An average gives readers a manageable view of a crowded polling calendar. It can also hide important choices: which polls qualify, how old they can be and how much influence each organisation receives.
A poll of polls is a calculation with a methodology, not a separate survey. Its usefulness depends on whether that calculation answers the question readers think it answers.
Several averages can all be calculated correctly
Take an illustrative week containing four polls from Organisation A at 30% and one from Organisation B at 24%. Giving each poll equal weight produces 28.8%. Giving each organisation equal weight produces 27%.
Neither number contains an arithmetic error. The first gives the frequent pollster four times as much influence; the second treats the two organisations equally. These are different decisions about how to summarise the evidence.
A sample-size weighted average makes another choice. It assigns more influence to larger surveys, although nominal sample size does not capture every feature of weighting or uncertainty.
What does UKDecides do?
Our Poll of Polls uses the latest qualifying national poll from each pollster within a rolling 30-day window, giving each included pollster equal weight. The methodology page describes the site’s wider approach to sourcing and data.
That prevents a prolific organisation from dominating simply by releasing more polls. It also means a poll near the beginning of the window can contribute alongside a much newer one. The average summarises recent evidence, rather than measuring everyone’s opinion on the date it appears.
Check a chart’s own definition before assuming a monthly series and a rolling current figure use identical calculations. A calendar-month summary and a rolling window can cover different interviews even when their labels look similar.
Why a rolling average can move without a new swing
The latest poll can replace an older poll from the same organisation. A new organisation can enter the window, or an old one can leave it. Each event can move the average without demonstrating a fresh change in opinion of the same size.
For illustration, an equal-weight average of 24%, 25% and 29% is 26%. Removing the 29% observation gives 24.5%, even though neither remaining observation changed. That is a composition effect.
Persistent differences between pollsters make these entries and exits especially relevant. Read changes alongside the underlying releases, rather than interpreting every decimal point as a new political development.
What does averaging improve?
Combining several measurements can smooth some idiosyncratic variation. It does less for a mistake shared across organisations, such as a common difficulty reaching a particular group or a similar turnout assumption.
The observations are not necessarily independent in every useful sense. They can share recruitment sources, reference data and methodological conventions. A narrow spread does not establish that a systematic error is absent.
How to report an average
Identify the time window, contributing organisations and latest fieldwork. Show enough of the inputs for readers to understand an unusual movement. Avoid presenting decimal precision as equivalent to certainty.
Most importantly, keep the inference proportionate. A national average describes voting-intention estimates. Turning it into seats requires further assumptions about where party support is located, as our guide to seat forecasts explains.
Sources and further reading
- British Polling Council: information needed to assess published polls
- British Polling Council: sources of polling error
Reviewed 7 October 2026. Worked examples are illustrative unless an election or poll is identified.