Survation’s track record: how accurate are our polls?
Survation has published a final poll before every general election, referendum and Scottish Parliament election since 2014. This page sets each one beside the result, including the ones we got wrong, with links to our tables and to independent records so that anyone can check.
- 2017 general election: the most accurate final poll. We had the Conservatives one point ahead of Labour; the result was 2.5. Ours was the only final poll from a British Polling Council member to point to a hung parliament.
- 2014 Scottish independence referendum: the most accurate final poll, at Yes 47%, No 53% against a result of 45% and 55%.
- 2026 and 2016 Scottish Parliament elections: in 2026 we had the lowest error of any company on the constituency vote and on MRP seat projections. In 2016 our final telephone poll had the Conservatives overtaking Labour on the list vote.
- 2019 general election: we put the Conservative lead at 11 points. It was 11.7.
- Independent rating: Election Data Vault rates Survation joint second of 20 UK pollsters (score 88, grade A-), on a measure built mainly from final-poll accuracy at the last three general elections.
- Constituency polls: we are one of very few companies that polls single seats by telephone. A selection of results is below.
Survation is a member of the British Polling Council and a Market Research Society Company Partner. Every published poll has full tables in our archive.
General elections
Our final polls at the last four general elections, against the result in Great Britain. Figures are those recorded by the British Polling Council, which publishes every member’s final poll beside the result after each election.
| Election | Survation final poll | Result (GB) | Lead: poll v result | Average error per party | Place among BPC final polls |
|---|---|---|---|---|---|
| 2024 | Con 20, Lab 38, LD 12, Reform 17, Green 7 (telephone, 1,679, 1 to 3 July) | Con 24.4, Lab 34.7, LD 12.5, Reform 14.7, Green 6.9 | Lab +18 v Lab +10.3 | 1.9 points | 8th of 19 |
| 2019 | Con 45, Lab 34, LD 9, Brexit 3, Green 3 (telephone, 2,395, 10 to 11 December) | Con 44.7, Lab 33.0, LD 11.8, Brexit 2.1, Green 2.8 | Con +11 v Con +11.7 | 0.9 points | 5th of 11 (joint 3rd on the lead) |
| 2017 | Con 41, Lab 40, LD 8, UKIP 2, Green 2 (telephone, 2,798, 6 to 7 June) | Con 43.5, Lab 41.0, LD 7.6, UKIP 1.9, Green 1.6 | Con +1 v Con +2.5 | 0.9 points | 1st of 9 |
| 2015 (published) | Con 31, Lab 31, UKIP 16, LD 10, Green 5 (online) | Con 37.8, Lab 31.2, UKIP 12.9, LD 8.1, Green 3.8 | Tie v Con +6.6 | 2.6 points | With the rest of the industry |
| 2015 (not published) | Con 37.3, Lab 30.6, UKIP 10.6, LD 10.0, Green 5.2 (telephone, 6 May) | As above | Con +6.7 v Con +6.6 | 1.3 points | Not counted: it was not published before the vote |
Sources: British Polling Council records for 2024, 2019 and 2017.
2017: the only final poll pointing to a hung parliament
The average final poll in 2017 gave the Conservatives a lead of about eight points and a comfortable majority. Final Conservative leads from other British Polling Council members ranged from 5 to 13 points. Ours was 1. The result was a lead of 2.5 points and a hung parliament.
No other member’s final opinion poll pointed to that outcome. (YouGov’s separate MRP model also indicated a hung parliament; its final poll did not.) Our own MRP that year, built with Dr Chris Hanretty for private clients, also showed no overall majority. How we got it right.
The editors of Political Communication in Britain (Palgrave Macmillan, 2019), the academic study of the 2017 campaign, describe Survation as “the only major polling company whose polls predicted the election correctly”. Our chief executive’s chapter in the same book, Fear, Groupthink and Outliers, sets out the method and the pressure not to publish a poll that disagreed with everyone else.
2019: the lead within a point
Our final telephone poll put the Conservatives on 45% and Labour on 34%. The result was 44.7% and 33.0%. We understated the Liberal Democrats by about three points.
