The current mistrust for polls started in Britain when pollsters got the 2015 general election and the 2016 Brexit referendum wrong. Now, it might appear that, despite intensive soul-searching, British pollsters are still unreliable: their numbers are all over the place, with Prime Minister Theresa May's Conservative Party commanding a lead of anywhere between 3 and 12 percentage points.

It's possible, however, that at least one polling organisation has figured out how to capture voting intentions in a way that best fits Britain's election system. And this organisation's model produces the least favourable results for May.

In YouGov's model, an unusually large number of respondents are polled: at least 5000 are surveyed every day to produce daily results. On Thursday, May's Conservative Party stood to receive 42 per cent of the vote and 317 of the 650 parliamentary seats, and Jeremy Corbyn's Labour Party was at 38 per cent and 253 seats. But the YouGov model is more than a poll. It's a hybrid of a traditional opinion survey and an exercise in big data analysis.

"We use the individual responses in each constituency to build a model of how voting is related to individual characteristics and the types of constituencies people live in," Benjamin Lauderdale, the London School of Economics professor who designed the YouGov model, said. "We then use that model to calculate an estimate for each constituency based on the kinds of people who live there and how they have voted in the past."

Poll accuracy depends a lot on the methods used by sociologists – for instance, on the relevance of their samples. But there's another reason why French polls were strikingly accurate ahead of the recent presidential election, while similar precision has proved elusive in Britian and the United States. For the purposes of the presidential vote, France is a single national constituency. Britain and the US don't work that way: the winning party in Britain must win the most constituencies, not the biggest share of the national vote, and a US presidential candidate must win the most electoral votes rather than garner a national majority. That makes things difficult for polling organisations: doing local polls everywhere is expensive and a lack of resources often makes such polling unreliable.

YouGov's approach helps overcome that problem. While the number of respondents in each constituency is rather small – fewer than a 100 over the duration of the project – the use of geographical, demographic and voting data makes it possible to estimate the outcome in each constituency and each party's number of seats.

Lauderdale started working on the method after the 2015 election, trying to nowcast results from existing polls and local trend data. The results were off. Further efforts proved more accurate. YouGov tested an iteration of Lauderdale's model on the Brexit referendum, and it predicted a victory for the "Leave" side, though by a thinner margin than it actually achieved.

The current snap election, called by May in the hope of destroying the opposition and taking a commanding majority, is the first one for which YouGov is officially using the model. Traditional polls still show May might get what she was after, though her lead has recently shrunk. Financial Times' polling average – which contains results from the YouGov model – on Thursday gave the Conservatives 44 per cent of the vote and Labour 35 per cent. Pollsters talk about the predicted youth turnout determining the difference: according to their conventional wisdom, if one believes that more young people than usual vote on June 8, he or she will predict a better Labour performance. Lauderdale also said that different assumptions about who will turn out to vote drive the difference in poll results.

He said, however, that the model he designed did not predict a higher than usual youth turnout. "The estimates assume a similar turnout by age distribution to 2010 and 2015 – which were very similar." "We use the 2010 and 2015 British Election Study to determine how strong this relationship is."

Obviously, Lauderdale and YouGov are not necessarily right about Britons' voter preferences: they could be mistaken about their voter typology and other assumptions. But election systems in Britain – and the US – demand that researchers make an effort to go beyond the national vote, to the local level. YouGov's approach to that problem is clever and innovative; if it works on June 8, pollsters dealing with inconvenient election systems could take a giant step toward greater accuracy.

In any case, it's worth keeping a closer eye on the YouGov model's results than on traditional polls, if only because it worked well for Brexit.  

Robert Barnes, the US lawyer who made hundreds of thousands of dollars betting on Brexit and Donald Trump's victory, has told me his money is on Labour outperforming projections because it's harvesting the anti-establishment vote.

That's exactly what the YouGov model is showing, too. The likelihood of a hung parliament, in which no party has a majority, is not negligible. May is in danger of snatching defeat from the jaws of victory. Even if she does get a majority, it appears highly unlikely that it will be strong enough to give her a free hand, both in Brexit negotiations and in domestic politics.

Bloomberg