SAMPLING
A random thousand
If half of everyone holds a view, there is a 95% chance a random sample of a thousand will put it between 47% and 53%.
See the numbersStart
WHAT THE GUIDANCE SAYS
A thousand people,
plus or minus three.
A good poll asks a small sample to stand in for everyone.
That works surprisingly well, but only within limits. UK polling and statistics guidance comes back to the same few checks: who was asked, how many, what exactly they were asked, and whether other polls agree.
± 3 POINTSfor about 1,000 people
Weigh a resultKeep in mind
- Check the sample
- Mind the margin
- Look for agreement
- Small habits, sharper reading
THE SHORT ANSWER
A sample is
a careful guess.
Polls estimate, they do not count everyone.
Every figure comes with a range around it.
MARGIN OF ERROR
51 to 49 is a tie
The British Polling Council says a 51% to 49% split does not show one side ahead. The public is more or less evenly divided.
Weigh a result
WORDING
The question matters
Good polls use plain language, ask whether people agree or disagree rather than leading them, and avoid asking two things at once.
Questions to askGeneral information only, not legal advice. For advice on your own situation, speak to a qualified lawyer or a free advice service.
A QUICK RESULT CHECK
What kind of result
are you looking at?
Pick the kind of poll finding you have seen to see what the guidance says about it. The bar shows how much weight the guidance suggests giving it.
Based on the British Polling Council's guidance for journalists (2023) and Full Fact's guide to polls.
TWO NUMBERS WORTH KNOWING
Precision gets
expensive fast.
One figure for a typical poll,
one for what it takes to be three times as precise.
± 3 points
Margin of error for a poll of 1,000 people
Full Fact says a poll of 1,000 people typically comes with a margin of error of plus or minus three per cent. The fewer people surveyed, the greater the margin of error.
TYPICAL POLL · FULL FACT
9,604
People needed for a margin of plus or minus one point
Narrowing the margin to one point either way would need a sample of 9,604 people, which Full Fact notes is rarely practical.
HIGH PRECISION · FULL FACT
The Office for National Statistics explains what 95% means: if 20 random samples were taken and an interval worked out for each, 19 of the 20 intervals would contain the true population value.
FOUR QUESTIONS TO ASK
Before you share
a poll.
British Polling Council members must publish key details within two working days of results being published. These are the ones worth checking.
Step 1
Who paid for it?
Members must say which client commissioned the survey, so you know who it was run for.
Step 2
Who and how many?
Look for the sample size, who it represents (for example all adults) and how people were interviewed.
Step 3
When was it done?
The dates of interviewing tell you whether the poll still reflects opinion now.
Step 4
What exactly was asked?
Where practical, members publish the full question wording. Check whether a question might lead people to a particular answer.
BEYOND THE MARGIN
Not every error
is random.
The margin of error only covers the chance of drawing an unusual sample. The ONS points out other errors too: people who cannot be reached or refuse, inaccurate answers, and processing mistakes. These would be present even if everyone had been surveyed, and are usually hard to measure.
Pollsters often weight their data so that groups such as men and women, or younger and older people, match their share of the population. The British Polling Council says weighting reduces, but does not eliminate, the risk of an inaccurate poll.
Questions
Questions on the way
What does margin of error mean?
It is the range around a poll figure that the true value probably falls within. For a random sample of about 1,000, a 50% result has a 95% chance of being within 47% to 53% of the whole population's view.
Why do polls only ask about 1,000 people?
Because precision improves slowly as samples grow. A poll of 1,000 typically has a margin of about 3 points, while getting down to 1 point would need 9,604 people.
Are online click polls reliable?
Not usually. The British Polling Council says self-selected polls on websites or social media should usually be avoided and ignored, because nothing is known about who responded.
Does British Polling Council membership mean a poll is good?
No. The Council says membership represents a commitment to transparency, but does not represent a quality mark in the conduct of polls.
How do 'don't knows' change a result?
They can change the picture a lot. In the Council's example, if 30% say they don't know, a figure of 60% among those with a view is only 42% of the whole sample.
Destination