The promises a survey can make about names
The survey and questionnaire design guide names the choice in a single clause of its section on consent and data handling: a survey says whether answers are anonymous, confidential, which it glosses as tied to the person yet never reported by name, or attributable. The same section notes that a survey which never asks for a name can still collect personal data, because a tracked invitation link or a customer ID carried into the survey is enough to tie answers to a person. This guide works that clause through.
The three promises sit on one line, from the least the business knows to the most it may share.
Anonymous. Nobody, at the business or elsewhere, can work out who gave an answer, from the survey or from anything joined to it. No identifier enters the survey, the link is the same for everyone, and no answer is detailed enough to point to a person.
Confidential. The business knows who answered, because an identifier travels with the response, but uses that knowledge only for analysis. Identities stay with the few people who need them, and nothing leaves the analysis with a name or a detail that would reveal one.
Attributable. The respondent agrees that what they said may be shared with their name, for a purpose they were told about, such as a follow-up from a colleague or a quote credited to them. Agreement is given per use and can be declined without leaving the survey.
Behind all three sits the professional standard the research methods section describes from the ESOMAR and MRS codes: taking part is a free and informed choice, people learn who is asking and for what, and what they say is kept confidential unless they agree to something else. Confidentiality is therefore the default the codes expect. Anonymous goes further than that default, and attributable steps back from it only with the respondent's agreement.
What each promise allows and rules out
Each promise opens some analyses and closes others, and the table sets them side by side.
| Row label | Anonymous | Confidential | Attributable |
|---|---|---|---|
| Answers set beside customer records | Not possible; nothing joins them | Possible, by the analysts, as disclosed in the invitation | Possible, for the uses agreed |
| Comparing respondents with the people invited | Only on what respondents say about themselves | On record fields such as tenure, plan or spend | On record fields |
| Removing duplicate answers from one person | Not reliably | By identifier | By identifier |
| Follow-up with a person about their answer | Not possible | Only with the person's separate agreement to be contacted | For the uses agreed |
| Reminders only to people who have not answered | Not possible; nobody knows who answered | Possible | Possible |
| Quotes in the report | By group, checked so nobody can be recognised | By group, checked the same way | Credited by name only where agreed |
| Individual answers passed to managers or account owners | Never | Never | Only where agreed, for that purpose |
Two rows deserve a note. Joining answers to records needs a Join KeyThe shared field, such as date and channel, a campaign ID or a click ID, on which rows from two data sources are matched in a blend or a query., the field both sources share, and in a survey that key is the identifier the invitation carries. The tutorial that tests survey scores against later churn shows how much that join is worth and where it stops: it limits itself to answers from surveys that told respondents about the link, and it is plain that answers promised anonymity cannot be joined to accounts. The follow-up row is narrower than it looks under every promise, because contacting someone about what they wrote is a use they need to have accepted.
Why a survey without names can still identify people
Four routes tie answers to people without any name field, and an anonymous promise has to close all of them.
Identifiers in the link or the record. A personalised invitation link, or a customer ID that travels into the survey, ties a response to a person, which is why the survey design guide treats either as making the answers personal data. A survey called anonymous whose link is tracked is not anonymous; it is confidential at best, and only if the business treats it that way.
What the survey records about each response. The personal data entry counts online and device identifiers as personal data, so check what is recorded about each response. The same entry separates identifiable, pseudonymous and anonymous data, and notes that data where names have been swapped for codes, with the key kept elsewhere, is still personal data. A confidential survey with a restricted key is in that position: protected, but not anonymous.
Free text. People write names, order numbers, the name of the colleague who helped them and details of their own situation into comment boxes. The tutorial on coding open-text answers removes those details before anyone codes the answers, and it warns that free text can identify someone once it is combined with other details.
Combinations of group fields. Region, plan, tenure and role identify nobody on their own. Together, in a small group, they can describe one person. The coding tutorial checks every quote so that its wording and group label together cannot single anyone out, and the same check applies to tables. It matters most where the population is small: the guide to researching business buyers shows how a job title paired with a tight account description can give one customer away where the business has few of that kind, and reports such answers by role and by a customer description wide enough to conceal the account.
An honest anonymous promise therefore needs more than a missing name field. It needs a shared link, no identifier carried in, a check of what the tool records, a warning in the comment box not to add personal details, and grouping that keeps every reported cell large enough to hide its members. Where a population is too small for that, a confidential promise kept strictly is more truthful than an anonymous one that cannot be kept.
