Exit polling is tricky at the best of times. Attempting it in Britain is a nightmare. The joint BBC/ITV exit poll correctly predicted the 66 seat majority in 2005. Don’t expect it to be as accurate tonight.
The experts have to work around duff information There’s no data on voting at individual polling stations. The census is nine years out of date. Local election ward returns are a flawed guide to voting patterns for a general election.
One in six voters refuse to respond to an exit poll It’s a mind your own business answer. And no one knows if these people disproportionately vote for one party.
Up to one in five votes will be cast by post These voters cunningly by-pass the exit pollsters. Around 12 per cent voted this way last time, and the proportion is rising. Most of them sent in their ballots around the peak of Cleggmania.
The traditional swingometer may be obsolete Even if they get the vote share correct, there’s a good chance the seat prediction will be flawed. See more on this here and here.
The exit polling sample barely covers Lib-Lab marginals Because there is no data from on individual polling stations, the wonks calculate the change from the 2001 and 2005 exit polls. It covers around 120 polling stations. But there’s only data on three Lib-Lab marginals. That’s why the Lib Dem vote share prediction was too low in 2005. The problem will be even greater this year.
Boundary changes muck up what historic data is available Since 2001, many boundaries have been revised. Some polling stations in the sample have been split in half, so how do you calculate the change in voting? On top of that, people move, meaning many polling stations serving very different communities.
It’s a mad rush to manage the data and make a prediction It’s a Thursday. Most people vote after they’ve been to work. That gives the boffins very little time to make the calculations. They’re working to a 10pm deadline and data is still streaming through until 7pm.
There are some seriously big brains deployed to build models that iron out these wrinkles. (One of their more basic forecasting equations is below.) You can read about their 2005 experience in this excellent paper, or Chris Giles’ astute take on the work of the exit poll boffins and the future of the swingometer.
John Curtice and Colin David Firth, who led the team in 2005, have said that they “fully acknowledge that getting the headline forecast exactly right owed much to luck”. They’ll need bags of it this evening. 
Jim Pickard
Kiran Stacey