Read beyond the noise.

A Survey Margin of Error Does Not Cover Every Kind of Error

Survey questionnaire and a small group of anonymised tally cards.
AI-generated photographic illustration; not documentary photography.

A poll reports support at 52 percent with a margin of error of three percentage points. The neat range can make the whole survey feel precisely measured. It describes one kind of uncertainty, not every possible problem in the result.

Pew Research Center’s explanation distinguishes sampling uncertainty from issues such as question wording and differences in survey procedures. A reported margin does not magically correct a biased question or an unrepresentative response pattern.

Find the methodology before calculating

Look for who was eligible, how people were selected, how the survey was conducted and when responses were collected. Check the sample size and whether weighting or design effects are included in the reported margin.

A conventional sampling margin should not be casually attached to an open online vote where anyone chooses whether to participate. Different survey designs need different explanations of uncertainty. If the report does not provide one, describe that limitation rather than inventing a range.

A subgroup has its own uncertainty

In an invented national poll, only a small portion of respondents may belong to a particular age group. The overall sample’s margin is not automatically the right margin for that smaller group. Read the subgroup base and the report’s guidance before comparing two narrow slices.

Likewise, the uncertainty around a difference between two candidates or two surveys is not obtained simply by looking at one percentage in isolation. Avoid declaring a clear lead from a tiny gap without the appropriate analysis.

Read what was asked

Locate the exact question and response options. “Support this proposal” and “make this proposal the highest priority” ask different things. A chart label may compress that distinction until it disappears.

A useful reading note keeps the estimate, uncertainty statement, population, dates and wording together. That does not require dismissing the poll. It helps you report what the survey actually supports, rather than giving one familiar statistical phrase more authority than it can carry.