Choosing the promise from the decision
The research design guide starts every study from the decision it serves, and its brief has an ethics and data field where the promise belongs. Choose the promise from what the decision needs to see, not from what feels safest or what the last survey did, and collect no more identity than that: the survey design guide's rule of asking only what the analysis needs, applied to names.
| What the decision needs | The promise that fits | What to check before choosing |
|---|---|---|
| Answers read beside behaviour, tenure or spend from the records | Confidential | That the invitation will say answers are linked and why |
| Respondents compared with the people invited on record fields | Confidential | That the comparison is planned, not a habit |
| Honest answers on a topic people would hold back on if known | Anonymous | That the population and the groups are large enough to hide people |
| A one-off read of opinion with no join and no follow-up | Anonymous or confidential | Which fields the analysis really uses |
| Follow-up with customers about what they said | Confidential, with a separate question asking who agrees to be contacted | Who follows up, under what written rule |
| Quotes or feedback credited to named people | Attributable, for those who agree | That agreement is asked per use and can be refused |
Two more questions narrow it. Is the population too small for anonymity to hold, as with a short list of business customers? Then a strictly kept confidential promise is the honest one. Would respondents expect a reply? An anonymous survey cannot give one, so offer another route beside it.
The direction of the choice matters too. The one-way and two-way doors framework groups decisions by how easily they can be undone. A promise of anonymity is a one-way door for the answers already given: once people have answered on that basis, no later need can justify tying their answers to them. A confidential survey keeps more doors open, because the business can decide later to report less, delete the key or never join the data, but it cannot move in the other direction for answers already collected. Whichever promise is chosen, the Lawful BasisOne of the six grounds the GDPR accepts for processing personal data, such as consent, contract or legitimate interests, chosen for each purpose. for any linking is decided for that purpose, before processing starts, by whoever owns data protection, and data gathered for one purpose is not simply reused for an incompatible one.
Saying it in the invitation
The invitation is where the promise is made, so it has to match what the survey actually does. If the link is tracked, the invitation cannot call the survey anonymous; if answers will be joined to records, it says so and why. Keep the promise short, plain and near the top, so respondents read it before the first question.
As general practice, each promise needs a few elements, and the final wording belongs to whoever owns data protection.
- Anonymous. That nobody at the business will know who gave which answer; a request not to put names or account details in comment boxes; and, because nobody can reply, a pointer to the right place for anything that needs a response.
- Confidential. That answers are linked to the customer record and why; who sees them in that form; that results are reported for groups and never by name; and that nothing said will be passed to a salesperson, account owner or local manager.
- Attributable. What will be shared, with whom and for what; that agreement is a separate choice for each use; and that answering the survey does not depend on agreeing.
Two separate items belong in the survey rather than inside the promise. The first is a route for complaints. A survey read for groups is a poor channel for a problem that needs handling, so say where complaints go instead. The second is an optional question asking whether the respondent is willing to be contacted about their answers. The voice of the customer guide closes the loop with individual customers only where they had said yes to contact about their feedback, and it is firm that anonymous surveys stay anonymous. A respondent who says no remains covered by the survey's main promise.
Finally, check the promise against the business's published Privacy NoticeThe published notice that tells people what personal data the business collects, why, and which tools and partners receive it; marketing keeps it current.. The notice describes what the business does with personal data but does not make a use acceptable on its own, and respondents are informed only when the invitation, the survey and the notice say the same thing.
Keeping the promise in the data and the report
A promise is kept or broken in the handling, long after the invitation has gone.
Separate identifiers from answers early. The survey design guide asks for this as soon as possible. In a confidential survey, give each response a code, keep the key that links codes to people where only the named analysts can reach it, and work from the coded file everywhere else. The tutorial on participant information and consent uses participant codes and a restricted key in the same way, together with a consent statement for each use and a deletion date set with the data protection owner. In an anonymous survey there is no key to keep.
Clean before anyone else reads. The cleaning and tabulation tutorial swaps names and contact details for the response ID in every place the analysis can do without them, and strips identifying details from open answers before they are shared. Under an anonymous promise, any identifying detail a respondent typed in has to go before the file leaves the person cleaning it.
Write the promise and the small-group rule into the plan. The survey analysis plan template asks the plan to record which of the three promises the survey makes and to set the count under which a group is shown as a number instead of a percentage, with the groups to merge or hold back. Write both before fieldwork, so nobody decides how to show a group after seeing who is in it. The guide to reading small numbers asks for the count behind every rate because small groups mislead; the same small cells can also reveal who is in them, so check every cut for a cell so small, or so particular, that a reader could name its members.