2024: right story, lead too large
Our final poll, like every final poll that year, showed a Labour landslide and overstated its margin. We had Labour 18 points ahead; the result was 10.3. The average error across parties was 1.9 points. We published a review of our telephone polling through the British Polling Council.
2015: the poll we did not publish
Our published online poll in 2015 showed a tie, as most polls did. The Conservatives won by 6.6 points. The evening before the vote we ran a telephone poll that put them 6.7 points ahead, almost exactly the result. It looked out of line with everything else and we held it back. We sent the raw data to the British Polling Council the following day. We said so the next day, and it changed how we work: since 2015 our final call at every general election has been made by telephone.
Referendums
| Vote | Survation final poll | Result | Error |
|---|---|---|---|
| Scottish independence referendum, 18 September 2014 | Yes 47%, No 53% (telephone, 16 to 17 September, for the Daily Record) | Yes 44.7%, No 55.3% | 2.3 points |
| EU referendum, 23 June 2016 | Remain 51%, Leave 49% (telephone, 1,003, 20 June, for IG) | Remain 48.1%, Leave 51.9% | 2.9 points |
2014: the most accurate final poll
Our final telephone poll for the Daily Record put Yes on 47% and No on 53%. Scotland voted No by 55.3% to 44.7%. Ipsos MORI published the same rounded figures; on the unrounded figures ours was the closest of any company, by a tenth of a point. Tables.
2016: close, but on the wrong side
Every final poll overstated Remain. The British Polling Council’s record shows an average error of 4.3 points across seven final polls. Ours was the second smallest, but we still had Remain narrowly ahead, and Leave won.
Scottish Parliament elections
Survation has polled Scotland continuously since before the 2014 referendum. Holyrood elections have two ballots, so each final poll is tested twice.
| Election | Ballot | Survation final poll | Result |
|---|---|---|---|
| 2026 | Constituency | SNP 39, Lab 19, Reform 17, Con 12, LD 10, Green 2 | SNP 38.2, Lab 19.2, Reform 15.8, Con 11.8, LD 11.4, Green 2.3 |
| 2026 | Regional list | SNP 29, Reform 17, Lab 16, Green 15, Con 13, LD 8 | SNP 27.2, Reform 16.7, Lab 16.0, Green 14.0, Con 11.8, LD 9.4 |
| 2021 | Constituency | SNP 49, Con 21, Lab 21, LD 8 | SNP 47.7, Con 21.9, Lab 21.6, LD 6.9 |
| 2016 | Constituency | SNP 49, Lab 21, Con 19, LD 7 | SNP 46.5, Lab 22.6, Con 22.0, LD 7.8 |
| 2016 | Regional list | SNP 43, Con 20, Lab 19, Green 7, LD 6, UKIP 2 | SNP 41.7, Con 22.9, Lab 19.1, Green 6.6, LD 5.2, UKIP 2.0 |
2026: lowest error on the constituency vote and on seats
Our final call used 5,025 interviews in Scotland, online and by telephone, between 20 April and 5 May 2026.
| Measure | Survation | Next best | Companies compared |
|---|---|---|---|
| Constituency vote, average error per party | 0.64 points (1st) | YouGov 1.05 | 6 |
| Regional list vote, average error per party | 1.26 points (2nd) | YouGov 1.06 (1st) | 6 |
| MRP seat projection, average error per party | 1.17 seats (1st) | YouGov and More in Common 2.33 | 6 |
Our MRP projected the SNP on 59 seats; they won 58. Projected and actual seats for the other parties: Labour 17 and 17, Reform UK 18 and 17, Greens 16 and 15, Conservatives 13 and 12, Liberal Democrats 7 and 10. Full comparison with every company’s figures.
2021: closest on the constituency vote
Our final poll was within about one point of the result for every party on the constituency ballot, the closest of the final polls.
2016: the Conservatives overtake Labour
Our final poll for the Daily Record, by telephone on 1 and 2 May, put the Conservatives ahead of Labour on the list vote, the change that made them Holyrood’s second party. It also had the SNP well ahead on both ballots. Like other polls it understated the Conservative share by about three points. Poll write-up.
