John Curtice: It pays to treat the exit polls with caution

As soon as the polls close at 10pm tonight the BBC, ITV News and Sky will tell us how many seats each party has won. Or at least they hope to.

How will they know? By conducting a jointly commissioned exit poll in which voters outside 130 polling stations across the country are asked by interviewers from GfKNOP and Ipsos-MORI to indicate how they have just voted by filling in a mock ballot paper. The results will then be analysed by a team of voting experts and statisticians in order to produce that 10pm final forecast.

The task facing the pollsters is a formidable one. There are few countries in which it is more difficult to conduct an exit poll successfully than it is in Britain.

First, unlike most other countries, elections are not counted separately in each polling station. We thus do not know how voters at each polling station voted last time. This means it is virtually impossible to establish whether any particular sample of polling stations is representative of the country as a whole.

To overcome this difficulty today's exit poll will mostly be conducted outside the same polling stations as a similar poll in 2005. The team of analysts will be able to compare the voting pattern uncovered this time with the results of the previous exit poll to establish how much each party's vote appears to be up or down.

The second problem is Britain's electoral system. In contrast to more or less any system of proportional representation there is no straightforward mathematical formula that determines how votes are translated into seats. The outcome depends not just on how many votes a party wins but also on where it does so.

So the exit poll analysts have to try and establish whether there are any crucial variations across the country in the pattern of voting – such as whether the Conservatives are advancing more in Labour-held marginal seats than elsewhere.

There is yet one more hurdle. Elections are held on a Thursday rather than at the weekend. As a result most people vote on the way home from work. So most of the data collected by the exit poll only becomes available very late in the day – leaving very little time for the analysts to work out what it all means. Given these problems it might be thought that the exercise is not much more than foolhardy and little notice should be taken of its forecast. That, however, would be too dismissive.

At the last election a similar exercise commissioned jointly by the BBC and ITV News forecast a Labour majority of 66 – exactly the outcome that eventually transpired. True, that same poll slightly underestimated the number of seats won by the Liberal Democrats – it forecast they would win 53 seats when in the event they captured 62 – but the 10pm forecast did give viewers a very good idea of the overall shape of the next House of Commons.

Of course there was a substantial element of luck in producing such an accurate forecast. There is certainly no guarantee tonight's forecast will be just as accurate. Exit polls have not always been so successful as last time. In 1992, for example, a BBC exit poll forecast the Conservatives would only just have more seats than Labour in a hung parliament. In the event, the Conservatives secured an overall majority of 21.

However the purpose of an exit poll is not simply to tell everyone what the result will be before the votes have been counted. It also provides a benchmark against which the significance of the early results can be identified and assessed.

If those early results are in line with the expectations of the exit poll, then what they portend is made clear. And if they are at variance with the exit poll? Well, it will be an excuse to stay up a little longer!

John Curtice is professor of politics at the University of Strathclyde. He is analysing the election opinion polls for The Independent