by Stephen Fisher and Jake Dibden, 26th June 2024
Seat projections for the Conservatives at next week’s general election range from really bad to totally dire. Given recent polls, traditional uniform-change projections suggest the Conservatives will win just around 190 seats. On average the Multilevel Regression and Poststratification (MRP) models suggest the Tories will win less than 100 seats.
The main difference between them is that the MRP models estimate that the Conservative vote has dropped more where the party was stronger in 2019. That is to say, the drop is broadly proportional to prior strength rather than uniform (the same) across all constituencies.
Some MRP models and other forecasters have come unstuck in the past by predicting proportional change when uniform change predictions have been a better guide to seat tallies for the big parties in the post-war period (see Appendices of the Nuffield Election Studies). The efficacy of uniform change projections was so well established that they became the basis of the swingometer for election night programmes.
Since past vote choice is such a strong predictor of future vote choice, MRP models in effect have a 2019-2024 vote-transition matrix model at the heart of them. That in turn means MRP models tend to project proportional drops for parties in decline. The MRP modellers need a lot of data across constituencies and careful modelling to identify any counter-balancing pattern towards uniform change at the constituency level.
Are we likely to see predictions of proportional change come unstuck again this year?
Continue reading Breaking the swingometer: historical precedents for proportional change