Constituency polls: selected results
Constituency polls cannot be ranked the way final national polls are. Nobody keeps a central record of them (the British Polling Council compares national final polls only), and a seat poll may be taken days or weeks before the vote. So this is a selection of our published seat polls, not a league table. Survation is one of very few companies that polls single seats by telephone, and most of that work is private. We have polled individual seats for all the major parties in these islands. One client alone commissioned 139 polls in 73 constituencies between August 2013 and April 2015, the start of a relationship that has run across four general elections.
| Seat | Survation poll | Fieldwork ended | Result |
|---|---|---|---|
| Clacton by-election, 13 August 2026 (two-candidate question) | Farage 69, Count Binface 31 (telephone, 502) | 20 days before | Farage 70, Binface 30 of the two-candidate vote |
| Makerfield by-election, 18 June 2026 | Burnham (Lab) 49, Kenyon (Reform UK) 39 (telephone, 518) | 17 days before | Lab 54.8, Reform UK 34.5 |
| Mid Bedfordshire by-election, 19 October 2023 | Within two points of every main party among those certain to vote (telephone, 559) | 34 days before | Lab 34.1, Con 31.1, LD 23.1: Labour gain |
| Hartlepool by-election, 6 May 2021 | Con 50, Lab 33 (telephone, 517) | 7 days before | Con 51.9, Lab 28.7: Conservative gain |
| Wrexham, general election 2019 | Con 44, Lab 29 (telephone, 405) | 12 days before | Con 45.3, Lab 39.0: Conservative gain |
| Great Grimsby, general election 2019 | Con 44, Lab 31 (telephone) | November 2019 | Con 54.9, Lab 32.7: Conservative gain |
| Workington, general election 2019 | Con 45, Lab 34 (telephone) | October 2019 | Con 49.3, Lab 39.2: Conservative gain |
| Rochester and Strood by-election, 20 November 2014 | UKIP 48, Con 33, Lab 16 (telephone, 1,012) | 23 days before | UKIP 42.1, Con 34.8, Lab 16.8 |
| Clacton by-election, 9 October 2014 | UKIP 64, Con 20, Lab 13 (telephone, 700) | 41 days before | UKIP 59.7, Con 24.6, Lab 11.2 |
| Newark by-election, 5 June 2014 | Con 42, UKIP 27, Lab 22 (telephone, 678) | 2 days before | Con 45.0, UKIP 25.9, Lab 17.7 |
Clacton, 2026. Asked to choose between the top two candidates only, our poll gave Nigel Farage 69% and Count Binface 31%. Their shares of the two-candidate vote on the day were 70% and 30%.
Makerfield, 2026. Ours was the first poll of the contest. It showed Andy Burnham ahead as a named candidate in a seat where Reform UK led by 11 points on party labels alone, and our second poll showed his lead growing. He won by 20. We also identified why every published poll understated him: people asked about their 2024 vote were often recalling their vote for him as mayor.
Mid Bedfordshire, 2023. Five weeks before polling day our poll showed the Conservative vote collapsing in a seat they had held since 1931. Among respondents certain to vote, our figures were within one point of Labour’s final share, two of the Conservatives’ and 0.2 of the Liberal Democrats’. Our review.
Hartlepool, 2021. A 17-point Conservative lead in a seat Labour had held since its creation, published when many expected Labour to hold on. The Conservatives won by 23.
The Red Wall, 2019. Our telephone polls in Workington, Great Grimsby and Wrexham each showed the Conservatives ahead in seats Labour had held for decades. All three changed hands.
The UKIP by-elections, 2014. We polled Clacton by telephone within a day of Douglas Carswell’s defection and found him on 64%; he won with 60%. In Rochester and Strood we had Mark Reckless winning with the Conservatives on 33% and Labour on 16%; they finished on 35% and 17%. In Newark, two days out, we were within three points on the Conservatives and one on UKIP.
