A headline gives you five or six party percentages. The tables behind it tell you who answered, what they were asked and which people counted towards those percentages. They can also explain an apparent discrepancy between the figure on a chart and the figure in a newspaper.
Read the question and its base together. A percentage without its denominator is incomplete information, however prominent it looks on the page.
Start with the fieldwork and the population
The title page or methodology note should identify the polling organisation, commissioner, interview dates, sample and geographical coverage. Check whether the survey covers British adults, eligible voters, likely voters or a narrower group. “National” does not necessarily mean the whole United Kingdom.
The British Polling Council’s disclosure rules set out the information members should make available. Those rules are useful when looking for the question wording, interview method and weighting information, rather than treating a clipped social-media graphic as the full poll.
Follow the denominator
In an illustrative survey of 1,000 people, 200 give no party preference. Of the remaining 800, 240 choose Party A. Party A has 24% of all respondents but 30% of those expressing a party preference. Both calculations are correct; they answer different questions.
The distinction becomes more complicated when turnout filters, follow-up questions or statistical modelling are introduced. Read the notes beneath the voting-intention table rather than assuming the headline is obtained by removing one row of “don’t knows”. Our guide to undecided voters and turnout follows those choices in more detail.
YouGov’s published explanation of its method explicitly distinguishes its modelled headline from tables of answers before that modelling. A reader who tries to recreate the headline directly from an earlier table may therefore obtain a different number without finding an error.
Weighted bases are not headcounts
A table may display an unweighted base of 150 and a weighted base of 200 for a subgroup. That means 150 people supplied answers, which collectively contribute the equivalent of 200 after adjustment. It does not mean another 50 people were interviewed.
The weighted figure is useful for following the calculation, while the unweighted figure helps assess how many actual respondents sit behind it. Neither alone gives a complete account of precision. Unequal weights and the sampling design can affect the effective amount of information; see how weighting works.
Be careful with crossbreaks
Columns for age, region or past vote invite comparisons, but their samples are smaller than the whole survey. A national poll containing 2,000 respondents can have a particular regional or demographic column based on only a small fraction of them.
A five-point movement within that column can be much less persuasive than a five-point national movement. Looking through enough columns will also reveal some striking results by chance. Margins of error explains why subgroup estimates need particular care.
Check what changed between polls
Before comparing two releases, check that the question, population and processing are comparable. A new party prompt or turnout model can break the series. Different fieldwork dates can also make two apparently conflicting polls less contradictory than they first appear.
Record the actual percentages and bases, rather than the rounded change alone. Rows may total 99% or 101% because each figure has been rounded independently. Missing figures should remain missing: an unpublished party share cannot be filled with zero.
Use the original source links in our polling database to inspect releases. For differences which persist beyond a single set of tables, move on to why pollsters disagree.
Sources and further reading
Reviewed 7 October 2026. Worked examples are illustrative unless an election or poll is identified.