Quote by group. Attribute quotes by group, as the coding tutorial does, and read each with its label to be sure nobody who knows the customers could recognise the writer.
Pass on themes, not people. Teams who act on findings receive themes with counts and bases; under an anonymous or confidential promise, no individual answer goes to a manager, salesperson or account owner.
Say what was promised in the report. State the promise beside the method, so later readers know what the data may be used for and whether it can ever be joined to anything.
When the promise has to change
Requests to change the promise arrive after the fact, with good reasons: an analyst wants to join the last wave's anonymous answers to churn, a manager wants to know who wrote a critical comment, a product team wants to call the people who mentioned a fault. For answers already given, the answer to each is the same. The promise holds for everything collected under it. What can change is the next survey.
A change for the future is made openly. Decide it from the decision, as above, and put the new promise in the new invitation. For a repeated survey, the change touches the instrument itself. A Survey WaveOne round of fieldwork in a repeated survey or tracker, drawn, worded and coded the same way as earlier rounds so the results can be compared over time. repeats the earlier rounds' draw, wording and coding so its results can be compared, and the customer survey wave SOP keeps the questionnaire frozen or starts a new series. A different promise is a reason to start a new series, or at least to mark a break in the old one, because what respondents are told about their names can change who answers and what they are willing to say. Record the change with the draw, which the customer-list sampling tutorial writes down in full so a later wave can follow it.
Some requests are really about something else. A worrying comment in an anonymous survey cannot be traced, and the wave SOP keeps anonymous answers untraceable even when a comment is alarming; the remedy is the complaints route the invitation pointed to, and a confidential design next time if contact matters. A request to follow up respondents for marketing sits outside every promise here: the survey design guide does not allow research answers to feed marketing follow-up unless respondents separately agreed. When a request does not fit, say no, explain which promise it would break, and offer what the promise allows, such as a theme with its count.
A gym chain chooses how its member survey treats names (illustrative)
Grixomba Gyms chooses a confidential member survey
Grixomba Gyms, an invented chain of 9 gyms, plans a member survey about changing rooms and class times before deciding which clubs get new changing rooms first and whether to move evening classes. The decision needs answers read by membership type and by how often members visit, both of which the business holds in its records, so an anonymous survey would leave the analysis without the two fields it rests on.
The survey owner chooses a confidential survey. Each invitation carries a member ID, the key that links IDs to names stays with one analyst, and everyone else works from response codes. The invitation says that answers are linked to the membership record so they can be read by membership type and visit pattern, that results are reported only for groups and never by name, and that nobody at any club will see who said what. It adds that complaints about a club go through the separate complaints form, because survey answers are read in groups; the survey ends with an optional question asking whether the member is willing to be contacted about their answers.
Of 4,200 members invited, 731 answer, and 96 agree to be contacted. The plan, written before launch, sets a count below which a group is shown as counts only and a lower one below which it is merged into a wider group or, where none fits, held back. Off-peak members at the smallest club give 7 answers, under the lower count, so they are merged into an off-peak column across all clubs. Open answers are cleaned of names before coding, and quotes are labelled by membership type and club size only.
After the readout, a club manager asks to see who wrote a critical comment about evening classes. The survey owner refuses, explains that the invitation promised no one at a club would see who said what, and sends the theme instead: 41 members at that club raised class times. The customer service lead contacts only those of the 96 whose answers met the written follow-up rule.
Common mistakes
- Calling a tracked survey anonymous. A personal link or a customer ID makes the answers traceable, whatever the invitation says.
- Choosing the promise by habit. Anonymous by default can strip out the fields the decision needs; confidential by default can deter honest answers on a sensitive topic.
- Joining anonymous answers to records later. The promise covers the answers already given, whatever the analysis would gain.
- Forgetting the comment box. Names and account details typed into free text undo an anonymous promise unless they are removed before anyone else reads them.
- Small cells in the tables. A cut that leaves a handful of people in a group can tell readers who they are.
- Passing individual answers to managers. A confidential survey shares themes, never the person behind a comment.
- Changing the promise mid-series without a mark. A wave asked under a different promise is not the same measurement as the waves before it.
- An invitation that disagrees with the privacy notice. Respondents cannot be informed when the documents contradict each other.
