How accurate the pollsters were

For every pollster: the average difference between its final poll and the result, how many elections that rests on, and the period those polls span.

Pollster Average difference Elections Period
Forecasting Statistics 0.9 percentage points 1 too few elections 1959-10-01 – 1959-10-01
Verian 1.1 percentage points 3 2017-06-01 – 2024-06-28
ORC 1.3 percentage points 3 1970-06-13 – 1974-10-09
Survation 1.3 percentage points 4 2015-05-06 – 2024-07-01
RNB 1.4 percentage points 1 too few elections 2010-04-30 – 2010-04-30
Number Cruncher Politics 1.4 percentage points 2 too few elections 2019-12-08 – 2024-07-03
Qriously 1.5 percentage points 2 too few elections 2017-06-04 – 2019-12-05
Ipsos MORI 1.5 percentage points 12 1979-05-02 – 2024-07-01
More in Common 1.5 percentage points 1 too few elections 2024-06-28 – 2024-06-28
Gallup 1.5 percentage points 16 1945-06-24 – 2001-06-06
Deltapoll 1.7 percentage points 2 too few elections 2019-12-09 – 2024-06-29
Business Decisions 1.7 percentage points 2 too few elections 1974-02-21 – 1974-10-02
YouGov 1.7 percentage points 7 2001-06-04 – 2024-06-25
Opinium 1.8 percentage points 5 2010-05-04 – 2024-07-01
Populus 1.8 percentage points 3 2005-05-02 – 2015-05-05
JL Partners 1.9 percentage points 1 too few elections 2024-07-02 – 2024-07-02
Panelbase 1.9 percentage points 3 2015-05-01 – 2019-12-10
ComRes 1.9 percentage points 5 2005-04-23 – 2019-12-09
We Think 1.9 percentage points 1 too few elections 2024-07-02 – 2024-07-02
Focaldata 2.0 percentage points 2 too few elections 2019-11-27 – 2024-06-27
ICM 2.1 percentage points 8 1992-04-08 – 2019-12-06
SurveyMonkey 2.1 percentage points 2 too few elections 2015-04-30 – 2017-06-04
Stonehaven 2.1 percentage points 1 too few elections 2024-06-28 – 2024-06-28
Find Out Now 2.1 percentage points 1 too few elections 2024-07-03 – 2024-07-03
Rasmussen 2.2 percentage points 1 too few elections 2001-06-02 – 2001-06-02
BMG 2.2 percentage points 4 2015-05-03 – 2024-06-30
Harris 2.2 percentage points 9 1970-06-14 – 2010-05-04
Savanta 2.2 percentage points 2 too few elections 2019-12-09 – 2024-07-02
Norstat 2.3 percentage points 2 too few elections 2017-05-31 – 2024-07-01
ORB 2.3 percentage points 1 too few elections 2017-05-31 – 2017-05-31
NOP 2.4 percentage points 13 1959-10-02 – 2005-05-01
Marplan 2.4 percentage points 6 1970-06-11 – 1987-06-09
Techne 2.5 percentage points 1 too few elections 2024-06-28 – 2024-06-28
Whitestone Insight 2.5 percentage points 1 too few elections 2024-07-01 – 2024-07-01
Redfield & Wilton 2.7 percentage points 1 too few elections 2024-06-28 – 2024-06-28
Lord Ashcroft 2.7 percentage points 2 too few elections 2015-05-05 – 2024-06-27
Research Services Ltd 2.7 percentage points 4 1951-10-19 – 1979-04-24
BPIX 2.8 percentage points 2 too few elections 2005-04-28 – 2010-04-30
Nielsen 2.8 percentage points 1 too few elections 1992-03-28 – 1992-03-28
TNS BMRB 2.9 percentage points 2 too few elections 2010-04-29 – 2015-04-30
NMR 3.1 percentage points 1 too few elections 1992-03-24 – 1992-03-24
Daily Mail 3.2 percentage points 1 too few elections 1950-02-05 – 1950-02-05
ASL 3.4 percentage points 3 1983-06-07 – 1992-04-03
People Polling 3.9 percentage points 1 too few elections 2024-07-02 – 2024-07-02
Angus Reid 4.0 percentage points 1 too few elections 2010-05-04 – 2010-05-04
Centre of Public Opinion 4.1 percentage points 1 too few elections 1945-06-15 – 1945-06-15
OnePoll 8.0 percentage points 1 too few elections 2010-04-29 – 2010-04-29

A pollster with fewer than 3 scored elections is listed here with its figures and marked as such. Its average is shown for completeness, not as a verdict: an average over one or two elections says more about those elections than about the pollster.

Every election we hold a result for

How this is measured: for each past election we take each pollster’s last poll in the 30 days before polling day and compare it, party by party, with the result. The figure is the mean absolute difference across the parties that appear in both. Only parties the pollster actually reported are counted. A late shift in opinion is not the same thing as a polling error.