Polls predicting the next British election are not to be trusted

“Sky News can now reveal the latest polling projection from YouGov,” announced the channel’s presenter. “It puts Reform UK on course to be the next government and Nigel Farage, of course, to be the next prime minister.” The segment, which aired on September 26th, was reporting on a polling projection known as MRP (multilevel regression and post-stratification), from YouGov. It was framed as if it were an exit poll—yet an election is probably years away. Any poll that claims to show results constituency by constituency should be treated with scepticism.

MRPs are a sophisticated statistical technique which use public-opinion surveys to estimate voting intentions around the country. They work by finding the individual characteristics (such as age and past voting behaviour) and local factors which predict how people are likely to vote. Pollsters project these characteristics in each constituency, weighted according to their population, to judge how they might swing if an election were held today.

The technique has been effective in the past. In 2017 YouGov went out on a limb to predict (correctly) a hung parliament. Since then MRP polling has exploded: 46 such forecasts were published between the 2019 and 2024 general elections, showing a wide range of predictions (of varying accuracy). Journalists lap them up. This creates opportunities for spin doctors, who judge that the cost of an MRP is good value for guaranteed media coverage. A recent MRP projection which suggested the Conservatives would win just 14 seats was leaked, in an apparent attempt to destabilise Kemi Badenoch, the party’s leader.

The gluttonous consumption of MRPs might not be wise. “[MRP] is not capturing a lot of the things we know are going to change between now and the next election,” says Jane Green, a professor of politics at the University of Oxford. One is the fortunes of the parties over the next few years. At the local level tactical voting could create even more uncertainty.

As a snapshot of current opinion, MRPs are interesting. Indeed, The Economist’s statistical modelling of the next election incorporates data from such forecasts. But unlike an MRP, the focus of our modelling is uncertainty, which flows from Britain’s voting system. This far out from an election our fuzzy range of plausible outcomes is less sexy but more realistic.■