London mayoral elections
We have named the winner at each of the three London mayoral elections we polled. In 2024 our poll for ITV News had Sadiq Khan on 44% five weeks out; he won with 43.8%. In 2016 we had Zac Goldsmith on 34%; he took 35.0%.
| Election | Survation poll | Fieldwork ended | Result |
|---|---|---|---|
| 2 May 2024 | Khan 44, Hall 26 (1,019, for ITV News) | 37 days before | Khan 43.8, Hall 32.7 |
| 5 May 2016 | Khan 49, Goldsmith 34 (1,010) | 10 days before | Khan 44.2, Goldsmith 35.0 |
| 3 May 2012 | Johnson 42, Livingstone 31 (1,443) | 9 days before | Johnson 44.0, Livingstone 40.3 |
Need a seat polled? See telephone polling.
MRP seat projections
MRP (multilevel regression and post-stratification) is a statistical model that uses data from a large survey to estimate the result in every seat. Survation has run MRP models since 2017. What is MRP?
| Election | Survation MRP projection | Result | Verdict |
|---|---|---|---|
| Scottish Parliament 2026 | SNP 59, Reform UK 18, Lab 17, Green 16, Con 13, LD 7 | SNP 58, Reform UK 17, Lab 17, Green 15, Con 12, LD 10 | Lowest seat error of the six published MRP models |
| General election 2024 | Lab 470, Con 68, LD 59, Reform UK 15, SNP 14 | Lab 411, Con 121, LD 72, SNP 9, Reform UK 5 | Right on the landslide, too large on its size |
| General election 2019 (private clients) | Con 353, Lab 219, LD 11 | Con 365, Lab 202, LD 11 | 94.3% of seats called correctly, ahead of YouGov’s MRP on 93.8%. Dr Chris Hanretty’s review |
| General election 2017 (private clients) | Con 314, Lab 255, SNP 54, LD 6: no overall majority | Con 318, Lab 262, SNP 35, LD 12: hung parliament | Right on the outcome: the model gave a 60% probability of a hung parliament |
In 2024 our model overstated Labour’s majority, as our polling overstated Labour’s lead. Dr Chris Hanretty’s review of the model is published by the British Polling Council. The 2026 Holyrood model, built on a mixed online and telephone sample, followed that review. More on our MRP and statistical modelling.
Outside the UK
Survation also runs election research abroad. At Georgia’s parliamentary election on 31 October 2020 we ran the exit poll for the national broadcaster Rustavi 2. Our 8pm figures put Georgian Dream on 51.4% and the main opposition bloc on 25.4%; the official result was 48.2% and 27.2%. Every smaller party was within one point of its official share. Four exit polls were published that night, and between them they gave Georgian Dream anywhere from 41% to 55%.
We have also polled presidential and midterm elections in the United States, and an Italian referendum and an Italian general election for private clients.
Beyond elections, we run surveys in many countries. Recent examples:
- Britain, Canada, Chile and Germany: surveys of the public for a University of East Anglia study of climate knowledge, published in a peer-reviewed journal.
- Canada, the United States, the UK, France, Italy, the Netherlands and Argentina: visitors, business leaders and investors for a Caribbean investment agency.
- Europe and beyond: a multi-country consumer survey for a global consumer staples brand.
- Several countries, including markets in Asia: senior decision-makers for HSBC’s Funding the Future survey.
- Romania: young TikTok users for Global Witness.
- Spain: a national survey of 1,500 adults for a university research team.
What we got wrong, and what we changed
No polling company is right every time, and a track record that lists only the wins is not a record. These are ours.
| When | What happened | What we did about it |
|---|---|---|
| 2015 general election | Our published online poll showed a tie; the Conservatives won by 6.6 points. An accurate telephone poll went unpublished. | Moved our final election call to telephone and committed to publishing what the data says, outlier or not. That decision produced the 2017 result. |
| 2016 EU referendum | Our final poll had Remain ahead by two points; Leave won by four. | Reviewed contact patterns by age and time of day in telephone fieldwork. |
| 2024 general election | Our final poll overstated Labour’s lead by 7.7 points and our MRP overstated Labour’s seats by 59. | Published reviews of both through the British Polling Council; combined online and telephone samples for our 2026 Holyrood call. |
Check our record independently
You do not have to take our word for any figure on this page.
