A community survey begins before anyone opens a form. It begins when a team decides what it needs to know, what it already assumes, who could be harmed by the answers, and who will be missing if the survey is only shared through the easiest channels.
This guide is for organizations designing a survey in a German or EU civil-society context. It is general research guidance, not legal advice about a specific data-protection question. GDPR matters here at a general level because surveys can collect personal and sensitive data. The deeper point is simpler: people are more likely to trust a survey when they can see why it exists, what will happen to their information, and where the limits are.
1. Start with the decision the survey should inform
Do not begin with a question list. Begin with the decision.
Write one sentence: “We need evidence to decide whether…” The sentence might end with “our advice service should change its opening hours,” “the city should hear about barriers in a local application process,” or “our members experience language access differently across offices.”
Then separate:
- What you already know from daily work.
- What you suspect but cannot yet show.
- What a survey can realistically find out.
A good community survey is not a general information harvest. It is a focused research instrument tied to a clear purpose.
If the real need is to document discrimination incidents, a survey may be the wrong tool. Equal Voices Initiative covers incident documentation more directly. If the real need is to choose campaign tactics, Civic Futures Lab is the better lane. A survey can inform those processes, but it should not pretend to replace them.
2. Define who the survey is about, not only who you can reach
“The community” is not a sample frame. It is a relationship, a history, and often a contested boundary.
Name the population as precisely as you can: tenants in one district who contacted a housing advice service; parents whose children attend three local schools; volunteers in migrant-led organizations across one region; people using a particular public service in the last year.
Then ask how you will reach people beyond your usual circle. A survey shared only through an organization’s newsletter mostly measures the people already connected to that organization. A QR code at one event measures who came to that event. A German-only online survey excludes people who do not read German comfortably, people with limited internet access, and people who distrust digital forms.
Sampling is not only a technical issue. It is an inclusion issue. If you cannot reach everyone, say so in the analysis plan and later in the report. “This survey reflects people connected to our network” is weaker than “residents of the city,” but it is much more honest.
3. Write questions that do not push the answer
Biased survey questions distort results by smuggling a conclusion into the wording. They can make a problem look larger, smaller, simpler, or more agreed upon than it is.
Avoid loaded language:
- Weak: “How often has the office ignored your needs?”
- Better: “In your most recent contact with the office, were your main questions answered?”
Avoid double questions:
- Weak: “Was the service respectful and easy to access?”
- Better: ask respect and access separately.
Avoid answer options that trap people:
- Weak: “Why did you not complain?” with only “I did not know how” and “I was afraid.”
- Better: include options such as “I did complain,” “I did not think it would help,” “I did not have time,” “I was worried about consequences,” “another reason,” and “prefer not to say.”
Avoid assumptions:
- Weak: “As a migrant, which barriers did you face?”
- Better: first ask whether the person wants to describe migration-related barriers at all.
What is known: question wording affects answers. This is not a minor editing issue. Small wording choices can change what people remember, what they feel invited to disclose, and whether they feel judged.
What is contested: there is rarely one perfect wording. Community reviewers may prefer direct language because it names harm plainly. Researchers may prefer more neutral language to reduce measurement bias. Both concerns are legitimate.
What we cannot say: a survey cannot prove motive just because respondents describe a pattern. It can show reported experiences, frequencies within the reached group, and associations that deserve attention.
4. Make accessibility part of the research design
Accessibility is not a final polish. It changes who can answer.
Plan for plain language. Use short sentences, define terms, and avoid administrative vocabulary unless it is necessary. If you use German terms such as Bescheid, Jobcenter, Aufenthaltstitel, or Widerspruch, explain them in English or the survey language.
Offer language options where the population needs them. Translation should be checked by someone who understands both language and context. Literal translation can make a consent statement technically correct but socially confusing.
Think about format. Some people need screen-reader compatible forms. Some need paper versions. Some need the option to complete a survey with support from a trusted person. Some will not answer sensitive questions on a shared family device. If you offer assistance, make sure the helper does not pressure the participant or see answers that should remain private.
Accessibility also includes time. A 25-minute survey may be acceptable for a paid research panel. It may be unreasonable for a parent between shifts, a person translating official letters, or a volunteer already asked for help every week.
