How The Secret General Election 2017 Exit Poll Is Produced

(Bloomberg) -- At a secret location in London on Thursday, eight men will sit in a sealed room, hunched over laptops. They are Britain's leading experts on the science of voting, and their mission is to produce the first authoritative projection of the day's election result.

The exit poll, published as Big Ben, the bell in Parliament's clock tower, strikes 10 p.m., has delivered shocks in the past two elections, heralding the first postwar coalition government in 2010 and an outright Conservative win in 2015 when the money was on another coalition. It is produced in conditions of high security, with those involved sworn to secrecy about the exact methods used.

Read More: Everything that pollsters could get wrong in this election

“Of course it's exciting,” said Rob Ford, professor of politics at Manchester University, who has been involved in producing the poll since 2005 and co-edited “Sex, Lies and the Ballot Box.” “We're the first people in the country finding out what's going to happen. For the best part of half a day, we're the only people who know what's coming.”

The results are more reliable than normal opinion polls because they're based on interviews with people who have actually voted, removing the problem of people who say they'll vote but don't, and those who think they'll vote one way but change their mind in the booth.

Last time, a total of 20,000 voters were interviewed. By using historical data from previous elections, it aims to measure shifts in voting patterns across the country. Those are then combined to predict how different seats will vote.

The team is led by John Curtice, a 63-year-old professor of politics from Strathclyde University in Glasgow who manages to be both balding and wild-haired at the same time and who has become a fixture on television screens as the BBC's polling guru.

Psephology -- which comes from the Greek for pebbles, used in antiquity to vote -- involves sciences, number crunching and ultimately a judgement call on how to interpret it all. For Curtice it's also an art form and a lifelong passion.

In a 2005 interview with the Guardian, he described how at age 11 he found himself mesmerized by the 1964 election that saw Harold Wilson lead the Labour party to victory: "I was up half the night listening to the results.”

Part of the art of the poll is identifying how different parts of the electorate are shifting. Rural voters might go one way, while urban voters go another. In this election, Labour candidates report high support among the young and those in city centers, and less in suburban areas.

1992 Fiasco

After a disaster in 1992, when the poll was 62 seats out predicting that John Major's Conservatives would be short of a majority -- and a lesser miss in 1997, when it was out on the scale of Labour's landslide by 29 -- the method has been changed. Between 2001 and 2010, it got the winning party's majority to within six votes. 

Secrecy is paramount to prevent the result leaking out -- even though it is a crime in Britain to publish information on results while people are still voting. “There are very, very strict security procedures,” said Ford. Mobile phones are confiscated from everyone in the room.

Three broadcasters -- the BBC, Sky and ITV -- jointly fund the poll, but very few of their staff are allowed to know what it says, and even those in on the secret are told only at the last possible moment. Because they need time to prepare visuals, the first people told are graphics staff, before even senior correspondents.

‘Why the Hell?'

In 2015, Nick Robinson was the BBC's political editor. He was given the numbers a few minutes before he was due to go live. It predicted the Conservatives would gain seats. In his memoir of the election, he described his reaction: “I want to shout, ‘Really? Labour go backwards and the Lib Dems collapse? Why the hell haven't the polls shown that happening?”'

But for the exit-poll team, 10 p.m. is barely halfway through what Curtice describes as “the long march.” Their next data points are the actual results. As these come in, starting at about 11 p.m., they offer solid information on how the vote is shifting, and the model increasingly relies on this.

It doesn't matter that the early declarations in U.K. elections tend to be from Labour-supporting seats in places like Sunderland in northeast England: Curtice and his team are looking for the change in votes, not just the headline result.

“Once 50 results have come in, that's what's driving things,” said Ford. “You'll see that the projections do change as the results come in. That's because the model is learning.”

To contact the author of this story: Robert Hutton in London at rhutton1@bloomberg.net.

To contact the editor responsible for this story: Flavia Krause-Jackson at fjackson@bloomberg.net, Eddie Buckle