3JUN

Reform short 894 seats on MRP

3 min read

10:25UTC

Reform UK won 14 English councils on 7 May, returning 1,448 councillors against a PollCheck/YouGov MRP projection of 2,342: a 38 percent undershoot, the worst projection failure in modern UK polling.

← Back to Reform's audit unit hits the spend wall Jump to analysis ↓

PoliticsDeveloping

Key takeaway

Five-party fragmentation has broken the uniform-swing MRP model that has projected FPTP elections for seven decades.

Reform UK returned 1,448 councillors across 14 English councils on Thursday 7 May 2026, finishing 894 seats below the PollCheck and YouGov final MRP projection of 2,342 published 24 hours before polls opened . The shortfall is a 38 percent undershoot, the largest projection error in modern UK polling.

MRP, the polling technique that combines a national vote-share estimate with constituency-level demographic data, came of age in 2017, when Chris Hanretty's model called a hung Parliament against the consensus. It has since become the trusted projection method for Westminster elections. The 7 May ballot was its first major five-party fragmented-FPTP test. First past the post (FPTP), which awards each council ward to the candidate with the most votes regardless of national share, behaves predictably under two-party competition. Under five parties polling at 26, 18, 17, 17 and 16 percent, ward-level outcomes turn on a few hundred votes either way, and uniform-swing assumptions, which redistribute votes proportionally across every ward, stop describing reality.

The operational cost landed within 48 hours. Reform had stood up transition teams for 22 councils and inherited 14. Civil-service induction packs sized for 22 went unused; coalition arithmetic prepared for 22 collapsed at 14. The party's projected 26 percent national vote share was confirmed at the count; what the model could not project was how that share converted to ward wins under fragmentation. The Greens, working from the same family of FPTP models, hit the inverse problem: 18 percent vote share projected to 696 council seats, against an actual 543 (event-06).

The consequence runs further than this council cycle. Every commercial and academic Westminster MRP currently in the field for the next general election uses the same uniform-swing under-the-hood assumption that produced the 894-seat miss. The 2026 polling industry now owes its readers a structural review of MRP performance under five-party fragmentation before the autumn party conference cycle, or its projections will be priced at a 38 percent error bar by every campaign that has read this result.

Deep Analysis

In plain English

MRP (Multilevel Regression and Post-stratification) is a statistical technique that takes a big national poll and breaks it down into projections for individual wards and councils. It does this by combining what people said they would vote with data about where they live, their age, and how their area voted before. It worked well in 2017 and 2019. But in 2026, Reform UK was surging in areas where it had never previously stood in large numbers. The model had very little local history to work with, so it extrapolated from national polls and got the seat count badly wrong. Reform won 1,448 council seats instead of the projected 2,342.

Deep Analysis

Root Causes

MRP models allocate seat projections by estimating, for each ward or constituency, the probability that a given party finishes first. That probability depends on local demographic and prior-vote covariates. When a party is simultaneously surging nationally and untested at ward level in its new target areas, the prior-vote covariate, historically the strongest predictor, carries low weight, and the demographic covariate is asked to do more work than the training data supports.

For Reform in 2026, the party had near-zero prior-vote data in many English wards (it stood in few 2023 local elections). The model therefore over-relied on national polls showing 26% support, translating that share into seat projections without adequate local calibration. The 2025 local elections were the first large-scale observation of Reform's ward-level conversion rate. Had those results been incorporated as a prior, the 2026 projection would likely have been significantly lower.

What could happen next?

Consequence
Polling firms will face pressure to rebuild MRP methodology from scratch for multi-party ward-level elections; the Electoral Commission is expected to commission a review of projection methodologies used in regulated elections.
Short term · 0.7
Risk
Media organisations that used PollCheck/YouGov MRP projections to structure election-night programming and overnight reporting may have significantly misled audiences about the scale of Reform's night; corrections and trust damage follow.
Immediate · 0.8
Precedent
The 2026 undershoot establishes that MRP requires a reform-of-method for any future UK local election cycle where a new or recently-transformed party contests more than 1,000 wards simultaneously.
Long term · 0.75

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: The 2017 UK general election saw YouGov's MRP correctly project a hung Parliament when every other model showed a Conservative majority, MRP's defining success. That success, however, was under three-party competition in most English seats. The 2026 MRP failure inverts 2017: a correct national vote-share estimate (26% for Reform) produced a catastrophically wrong seat total because FPTP's threshold effects in multi-party fields are non-linear and resistant to regression.

Counter-parallel: The 2019 UK general election MRP (YouGov, six days before polling) came within five seats of the actual Conservative total despite the Brexit Party standing in 275 constituencies. The 2019 counter-case shows MRP can handle fragmentation if the insurgent party is distributed evenly.

Reform's 2026 surge was geographically concentrated in specific council types, former Red Wall, coastal, post-industrial, which should theoretically make MRP more accurate, not less. The 2026 failure is therefore harder to explain than the 2019 success.

