The U.K.’s exit poll conducted for the major networks is unlike anything most Americans are used to. In the United States, exit polls are used to explain why each candidate or party won, but in the U.K., the exit poll’s main purpose is to make a prediction about how many seats each party will win.
Why the emphasis on seats?
First, election night is really, really long in Britain. Half the seats may not be called until after 4 a.m. London time. It’s nice to get a first read of the results before people go to sleep. We have a lot more data going into an election night in the U.S. from pre-election polling (which the U.K. has historically lacked on the constituency level), and most seats and states in America aren’t close.
Second, the emphasis shifted toward dedicating all resources to getting the right result after the 1992 election. That year, there were two exit polls: one for seats and one for analysis of why people voted the way they did. That was also the year that BBC and ITN exit polls missed the election result by more than 60 seats. They called for a hung parliament, yet the Conservatives won a clear majority.
Indeed, election night forecasts from the BBC and ITN have historically been subject to wide margins of error. According to data collected by Clive Payne at Oxford University, the margin between Labour and the Conservatives projected at 10 p.m. (when the polls closed) was off by an average of 39 seats from 1974 to 1997.
Since 2001, the exit poll results have been fairly accurate, thanks to a new technique instituted by a team of academics. (You can read more about that technique, but suffice it to say that it has a lot of moving parts, including data down to the polling station level that is then modeled across the country.) Instead of an average error of 39 seats, it’s been only seven seats.
This year, there are a lot of potential changes in voting patterns that could throw off the exit poll. There are arguably five major parties now (instead of three), and Scotland is being contested for the first time in ages. To try to adjust, 10 more constituencies and polling stations will be polled. That may help, though we don’t have past election data for these polling stations to extrapolate the swing from. (Election day data isn’t released at the polling station level. In most cases, previous exit poll data is available and used to compare with this year’s exit polls.)
John Curtice, president of the British Polling Council and a FiveThirtyEight contributor, helped design the exit poll. He has said to “prepare for disappointment.” In other words, we might be heading back to the 1974 to 1997 era of major exit poll errors.