ON ELECTION DAY in 1970 the BBC tried something “entirely new”: an exit poll at Gravesend in Kent, which was “shown by a computer to be the most ordinary constituency in England”. It correctly anticipated a surprise victory for the Conservative Party. Nearly half a century later, a nationwide exit poll has become an essential element of British election-night coverage. No polls may be published while voting is taking place, but once the booths close on the stroke of 10pm on December 12th, the result of the exit poll—a projected seat total for each party—will be broadcast on television. How exactly does it work?
It is perhaps the world’s most expensive one-question social survey. Britain’s three main broadcasters—BBC, ITV and Sky News—jointly commission the exit poll from Ipsos Mori, a pollster, and a team of academics, at a cost of some £300,000 ($395,000). The work takes place in two stages. First, pairs of fieldworkers are dispatched to 144 of Britain’s 40,000 polling stations between the hours of 7am, when voting begins, until it ends. At each location they ask approximately every eighth person leaving to fill in a dummy polling card containing a single question, “How did you vote?”, and deposit it anonymously into a cardboard box.
With no other information gathered about the 30,000 respondents to help make the projection (as happens in American exit polls, for instance), the accuracy of the exit poll rests upon surveying the same locations in each general election. These were chosen in 2001 to be broadly representative of Britain, but also to be telling enough in the event of a tight national race between the Conservatives and the Labour Party. The academics are not concerned with the level of support for parties in the selected places, but the change compared with the previous exit poll (there are no actual results for individual polling stations).
The second stage involves the academics working throughout the day to create models that explain why support for the different parties has changed. These are then applied to each of the 632 constituencies in Britain—the 18 seats in Northern Ireland are treated differently—to arrive at seat-level estimates of vote share and the probability that any given party will be victorious. Adding up the probabilities for each party in each seat produces a forecast for a party’s expected number of seats nationwide.
Like any statistical exercise, the exit poll comes with an amount of uncertainty. David Firth, of the University of Warwick, who helped pioneer the current technique, reckons that any forecast for the largest party’s seat total that is within ten seats of its final tally should be considered as “exceptionally accurate”. Remarkably, in all but one of the past five exit polls, in 2015, that error has been five seats or fewer (see chart).
This year’s exit poll might be a more difficult undertaking, for three reasons. First, the anti-EU Brexit Party is standing for the first time, but only in seats that are not already held by the Tories. Although support for the party has fallen in recent weeks to just 2% nationally, according to polls, modelling how they might affect the ability of the Tories to win Labour-held marginal seats could cause a headache for the academics. Second, tactical voting, particularly among those wanting to remain in the EU, could be more widespread in this election, making the modelling trickier. The greatest source of uncertainty comes from the postman: Britain’s first December election since 1923—elections are normally held in the spring—could lead to a higher share of postal votes. If the composition of postal voters is different from 2017, when 21.6% of voters submitted their ballots by mail, that could lead the exit poll astray.
The exit poll’s uncertainty is communicated in the message that is broadcast at 10pm. A prospective government needs 326 seats to win a majority, but with an approximate margin of error of 15 seats either side, a central estimate of anything under 340 seats for the winning party means that the exit poll cannot rule out the possibility of a hung parliament. In the past three elections, even with remarkably accurate exit polls, the announcement of “Conservatives the largest party” has preceded very different outcomes: in 2010 a hung parliament and a coalition government; in 2015 an outright, albeit, slim majority; and in 2017 another hung parliament and a minority government.
The latest polls and seat predictions suggest that the most likely outcome in 2019 is that the Tories win an outright majority. But with the polls tightening in the last few days of campaigning, it may still be that when Britain holds its breath at 10pm, its voters are once again presented with the possibility of a hung parliament.