| Source | What it holds |
|---|---|
| British Polling Council | Every member’s final poll beside the result for 2017, 2019 and 2024, the 2016 EU referendum, and companies’ own reviews of 2024 |
| Election Data Vault pollster ratings | An independent rating built from final-poll accuracy at the last three general elections (70%), longevity (10%) and BPC membership (20%). Survation is rated joint second of 20, with a score of 88 (grade A-). |
| Political Communication in Britain (Palgrave Macmillan, 2019) | The academic study of the 2017 campaign, edited by Dominic Wring, Roger Mortimore and Simon Atkinson. Includes Professor Will Jennings’s analysis of every final poll and a chapter by Survation’s chief executive. |
| Survation archive | Full data tables for every poll we have published |
How we measure accuracy
- Final poll: the last poll a company publishes before polling day. It is the only poll that can be tested against a real result.
- Average error per party: the gap between poll and result for each party, ignoring direction, averaged across parties. Lower is better.
- Error on the lead: the gap between the poll’s lead for the first party over the second and the real one. This is usually what decides whether a poll told the right story.
- Great Britain figures: general election comparisons use results for Great Britain, the basis the British Polling Council uses.
Questions and answers
- How accurate is Survation?
- Survation produced the most accurate final poll of the 2017 general election and of the 2014 Scottish independence referendum, and had the lowest error on the constituency vote and on MRP seat projections at the 2026 Scottish Parliament election. Election Data Vault rates Survation joint second of 20 UK pollsters, with a score of 88 and grade A-, on a measure built mainly from accuracy at the last three general elections. In 2024 Survation, like the rest of the industry, overstated Labour’s lead.
- Which pollster was most accurate at the 2017 general election?
- Survation. Its final poll put the Conservatives on 41% and Labour on 40%, against a result of 43.5% and 41.0%. It had the lowest error of the nine final polls recorded by the British Polling Council.
- Did any poll predict the 2017 hung parliament?
- Survation’s final poll was the only one from a British Polling Council member to point to a hung parliament. The others showed Conservative leads of 5 to 13 points.
- Which pollster was most accurate at the 2014 Scottish independence referendum?
- Survation’s final telephone poll, at Yes 47% and No 53%, was the closest to the result of Yes 44.7% and No 55.3%, narrowly ahead of Ipsos MORI.
- Which pollster was most accurate at the 2026 Scottish Parliament election?
- Survation had the lowest average error on the constituency vote (0.64 points) and on MRP seat projections (1.17 seats), and projected 59 SNP seats against a result of 58. YouGov had the lowest error on the regional list vote, with Survation second.
- Is Survation a member of the British Polling Council?
- Yes. Survation is a member of the British Polling Council and follows its disclosure rules: full tables for every published poll, who commissioned it, and how it was done. Survation is also a Market Research Society Company Partner.
- How accurate are constituency polls?
- Less precise than national polls. Samples are usually around 500, so the margin of error on each party is about four points, and by-elections can move late. There is no central record of them, and timing varies. As one example, Survation’s Hartlepool poll in 2021 showed a 17-point Conservative lead a week before a 23-point win.
- How accurate is MRP?
- It depends on the survey behind it and on the model. Survation’s MRP was the most accurate of six at the 2026 Scottish Parliament election, called 94.3% of seats correctly at the 2019 general election, and indicated a hung parliament in 2017. In 2024 it projected the Labour landslide but overstated its size.
- Can I commission a poll from Survation?
- Yes. Survation runs national polls, constituency polls, MRP and wider market research for companies, charities, public bodies and campaigns. Email researchteam@survation.com or call 020 3818 9661.
Want research you can check? We run national and constituency polls, MRP, focus groups and market research for companies, charities, public bodies and campaigns. Get a quote, email researchteam@survation.com or call 020 3818 9661. See also our voting intention tracker and services.