5. Explain consent before the first sensitive question
Participants should know what they are entering before they give data. A consent note should be short enough to read and clear enough to matter.
At minimum, tell participants:
- who is collecting the data;
- why the survey is being conducted;
- what topics will be asked about;
- whether any questions are optional;
- whether answers are anonymous, confidential, or neither;
- how long data will be kept in general terms;
- who will see raw responses;
- how results may be shared;
- whether participation affects access to services;
- how to ask a question or withdraw where withdrawal is possible.
Do not use “anonymous” if you collect names, email addresses, rare demographic combinations, detailed locations, or free-text stories that can identify someone. “We will not publish your name” is not the same as anonymity.
This answers “What should participants know before giving data?” They should know the purpose, the risks, the choices, and the consequences of non-participation. They should also know if the survey cannot help them individually. A research survey is not an emergency channel, legal representation, or a promise of service.
6. Collect less sensitive data than you are tempted to collect
Every demographic question should earn its place. Age group, gender, disability, income, migration history, residence status, religion, ethnicity, family situation, and health information can all be relevant. They can also create risk.
Ask three checks before including a sensitive item:
- Will this variable be used in analysis?
- Can the question be asked in a less identifying way?
- Could the answer expose someone if the dataset, table, or small subgroup were shared?
For example, asking for exact nationality, small neighborhood, age, gender identity, and a rare service use can make a person recognizable even without a name. This is especially serious in small towns, small organizations, shelters, schools, or tightly connected migrant communities.
Digital Dignity Lab is the better hand-off for detailed data safety practices. In survey design, the research principle is clear: do not collect data simply because it might be interesting later.
7. Reduce respondent burden
Respondent burden is the work you ask people to do: remembering, interpreting, translating, typing, deciding what is safe to disclose, and sometimes reliving difficult events.
Keep the survey as short as the research question allows. Put the most important questions early. Mark optional questions clearly. Use skip logic carefully so people are not forced through irrelevant sections.
Avoid asking for long trauma narratives in a general survey. If narrative material is essential, explain why, make the question optional, and provide a less exposing alternative such as categories with an “add detail if you want” box. MindForward Collective is the appropriate sibling resource for distress and aftercare questions when research touches painful experiences.
Payment or small compensation can be appropriate, especially when communities are often asked to provide expertise for free. Be transparent about it. Compensation should recognize time without becoming pressure to disclose more than someone wants.
8. Pre-test with people who can disagree safely
Before launch, test the survey with a small group that includes people similar to the intended respondents and people who are willing to challenge the wording.
Ask testers:
- Which question felt unclear?
- Which answer option was missing?
- Which question felt intrusive?
- Where did you wonder who would see the answer?
- Did the introduction make the purpose clear?
- How long did it actually take?
Do not treat pre-testing as approval theatre. If testers point out that a category is stigmatizing or a consent note is unclear, revise. If the survey is translated, test each language version. A form can be easy in German and awkward in Arabic, Turkish, Russian, Ukrainian, English, or another community language.
9. Write the analysis plan before looking at responses
An analysis plan does not need to be academic. It can be a one-page note.
State the main questions, the key comparisons, the subgroups you will and will not report, and how you will handle open-text answers. Decide in advance what minimum group size is too small to show in a table. Decide how you will report uncertainty. Decide what claims you will not make.
This protects against cherry-picking. Without a plan, teams are tempted to search for the most dramatic result and build the story around it. A striking quote or percentage may still be important, but it should sit inside the design limits.
10. Report results with limits attached
A trustworthy survey report says what was found and how it was found.
Include the field period, languages, recruitment channels, number of respondents, eligibility rules, and major limitations. If the survey used convenience sampling, say so plainly. If certain groups were underrepresented, say that too. If open-text responses were coded by staff, explain the broad coding process.
Limitations do not weaken the work. They make the work usable. A city official, journalist, funder, or community member can only judge a finding if they know what kind of evidence it is.
The strongest sentence is sometimes not “this proves.” It is “among the people we reached, this pattern was common enough and consistent enough to require attention.” That sentence can still matter.