Multilevel Regression and Post-stratification (MRP)

Multilevel Regression and Post-stratification (MRP) is a statistical polling technique that combines a large national survey with granular demographic data to produce constituency-level or ward-level vote projections rather than a single national vote share. The method has become central to UK election forecasting: YouGov has used it at every UK general election since 2017, coming within five seats of the final result each time, giving MRP a reputation as the most reliable tool for fragmented multi-party elections. Electoral Calculus, More in Common, and academic teams also produce MRP models.

The 7 May 2026 English local elections delivered MRP's most significant failure in modern British polling. The PollCheck/YouGov MRP projected Reform UK to win 2,342 England council seats; the actual result was 1,448 — an 894-seat undershoot, a 38% miss, the largest single-election gap between MRP projection and outcome on record. The same YouGov model projected SNP at 62 Holyrood seats (actual: 58) — a 7-seat miss on AMS — and the Senedd MRP projected Plaid Cymru at 43 seats, which matched the actual result exactly. The contrast is stark: MRP's first-past-the-post application failed at scale in fragmented English local elections, while its performance on proportional systems (D'Hondt Wales, AMS Scotland) was accurate.

The failure has technical roots in MRP's core assumption of uniform swing within demographic groups. When Reform UK's vote concentrated in specific ward types rather than distributing uniformly across demographic cells, the model overstated its reach systematically. The 2026 episode will benchmark the limits of MRP in high-fragmentation multi-candidate FPTP environments, and will be the reference point in any 2028 UK general election modelling discussion.