Gold standard What is an MRP poll and why is it so reliable?

The Yougov poll published today is the most significant warning for the Tories from any poll since 2019. It was carried out using the Multilevel Regression with Poststratification (MRP) method, recognised as the gold standard of election research, which combines census data with large-scale public opinion polling. Between Dec 22 and Jan 3, Yougov conducted a nationally representative poll of 13,910 adults in Britain on a number of topics including how they would vote if an election were to be held tomorrow. Other recent polls typically sample around 2,000 people – too few to break the overall numbers down to individual constituencies, which is how general elections are decided. Such a large sample size enables Yougov to provide a granular break down by constituency. Yougov’s MRP model has been credited with correctly predicting a closer race in the 2017 general election than traditional polling, and has been seen as among the most reliable predictors of UK elections. It recently forecast the Australian and Spanish elections with great precision. In 2019, a Yougov MRP poll two weeks before the election predicted that Boris Johnson would win 359 seats – just six short of the eventual outcome. With UK census data, Yougov uses a wide variety of variables that capture different voting patterns. These include demographic characteristics of individuals and constituencies, past vote choices of individuals and shares of constituencies, information about the candidates standing for different parties, and a range of other features of constituencies. Crucially, it also accounts for how the large number of voters who currently say they have not decided who to vote for might vote, by assuming they will vote similarly to the way people politically and demographically similar to them are planning to. Many other polls simply exclude these voters from their headline figures, giving a misleading picture of what the final result is likely to be.