all halftime scores fell within this range; with the exception of Michigan and Wisconsin, most states were correctly assigned to Trump in pre-election analyses.
Are Trump Voters Too Poll-Shy?
It wasn’t always like that. Trump’s support was systematically underestimated in both 2016 and 2020. In 2016, polls predicted narrow wins for Hillary Clinton in the battleground states of Pennsylvania, Michigan, and Wisconsin. Four years later, polls again overestimated Biden’s support, missing the mark by about four percentage points nationally and in many states.
An expert panel from the American Association of Pollsters pointed to three reasons behind these incorrect predictions: First, a significant number of undecided voters swung toward Trump shortly before the election. Second, voters without a college degree were underrepresented in the sample. Third, some respondents chose not to identify as Trump supporters—either they distrusted polls or felt embarrassed by their choice. This likely led to a sample skewed toward Democrats, the panel concluded.
In response, many polling firms adjusted by placing more weight on educational demographics and asking respondents about their voting history. The hope was that past Trump voters would serve as an indicator for future predictions.
Misjudge Latinos, Black voters and women
Whether these adjustments helped avoid major miscalculations this time remains uncertain. Preliminary results indicate that the vote was more accurate than in the previous election. But early analyses indicates a shift: Latino and black men voted for Trump in greater numbers than they did four years ago. Meanwhile, the expected wave of support for Kamala Harris among women never materialized. Polling agencies may have underestimated the appeal Trump held among groups traditionally aligned with Democrats.
Questions also swirled about whether the “Bradley effect” might be affecting Harris. This phenomenon, in which respondents overestimate support for female or nonwhite candidates in public opinion polls, can result in such candidates underperforming on Election Day compared to their poll numbers.
Whether accurate surveys are based on the sincerity of the respondents can be seen from the experiments with new survey methods. Many polls are now done online rather than by phone or in person, but studies indicates that important information is often omitted or incorrectly reported in online formats. This can skew samples by overrepresenting certain groups, skewing the data.
Finally, pollsters face a constant challenge. They cannot predict with certainty whether the respondents will actually vote. Despite all the metrics, only election day reveals who shows up at the polls and how representative the sample was. Again, Trump’s support was underestimated.
Latest articles
Global reporting. Journalism of Swiss quality.
In today’s increasingly polarized media market, Switzerland-based NZZ offers a critical and fact-based view from the outside. We are not in the latest industry. We offer thoughtful, well-researched stories and analysis that go behind the headlines to explain relevant events in the US, in Europe and around the world. To produce this work, NZZ has an industry-leading network of expert reporters around the world who work close to our main newsroom in Zurich.
Sign up for our free newsletter or follow us further Chirp, Facebook or WhatsApp.
