Picking apart the MRPs

Different results hide some clear underlying patterns

One of the great innovations in election prediction in the last decade has been the emergence of MRP models, pioneered by YouGov. Their 2017 model correctly identified the UK was heading towards a hung parliament while conventional polls and models were predicting a clear Conservative majority.

MRPs take large polling samples and identify the demographic and political characteristics affecting people’s voting intentions and then project them onto constituencies based on their demographic and political characteristics. While the exact variables vary from model to model, in an era of more complex political behaviour, they offer a better way of predicting individual seat races than traditional swing models that assume uniform behaviour in each seat.

In the last week, four polling companies have released the results of their early campaign MRP models. While offering quite a varied set of results, they really just reflect the range of recent polls and their implications. More in Common’s echoes the narrower end of the pack, a reasonable best case for the Conservatives and a Labour majority of a mere 114; Find Out Now (with Electoral Calculus) and Survation’s 300+ Labour majorities are more what you would expect if the polls with the largest Labour leads came true; while YouGov’s model and its headline of a nearly 200-seat Labour majority is slap-bang in the middle and how current polling averages would likely translate into seats. Each of the pollsters offers oodles of analysis of their own results in isolation, so what we are going to do is take a look at them collectively – what do they all tell us about the election and where do they agree and disagree.

It should be noted that all four MRPs were done before Nigel Farage became leader of Reform and before Sunak’s D-Day blunder, so should be seen in that light.