United Kingdom · Pollster ratings
Every rated pollster, scored for reliability.
These scores were computed on 24 March 2026 for the previous UK average (wra-v1.0). The current average, in use since 22 October 2026, does not use them. The scores are bounded between 0.3 and 0.7 and run from 0.42 to 0.57. Under wra-v1.0 the most trusted pollster carried 1.4 times the weight of the least. Recency and sample size mattered more than the score. The table lists the 28 pollsters that have a score, with their 2,115 polls. The database holds 2,136 polls from 31 pollsters; the other 3 pollsters, with 21 polls, have no score yet.
Scores computed 24 Mar 202628 rated pollstersHow the score is built
| Pollster | Score | Polls | Avg sample | British Polling Council | Latest poll |
|---|---|---|---|---|---|
| Techne | 0.57 | 167 | 1,658 | Yes | 15 May |
| More in Common | 0.57 | 151 | 2,971 | Yes | 9 Nov |
| JL Partners | 0.57 | 29 | 2,861 | Yes | 11 Aug |
| Survation | 0.57 | 118 | 3,260 | Yes | 12 Nov |
| BMG Research | 0.56 | 57 | 1,544 | Yes | 29 Oct |
| YouGov | 0.56 | 340 | 2,754 | Yes | 9 Nov |
| Deltapoll | 0.55 | 136 | 1,606 | Yes | 9 Nov |
| Ipsos | 0.55 | 56 | 1,434 | Yes | 22 Oct |
| Opinium | 0.55 | 203 | 1,907 | Yes | 6 Nov |
| Focaldata | 0.55 | 27 | 7,311 | Yes | 10 Mar |
| Savanta | 0.54 | 181 | 2,264 | Yes | 3 Jul 2024 |
| Verian | 0.52 | 44 | 1,362 | Yes | 23 Mar |
| Redfield & Wilton | 0.52 | 237 | 2,266 | Yes | 2 Jul 2024 |
| Find Out Now | 0.52 | 119 | 4,039 | Yes | 12 Nov |
| We Think | 0.52 | 51 | 1,595 | Yes | 8 Aug 2024 |
| Whitestone Insight | 0.51 | 9 | 3,295 | Yes | 23 Jan 2025 |
| Norstat | 0.51 | 5 | 1,848 | Yes | 3 Jul 2024 |
| Lord Ashcroft | 0.50 | 33 | 5,069 | No | 2 Nov |
| Freshwater Strategy | 0.49 | 23 | 1,262 | Yes | 1 Nov |
| Good Growth Foundation | 0.49 | 3 | 2,000 | Yes | 7 Jun |
| Stack Data Strategy | 0.47 | 1 | 9,301 | Yes | 3 Aug 2025 |
| Omnisis | 0.47 | 61 | 1,235 | Yes | 18 Aug 2023 |
| PeoplePolling | 0.46 | 44 | 1,346 | No | 2 Jul 2024 |
| Number Cruncher Politics | 0.45 | 12 | 1,427 | No | 3 Jul 2024 |
| Kantar Public | 0.44 | 0 | 1,112 | No | — |
| Panelbase | 0.43 | 4 | 3,055 | No | 19 Nov 2021 |
| Labour Together | 0.42 | 2 | 5,337 | No | 29 Apr 2024 |
| Stonehaven | 0.42 | 2 | 2,060 | No | 9 Dec 2024 |
How the score is built
Four inputs, combined with fixed weights: accuracy against election results, with the penalty reduced when the whole industry missed together (40%); whether the pollster's movements track the underlying trend or are idiosyncratic noise (20%); methodological trust, meaning transparency, track-record depth, sample quality and recency of activity (25%); and the size and stability of the pollster's house effect (15%). Each input is shrunk toward an informed prior before they are combined. The full method, with the formulas, is on the methodology page.
What the current average uses instead
The current average does not read these scores. A pollster's say in setting the average's level depends on an anchor weight between 0 and 1, which builds up from the pollster's polls of the past 365 days and reaches 1 at about one poll a month. Only pollsters that meet our disclosure standard (British Polling Council membership, or published tables within two working days) carry anchor weight. Every eligible poll still moves the average, and pollsters with more and larger polls move it more. Each pollster's house effect, its lean against the average pollster, is estimated and removed.