Exit polls: What are they, and how do they work? All you ever wanted to know just in time for the election result

David Cowling, who has worked on exit polls for ITV and the BBC for decades, explains what lies behind tonight's big moment at 10pm

National exit polls have been used to predict the results of British general elections for forty years. In the October 1974 election, the BBC commissioned Louis Harris to conduct their exit poll and ITN used ORC. In 1979, ITN persisted with ORC but the BBC abandoned an exit poll in favour of a “poll of polls” which they used to predict the result. In both 1983 and 1987, BBC seat prediction was based on a national opinion poll conducted by Gallup.

Read the final exit poll here

From 1992 onwards we commissioned exit polls from NOP. ITN used Harris Research 1983-92 and MORI from 1997 onwards. In 2005, the BBC and ITN agreed to a joint exit poll, conducted by MORI and NOP. In 2010, Sky News joined the other two broadcasters in sponsoring the exit poll and this triumvirate continued in 2015; and they have jointly commissioned the 2017 general election exit poll, which will be conducted by MORI and NOP.

In some elections before 1997 there were two exit polls: one for seat prediction, which simply asked voters who had just exited polling stations how they had voted; and the other, the analysis exit poll, which asked voters a short series of questions to discover why the nation had voted the way it did. Both types were based on a small selection of polling stations throughout Britain (no sampling has ever been conducted in Northern Ireland) but the samples were distinctly different.

The seat prediction exit poll only sampled polling stations in marginal seats because its sole purpose was to predict the actual outcome of the election that would be decided in the relatively small number of seats with a history of changing hands from one party to another. The analysis poll was conducted in a set of polling districts designed to provide a representative national sample because its purpose was to deliver a snapshot of the views of all GB voters – those voting in every type of seat, not just in marginal ones.

Analysis polls were abandoned after the 1992 general election and since then there have only been seat prediction exit polls. Often colleagues ask for the demographic details of voters in the exit poll – what was the gender split between Conservative and Labour voters, how did the different age groups vote, etc.? The simple fact is that none of this data is available from the seat prediction poll: those who participate in these exit polls are simply asked to replicate the vote they have just cast – they do not fill in a census form at the same time, listing gender, age, ethnicity, or any other personal detail.

Exit polls do not engage every voter at the selected polling stations. The pollsters approach one voter out of a given number – the ‘given number’ (i.e. one out of every six/eight/ten etc.) based on the estimated national turnout agreed in advance between the polling companies and the broadcasters.

And because the exit poll is conducted in a small fraction of the 39,000 polling stations in Britain it is inevitably subject to a margin of error.

Exit poll results 1974-2015

The tables below set out the record of exit polls in general elections since October 1974. For nearly all this historic data I am totally indebted to my friend, Chris Long, the evil genius behind three decades of ITN election programmes (and much else besides).

We can see the extraordinary BBC prediction in the October 1974 election (a predicted 132 Labour majority versus an actual three seat majority) and then compare it to the remarkably accurate BBC predictions in 1983 and 1997. The 1992 car crash is also there in all its gory detail. I was with ITN at that election but suffered the same nightmares as our BBC colleagues. But interestingly, as Nick Moon observes, the 1992 predicted majorities for both the BBC and ITN were more accurate than either broadcaster’s 1987 prediction. The brutal difference was that in 1987 the less accurate predictions still identified the winning party: in 1992, they did not. We can also see the 2001 result where both the BBC and ITN produced very respectable predictions.

The 2005 result illustrates how random chance plays a part in polling as in so much else in life. The poll was out by eleven seats for the Conservatives and out by nine seats for the Lib Dems. But because it correctly predicted the exact number of Labour seats, the headline majority broadcast at 10pm was spot on (the majority was the outcome of amalgamating every predicted non-Labour seat and subtracting the resulting total from the predicted total of Labour seats).

BBC/ITV News exit poll 2005 general election

The 2010 exit poll result was as near perfection as one is likely to reach this side of Heaven.

BBC/ITV News/Sky News exit poll 2010 general election

BBC/ITV News/Sky News exit poll 2015 general election