Consensus view: Ben Lauderdale (UCL, formerly LSE) and Pippa Norris (Harvard Kennedy School) both assess that MRP models calibrated on 2015-2024 UK general elections embed an assumption of two-dominant-party competition in each constituency.

Under five-party fragmentation, Conservative, Labour, Reform, Green, and Lib Dem all above 10% nationally, the model's constituency-level regression underestimates vote-splitting effects, producing systematically high seat totals for any party projected to break through in a large number of wards simultaneously.

Counter-view: YouGov's data science team notes that their 2017 and 2019 MRPs were accurate under comparable fragmentation conditions (UKIP at 12% in 2015, Brexit Party present in 2019). The 2026 undershoot may reflect not a modelling failure but a late-campaign differential-turnout effect: Reform voters, polled at high enthusiasm levels in late April, may have converted to votes at a lower rate than Green or Lib Dem voters, who were more likely to have postal voted early.

Key tension: Whether the 38% undershoot reflects a structural MRP limitation under five-party FPTP, or a turnout-differential effect that additional weighting could correct in future cycles.

Sources:Wikipedia (citing BBC News and Sky News results pages)·Wikipedia (citing BBC News and Sky News results pages)·Wikipedia (citing BBC News North East results)

Mentions:YouGov →Reform UK →PollCheck →Multilevel Regression and Post-stratification (MRP) →First past the post →England →UKIP →Brexit →Westminster →

First Reported In

Update #7 · Reform's 14 councils, 894 seats short

Wikipedia (citing BBC News and Sky News results pages)· 9 May 2026

Read original →

Causes and effects

Caused by

Reform projected to 2,342 council seats

PollCheck/YouGov MRP projection of 2,342 Reform seats was the baseline against which the 894-seat undershoot was measured

Occurred 5 May 2026

Read story →

Essex audit unit meets the spend wall

Reform's 1,448-seat win (ID:3146) was the mandate that put the Essex DOGE pitch before a statutory-spend ledger it cannot move

Occurred 28 May 2026

Read story →

This Event

Reform short 894 seats on MRP

MRP (multilevel regression and post-stratification) became Britain's trusted seat-projection tool after Chris Hanretty's 2017 model called the hung parliament. Its first major test under five-party FPTP fragmentation has just produced an 894-seat error, and every Westminster poll built on the same uniform-swing assumption now carries that error bar.

Led to

Reform sweeps Sunderland and Wakefield councils

The overall 894-seat FPTP undershoot explains why Sunderland and Wakefield metro boroughs, where Reform overshot, diverged from county councils that undershot

Occurred 6 May 2026

Read story →

Essex Reform elects Harris, AGM 28 May

Reform winning Essex among 14 councils (3146) enabled Harris's election as group leader on 11 May.

Occurred 10 May 2026

Read story →

Nine of 14 raise tax, eight drop climate

Reform winning 14 councils (3146) is what gave the pledge-breach pattern something to examine; 9 of 14 councils raised council tax.

Occurred 11 May 2026

Read story →

Different Perspectives

Welsh Government (Plaid Cymru under Rhun ap Iorwerth)

Plaid's Cardiff minority government relies on Green confidence-and-supply with no written agreement, the same arrangement that collapsed in Scotland in 2023. Green Westminster polling fell from 17% to 15% in two weeks as Greens took governing responsibility; whether that deflation reaches Cardiff is the near-term test for ap Iorwerth's majority.

Reform-run English county councils (Essex, Suffolk, Norfolk)

Essex named a City-trained efficiency lead over a budget where statute has already committed roughly 98% of spend; Suffolk simultaneously issued a pre-action letter against the reorganisation that will dissolve it. Reform-controlled authorities are spending public money on litigation their own sector lawyers expect to fail while their DOGE units face statutory constraints they cannot override.

Scottish Government (SNP under John Swinney)

Swinney rested the 72-55 Holyrood mandate on the combined SNP-Green bloc rather than his own party's 58 seats, seven short of the trigger he named; he has publicly conceded he has no plan if Westminster holds its veto. The constitutional argument is made; the enforcement route does not exist.

UK Government (MHCLG and Downing Street)

MHCLG has until 12 June to respond to Suffolk's pre-action letter and faces three further counties at the same stage; Downing Street rejected Holyrood's Section 30 demand as a spokesperson lobby line rather than a written statement, declining to open formal inter-governmental correspondence. Both decisions compress Reform's two main legal challenges into the same two-week window.

UK Government (Labour)

Westminster framed the youth justice transfer as a culmination of prior work rather than a precedent, refused a Section 30 order before the request arrived, and omitted both the Representation of the People Bill and any Wales Bill from the 13 May King's Speech. Starmer is described as open to a devolved-leaders summit in June.

Reform UK

Essex, Norfolk and Suffolk filed pre-action protocol letters framing their LGR challenge as mandate-consistent, while 22 Reform councillors departed in 14 days at an annualised rate nearly three times the 10 percent projection. Richard Tice defended the Harborne 5 million pound gift as unconditional, with no acknowledgement of Farage's two contradictory accounts on record.