A poll of polls combines results from several surveys to give a broader view of party support. It can make the trend easier to follow when individual polls bounce around or disagree.
The result depends on which polls are included and how much influence each receives. Two averages can produce different figures from the same set of polls.
Which polls count?
Some averages include every poll published during a chosen period. Others take only the latest from each polling organisation. They may also give newer polls or larger samples more weight.
For example, one firm publishes four polls putting a party on 30%, while another publishes one putting it on 24%. Averaging all five gives 28.8%. Giving the two firms equal influence gives 27%.
The difference comes from how often the first firm polled, rather than any disagreement over the arithmetic.
How the UK Decides average works
Our Poll of Polls takes the most recent qualifying national Westminster poll from each pollster within a rolling 30-day window. Each included pollster contributes equally.
We calculate an arithmetic mean for each party, without sample-size weighting or a house-effect adjustment. Restore Britain is averaged across qualifying polls that publish a separate Restore figure; an unpublished figure is not counted as zero.
This approach prevents a frequent pollster from gaining extra influence simply by releasing more surveys. A poll near the start of the window can still sit alongside one conducted much more recently.
Why an average moves
A new release can replace a pollster’s previous result. An organisation can also enter the window or leave it when its last qualifying poll becomes too old.
An average of 24%, 25% and 29% is 26%. If the 29% poll drops out, the average becomes 24.5%, although neither remaining poll has changed.
That kind of movement matters when firms have persistent differences. The contributing polls help explain whether the average moved because of new findings or because its membership changed.
What can it tell us?
Averaging can reduce some of the noise in individual polls. It cannot remove an error shared by several firms, such as difficulty reaching a group of voters or misjudging turnout.
A calendar-month average may also use different interviews from a rolling 30-day figure. The dates and calculation alongside each chart explain the comparison.
The national average estimates recent voting intention. Turning that into MPs requires a further calculation about where the votes are, which is the job of a seat model.
Reviewed 19 October 2026.