Skip to main content

Survey Data Handling for Journalists

How to read, interrogate, and report survey results responsibly — from YouGov polling to academic surveys.

Last reviewed: Next review due:

Why survey reporting goes wrong

Survey data is among the most frequently misreported data in journalism. The most common errors are: treating a sample finding as if it applies to the whole population without mentioning uncertainty; reporting sub-group findings from too small a sample; ignoring weighting; and not questioning whether the question wording could have influenced the result.

Before reporting any survey finding, ask: who was asked, how many, how were they recruited, how were the results weighted, and who commissioned the poll?

Key concepts

Margin of error

The margin of error tells you the range within which the true population value probably lies. At 95% confidence with a sample of 1,000:

Margin of error ≈ ±3 percentage points

If poll shows 42% support for Party A with ±3pp:
True support is probably between 39% and 45%.

A lead of less than the margin of error (e.g. 43% vs 41%)
is "within the margin of error" — too close to call.

Sample size and sub-groups

Sub-group analysis has a much larger margin of error than the headline figure:

Total sample: 2,000 respondents  → ±2pp
Scotland sub-group (8.5%): ~170 → ±7.5pp
Under-25 sub-group (15%):  ~300 → ±5.6pp

Rule: sub-groups under ~200 respondents should not
be reported as representative findings.

Question wording bias

Leading questions, the order of response options, and framing can all influence results. Compare these two questions:

  • “Should the UK government keep the triple lock on pensions?” — likely higher support
  • “Should pensioners continue to receive guaranteed increases even when working-age benefits are frozen?” — likely lower support

Both are technically the same policy. Always check and include the exact question wording.

UK polling firm methodology notes

YouGov

Online panel; quota sampling; MRP (Multilevel Regression and Post-stratification) for constituency-level estimates.

Ipsos

Mixed mode (online + telephone); random probability sampling for some products; quota-based for others.

Survation

Online panel; telephone polling for some political surveys; weighted to ONS population estimates.

When survey data requires extra care

  • 1A press release leads with a sub-group finding (e.g. "60% of Northerners support X") without mentioning the sub-group sample size.
  • 2The poll was commissioned by an organisation with a financial or political interest in the result.
  • 3The question wording is not disclosed alongside the finding.
  • 4The margin of error would make the headline finding statistically unreliable.
  • 5The survey used self-selected respondents (e.g. a website clickpoll) rather than a properly sampled panel.

Red flags

  • No sample size disclosed.
  • A "click poll" or "opt-in" survey — self-selected respondents are not representative.
  • The commissioner of the poll stands to benefit from the result.
  • Sub-group findings with no mention of sub-group sample size.
  • Results presented as exact percentages with no mention of uncertainty.

Survey reporting checklist

  • I have the full tables, not just the press release headline.
  • I have checked the sample size and calculated or noted the margin of error.
  • I have included the exact question wording in my article or methodology note.
  • I have noted who commissioned the poll.
  • I have checked that any sub-group finding is based on a sufficient sub-group sample.
  • I have not reported a result as definitive when it falls within the margin of error.
  • I have checked the polling firm's membership of the British Polling Council (BPC), which requires methodology disclosure.

Need data that surveys cannot provide?

Administrative data from ONS, NHS England, or local authorities often provides more reliable evidence than survey data for many story types. Use our data sources guide to find it.

UK Data SourcesStatistics for Journalists

Common mistakes

  • Treating an online opt-in poll as a representative survey.
  • Reporting sub-group findings without mentioning the sub-group sample size.
  • Not mentioning who commissioned the poll.
  • Using the exact percentage without a margin of error caveat.
  • Assuming all polling firms use the same methodology — they do not, and differences matter.

Related guides

Statistics for Journalists
Means, percentages, and margins of error explained.
Regression & Correlation
Going beyond survey results to model relationships.
UK Statistical Bodies
Official alternatives to survey data for many story types.
Data Storytelling
How to communicate uncertainty to a general audience.

Primary sources

Frequently asked questions

What sample size do I need for a nationally representative UK poll?
For a nationally representative poll of UK adults, a minimum of 1,000 respondents is conventional. This gives a margin of error of roughly ±3 percentage points at 95% confidence. Polls of 2,000+ respondents are more reliable. Smaller samples (e.g. 500) have margins of error around ±4.5pp and should be reported with that caveat. Sub-group analysis (e.g. only responses from Scotland, or only responses from under-25s) requires much larger samples or the margin of error becomes too wide for reliable reporting.
What does weighting mean in a survey?
Weighting adjusts survey responses so the sample matches the known characteristics of the population. For example, if the survey over-sampled graduates (because they are more likely to respond), the firm will down-weight graduate responses so the final result reflects the UK population's actual education profile. YouGov, Ipsos, and Survation all publish their weighting methodology in their survey tables. When results are unweighted, they may not be representative even if the sample size is large.
How do I get access to the full tables for a YouGov poll?
YouGov publishes full cross-tabulation tables (usually as Excel or PDF) for most of its publicly commissioned polls. Search YouGov.co.uk > Data, or look for a link in the published article. Full tables let you check: the unweighted sample size; how respondents of different demographics answered; and whether sub-group differences are large enough to report. Always ask the polling firm or commissioner for full tables before reporting sub-group findings.

Related guides

Data JournalismStatistics for Journalists — Margins of ErrorRead →Data JournalismData Cleaning TechniquesRead →FOIFiling FOI Requests for Survey MicrodataRead →Media LawData Protection Law and Survey RespondentsRead →InvestigativeInvestigating Polling Data and Commissioned SurveysRead →

Primary sources