How to run a representative resident survey
Published November 16, 2026 · 9 minute read
A resident survey is only useful if its results describe the whole community, not just the people who like answering surveys. This guide walks through the decisions that make the difference, from who you invite to how you report the numbers.
1. Decide what the results must support
Start with the decisions the survey will inform: next year’s budget priorities, a service review, a performance indicator you report every year. Each decision tells you which groups you must be able to compare (for example each ward, age group or tenure type) and how precise the results need to be.
If you will compare wards, you need enough responses in each ward, not just in total. That single requirement usually drives the size of the survey more than anything else.
2. Choose a sampling approach
- Probability sample: invite a random selection of addresses or residents (for example from an address register) by post, with an online link and a phone or paper option. This is the most defensible approach for headline indicators.
- Standing citizen panel: recruit residents once, keep their profile (ward, age, gender) and invite balanced samples for each survey. Faster and cheaper per survey, and well suited to frequent consultation.
- Open link: publish the survey for anyone to answer. Good for engagement and reaching interested residents, but not representative on its own; report it as “what participants said”, not “what residents think”.
3. Work out how many responses you need
For a percentage, the margin of error at 95% confidence is about 1.96 × √(p(1−p)/n). With 400 responses it is about ±4.9 points; with 1,000 about ±3.1; with 2,500 about ±2.0. If you need ±5 points in each of 10 wards, plan for roughly 400 completed responses per ward.
Allow for response rates: if you expect 25% of invited residents to respond, invite four times the number of completes you need, and more in areas where response is usually lower.
| Completed responses | Margin of error (95%) |
|---|---|
| 100 | ±9.8 points |
| 400 | ±4.9 points |
| 1,000 | ±3.1 points |
| 2,500 | ±2.0 points |
4. Write questions that measure what you think they measure
- Ask one thing per question (“How satisfied are you with snow clearing on your street?”, not “…with snow clearing and road repair?”).
- Use balanced scales with labelled end points, and keep the same scale across the survey and across years.
- Avoid leading wording and loaded examples; put sensitive questions after easier ones.
- Keep it short: completion falls noticeably after about 10 to 15 minutes.
- Keep tracking questions word for word identical each year, so change over time is real change.
5. Make it accessible and multilingual
Public-sector surveys should meet WCAG 2.2 AA: every question usable by keyboard and screen reader, clear error messages, no time limits, and layouts that work on a phone. Offer the languages your community speaks, and an alternative (phone or paper) for residents who cannot answer online.
6. Run fieldwork with quotas and reminders
- Send reminders a few days apart to people who have not responded; most surveys gain a third or more of their responses from reminders.
- Watch responses by ward and age group as they arrive, and target extra invitations at groups that are behind.
- Use quotas to stop over-represented groups from dominating once they reach their target.
7. Weight the results to the population
Even a well-run survey over-represents some groups. Weighting adjusts each response so the sample matches known population figures (for example census counts by age, gender and ward). Raking adjusts to several of these at once.
Report the effective sample size after weighting: heavy weights reduce precision, and the margin of error should reflect that.
8. Report clearly and protect privacy
- Lead with the findings decision makers need, then the detail.
- Show margins of error, and avoid reading meaning into differences smaller than them.
- Hide results for very small groups (commonly fewer than 5 or 10 responses) so individuals cannot be identified.
- Publish a short “how the survey was done” note: who was invited, when, how, and how results were weighted.