Margin of error and sample size: a practical guide
Published November 16, 2026 · 6 minute read
The margin of error tells readers how far a survey result might be from the true figure because only a sample was asked. It is the single most useful number to put next to a percentage, and the most commonly misread.
The formula
For a percentage p from a simple random sample of n people, the margin of error at 95% confidence is 1.96 × √(p(1−p)/n). It is largest when p is 50%, so researchers usually quote that worst case.
If your sample is a large share of a small population (say more than 5%), multiply by the finite population correction √((N−n)/(N−1)), where N is the population size.
Quick reference (p = 50%, 95% confidence)
So you need about 385 completed responses for ±5 points and about 1,068 for ±3 points, whatever the size of a large population.
| Responses (n) | Margin of error |
|---|---|
| 200 | ±6.9 points |
| 385 | ±5.0 points |
| 600 | ±4.0 points |
| 1,068 | ±3.0 points |
| 2,000 | ±2.2 points |
Subgroups need their own sample size
The margin of error applies to the group you report on. A survey of 1,000 with 100 people aged 18 to 24 has about ±9.8 points for that group. Plan the sample around the smallest group you need to report.
Weighting reduces precision
Weighting corrects bias but makes results less precise. The design effect is approximately 1 + CV², where CV is the coefficient of variation of the weights; the effective sample size is n divided by the design effect. Report the margin of error using the effective sample size.
Is a difference real?
- Two results whose margins of error overlap are not necessarily the same, and two that do not overlap are not necessarily different: use a significance test (a chi-square test for a whole question, or a two-proportion z-test for one answer).
- Changes between waves of a tracking study should be tested the same way before you report a “rise” or “fall”.
What the margin of error does not cover
It only measures sampling error. It says nothing about bias from who chose to respond, question wording or mode effects. A large, biased sample can be precisely wrong; weighting and good fieldwork matter as much as size.