British ‘mega poll’ leans toward Harris victory

The British are a-coming. This time, they’re wielding a new polling technique with a huge survey sample that some say is more accurate than anything currently in widespread use in U.S. politics — and it’s tentatively predicting a Kamala Harris win.

Focaldata, a U.K. polling company, questioned more than 31,000 voters across the United States over the past month for an innovative type of survey called “MRP” that has become influential in Britain in recent years.

Combining the MRP results with their large-scale online swing state polling Focaldata assesses Harris is likely to take Michigan, with a lead of nearly 5 points, Nevada with a lead of about 2 points over Trump, and with a slight edge in Pennsylvania and Wisconsin.

The Focaldata research suggests Trump just scraping past Harris in Georgia and North Carolina. The polling data and MRP modeling are split on Arizona.

Harris’ advantage comes in part from her strong appeal to older white women, a historically right-leaning group. Trump has made gains with younger Black and Hispanic men.

“Our MRP model has shown a Trump win throughout the campaign and only in the final update has it nudged Democrat,” said James Kanagasooriam, Focaldata’s chief research officer.

Even though the research points to a Democratic victory emerging from Tuesday’s election, the researchers remain wary of nailing their colors to the mast. That’s not unusual in a contest that’s become defined by the phrase “toss-up.”

“We are ‘lean Democrat’, but only by the barest of margins,” Kanagasooriam told POLITICO. “Even a polling error a third of the size seen in 2016 and 2020 would put Trump back in the White House.”

MRP — which stands for “multilevel regression and post-stratification” — is a statistical model that pollsters use to estimate state-level election results. It takes a massive nationwide sample and then uses demographic data to estimate seat or state-level outcomes.

The method — which gained the moniker of “mega poll” — burst onto the scene in the U.K. in 2017 by correctly predicting Prime Minister Theresa May would lose her majority in Parliament at a time when most polls suggested she would win big.

At this year’s general election in the U.K., MRP surveys also dominated the polling coverage, with many predicting that Keir Starmer’s Labour would win a huge victory and the Conservatives would crash to a historic defeat. Even so, the MRP projections in the U.K. were mixed, with some a long way off the final election results.

Backers of the method say it is likely to be even more effective in the U.S. where there are only 50 states. In Britain, the MRP model needs to produce forecasts for 650 constituencies represented in Westminster’s parliament, a more complicated operation.

Focaldata has been working on the U.S. polls for months. According to swing state polls, conducted online among 10,992 voters earlier in October, Trump had a big lead on handling the most important issues facing the nation. Among the four in 10 voters who chose immigration as a top issue, Trump led Harris by 74 points as the best placed candidate to handle it.

He was also ahead on how he would deal with inflation and the cost of living, jobs, tax crime and national security. The vice president led Trump on climate change, abortion, democracy and healthcare, among other issues. But Harris’ strengths lay in areas that voters less frequently named as the top issues facing the country.

Focaldata’s two data streams – state-level polling and MRP modeling – produced marginally different results. The MRP results were slightly less favorable to Harris than polling in Michigan, Pennsylvania, Wisconsin, and North Carolina, but better for her in Arizona and Nevada. The company thinks the likeliest outcomes are between the state polling and MRP modeling.

U.S. election pollsters in recent times have had a mixed record. They underestimated Trump’s support in both 2016 and 2020 and also largely missed the Democratic Party’s strength in the 2022 midterm elections.

This time, according to Focaldata’s research, the pollsters may be “herding” — clustering their results around the idea that the race is 50-50 in order to avoid looking like they’re wrong. Some analysts have suggested the polls could even be understating support for Harris.

Analysis by Focaldata showed clear grouping of poll results in the middle of the distribution chart, suggesting the contest is virtually tied. There was a strange lack of “outlier” results showing a clear win for either candidate that would normally be expected from time to time.

“We, along with others, have seen evidence of herding among pollsters at this election — likely out of fear of underestimating Trump for a third cycle in a row,” said Kanagasooriam. “The result is the current public polling is … displaying much too high levels of confidence in a close race when we are a normal-sized polling error away from a clear victory for either candidate.”

In other words, it is totally possible that the election isn’t going to be close at all, despite what the pollsters are saying.

Another British polling firm, YouGov, made the running with the use of MRP models almost a decade ago. They bucked the trend in 2016 by correctly predicting both the shock Trump win and the unexpected U.K. Brexit referendum vote to leave the European Union in the same year.

But their results were so out of keeping with expectations that YouGov didn’t trumpet them in the public, recalled Joe Twyman, who was working at YouGov at the time.

YouGov’s 2017 MRP that predicted the loss of May’s majority in the U.K. election of that year “suddenly convinced people who should have known better … that MRP was some sort of magic bullet, that it could see things that other polls couldn’t,” Twyman told POLITICO. In the elections that followed, more and more pollsters produced their own MRPs. But in fact, these models are just as fallible as more conventional polling, said Twyman.

“There are good MRPs and bad MRPs, just like there are good polls and bad polls,” he said. “Polling is about asking the right people the right questions in the right way at the right time, and if you get those elements wrong, if you’ve asked the wrong questions, then it doesn’t matter how good your MRP is, it isn’t going to fix it.”