During much of the , we often heard that the contest would be too close to predict the final result.
However, on Wednesday (06/11) the results showed Donald Trump against . He won at least six and performed unexpectedly well elsewhere.
The Republican had 50.7% of the votes and won 295 delegates; the Democrat had 47.6% of the votes and took 226 Electoral College delegates.
Now, he is poised to become the first Republican in two decades to win the popular vote and could take over the White House with a House and Senate controlled by supporters.
So were the polls wrong to detect a close race?
On a national level, they certainly seemed to underestimate Trump for the third election in a row.
But in swing states like Pennsylvania, Trump’s margin of victory was typically close to his performance in the polls, even when they showed lower numbers than the final result.
The average difference between what the research showed and the final result was not really that big. However, in heated campaigns, small changes can make a big difference.
Before the results, media outlets such as the BBC warned that, despite the close race portrayed by the polls, it could end up being a landslide victory for either candidate, given the margin of error.
In some less-watched parts of the U.S., polls have underestimated support for Trump more significantly — a sign that there are some blind spots, according to Michael Bailey, a professor at Georgetown University.
In Florida, for example, polls tracked by RealClearPolitics in the final weeks of the election had Trump ahead by about five percentage points. He won by a margin greater than 13 points.
In New Jersey, Kamala was expected to win by nearly 20 percentage points, based on the two most recent polls tracked by the site. His margin was narrower, closer to 10.
“Imagine if we knew this or had a better sense of this a month ago. I don’t know if it would change the election, but it would certainly change our expectations,” says Bailey.
For the professor, research institutes may have repeated failures observed in 2020, failing to capture, for example, the depth of the change in favor of Trump among Latino and young voters.
But researcher Nate Silver, founder of survey aggregator 538, says there were warning signs for those looking in-depth at the data.
He cites a New York City poll from last month that indicated Trump was making gains in traditional Democratic strongholds.
“This is a problem that the party [Democrata] should have been prepared, because there was plenty of evidence of this in polls and election data,” Silver wrote.
The debate over election polls will certainly continue in the coming months.
This is particularly sensitive in a year when figures like Trump and his billionaire backer Elon Musk have promoted bets — many of which assumed a Trump victory — as a more accurate alternative.
Experts recognize that the electoral research market faces challenges.
Survey response rates dropped as it became easier for people to identify calls from unknown numbers.
The drop also coincides with growing distrust of the media and institutions — a trait particularly present among Trump supporters, which some argue has led to the underrepresentation of these voters.
Bailey says the big error in pollster Ann Selzer’s much-discussed Iowa poll — which was released days before the election and showed a three-point lead for Kamala in the state — showed the risks of the traditional approach.
To compensate for these problems, many surveys are adopting models, which can weight responses from different groups and make assumptions about factors such as turnout.
Many are also moving toward online surveys, although others in the field are reluctant about how reliable they are.
This year, voters who were inclined to answer online surveys were more likely to be Democrats, James Johnson of London-based polling firm JL Partners told the Times of London.
They were “more likely to be young, more likely to be highly engaged, more likely to work from home,” he said.
Bailey says researchers have had to overcome random sampling and become more intimate with the models, as well as improve tests and explanations for their assumptions.
But Stanford University professor Jon Krosnick says that without truly random samples, surveys remain vulnerable to error.
“What we need to do is go back to basics and spend the time and money necessary to do accurate research,” argues Krosnick.
