How to read opinion polls in the news

By Omid Jamshidi · Published 23 October 2026 · 5 min read

What a poll's sample size and margin of error really mean, why a two-point lead is often no lead at all, and seven questions to ask before you trust a poll headline.

"Candidate A takes the lead", "Party B slumps to record low" – headlines like these are often based on a single opinion poll. Polls are useful: they are one of the few ways to learn what millions of people think. But a poll is an estimate, not a count, and many headlines treat it as more precise than it is. This guide explains the basics you need to read polls with confidence.

How a poll works

A polling company asks a sample of people – often around 1,000 – the same questions and uses their answers to estimate the opinion of the whole population. This works only if the sample resembles the population. Pollsters reach people by phone, online panels or, more rarely, face to face, and then weight the answers: if young people or a region are under-represented in the sample, their answers count a little more, so that the result matches the population by age, gender, region, education and often past voting.

This is why two polls taken on the same day can differ. They may use different methods, different weighting and different questions – and each of these choices affects the result.

The margin of error

Because only a sample is asked, every result has a margin of error. For a sample of 1,000 people and a value around 50%, it is about ±3 percentage points (at the usual 95% confidence level). Roughly:

• 500 people: about ±4.4 points
• 1,000 people: about ±3.1 points
• 2,000 people: about ±2.2 points

So if a poll of 1,000 people puts a candidate at 48%, the real value is very likely somewhere between about 45% and 51%. Larger samples help, but slowly: four times as many interviews only halve the margin.

Three consequences are often missed in news reports:

1. The gap between two candidates is less certain than each share. If A has 46% and B has 44% in a poll of 1,000, the margin for the gap is almost ±6 points. A two-point lead is, statistically, no clear lead at all.
2. Subgroups are much less precise. If a headline says "young voters turn away", but only 200 young people were asked, the margin is about ±7 points.
3. The margin of error covers only chance. It says nothing about other problems: people who refuse to answer, online panels that are not representative, leading questions or voters who change their mind. Real polls are therefore less precise than the margin suggests.

Trends matter more than single polls

One poll is a snapshot with noise. A change of one or two points from last month is usually within the margin and can be pure chance. More reliable signals are:

• several polls from different companies pointing in the same direction;
• a trend over weeks, not a jump in one poll;
• poll averages, which many media publish before elections and which smooth out the noise of individual polls.

A poll that differs strongly from all others is more likely an outlier than the first sign of a sensation – even though it often gets the biggest headline.

Different countries, different rules

• United States: national polls measure the popular vote, but the president is chosen by the Electoral College. That is why polls in a few contested states are often more telling than the national figure.
• Germany: the weekly "Sonntagsfrage" asks which party people would vote for if there were a federal election next Sunday. Because parties generally need 5% of the vote to enter the Bundestag, small parties near that threshold get much attention – yet with 1,000 respondents, a party at 5% could really be anywhere between about 3.6% and 6.4%. On election night, the first figure at 6 p.m. is a forecast based on exit polls; later "projections" are based on counted votes.
• Spain: the public Centre for Sociological Research (CIS) publishes regular surveys alongside private pollsters, and electoral law forbids publishing election polls in the five days before a vote.

In countries where freedom of expression is restricted, polls face an additional problem: respondents may not say what they really think, especially by phone. Results from such countries should be read with extra caution, whoever conducted the poll.

Seven questions to ask about a poll

1. Who conducted it, and who paid for it? A poll commissioned by a party or campaign is not automatically wrong, but it deserves a closer look.
2. When was it conducted? Events since then may have changed opinions.
3. How many people were asked, and how? Phone, online panel, face to face?
4. What exactly was the question? Wording can change answers considerably.
5. What is the margin of error – and is the reported change larger than it?
6. What do other polls say? Compare with other companies and averages.
7. How are undecided voters handled? A "lead" can look very different depending on whether they are left out.

Reputable polling companies publish this information, and serious media report at least the date, the sample size and the company.

Polls on Omid57.info

Omid57.info shows headlines from many outlets, so you will often see several reports about different polls side by side – sometimes with opposite headlines on the same day. The questions above help to compare them. Our guide on how to read the news critically has more checks for any news story, and how news agencies work explains why many outlets report the same poll in almost identical words.

Conclusion

A poll is a careful estimate with built-in uncertainty. Look at the sample size and margin of error, treat small leads and small changes with caution, prefer trends and averages to single results, and always ask who asked whom, when and how. Then poll headlines become what they should be: one useful piece of information, not a prediction.