In this article
Opinion polls appear constantly, especially before elections. Used carefully, they are one of the best tools we have for understanding what the public thinks. Read carelessly, they can mislead.
Key facts
- A snapshot: a poll measures opinion when it was taken, not on election day.
- Margin of error: a typical poll of about 1,000 people has a margin of error of roughly plus or minus 3 percentage points.
- US 2020: national polls overstated Joe Biden’s winning margin by 3.9 points on average.
- UK 2024: final polls put Labour at about 39 percent; it won about 35 percent in Great Britain.
How a poll works
It is impossible to ask everyone, so pollsters question a sample, often around 1,000 people, and use statistics to estimate what the whole population thinks. The sample must resemble the population. Pollsters therefore adjust, or weight, their results so that the mix of ages, regions, education levels and past votes matches official figures.
Weighting is where much of the skill, and much of the error, lies. If certain kinds of people are less likely to answer surveys, and weighting does not fully correct for that, the poll will be off in a predictable direction.
What the margin of error means
Every sample-based poll has a margin of error. For a sample of about 1,000, it is roughly 3 percentage points either way, with 95 percent confidence. If a poll shows Party A on 42 percent and Party B on 40 percent, the true figures could plausibly be anywhere from 39 to 45 and 37 to 43. The poll cannot say with confidence who is ahead.
| Sample size | Approximate margin of error |
|---|---|
| 500 | plus or minus 4.4 points |
| 1,000 | plus or minus 3.1 points |
| 2,000 | plus or minus 2.2 points |
| 5,000 | plus or minus 1.4 points |
The margin of error only covers random chance. It does not cover problems such as a biased sample, a badly worded question or people changing their minds.
When polls miss
Polls are usually close but not perfect, and their errors often lean the same way. After the 2020 US election, the American Association for Public Opinion Research found that national polls had overstated Joe Biden’s margin over Donald Trump by 3.9 points on average, the largest error in decades. Biden still won, but by less than the polls suggested.
In the UK’s July 2024 general election, the final average of polls put Labour at about 39 percent. It won about 35 percent in Great Britain. Labour still won a large majority, but its share was overestimated by around 4 points, the second-largest such error on record for the party.
Five checks before you share a poll
- Who did it, and who paid? Look for an established pollster and a clear sponsor.
- How many people, and when? Bigger and more recent is better.
- How were they reached? Online panels, phone calls and door-to-door surveys all have strengths and weaknesses.
- What exactly was asked? Wording and question order can change answers.
- Is the gap bigger than the margin of error? If not, the race is too close to call.
Tip: an average of several reputable polls is almost always a better guide than any single survey. Individual polls bounce around; averages smooth out the noise.
In Praise of Reading Past the Headline
Polls and the news
A poll showing a small change is often reported as a dramatic shift. Before reacting, read beyond the headline; our columnist makes the case for reading past the headline. Polls also matter after elections, because parties in a coalition watch them closely when deciding whether to stay in government.
Why polls disagree with each other
Two polls taken in the same week can show different results without either being wrong. Common reasons include:
- Method: online panels, phone calls and face-to-face interviews reach different kinds of people.
- Weighting: pollsters make different choices about how to balance their samples, for example by past vote or education.
- Likely voters: some polls report everyone, others only people judged likely to vote.
- Question wording: whether undecided voters are pushed to choose, and whether party leaders are named, can shift results.
- Random chance: even two identical polls will differ within the margin of error.
Frequently asked questions
Are online polls reliable?
Many reputable pollsters use online panels of people recruited in advance and weighted carefully. Open “click to vote” website polls are not reliable because anyone can take part, often more than once.
Why do pollsters ask only about 1,000 people?
Beyond around 1,000 to 2,000 people, adding more respondents reduces the margin of error only slightly while raising costs sharply. Getting the right mix of people matters more than getting a huge number.
Do polls influence how people vote?
Researchers disagree on how much. Polls can affect campaign strategy, media coverage and tactical voting, which is one reason reading them carefully matters.
