Should a survey be anonymous? When it helps and when it hurts

The two words get mixed up. Penn State's assessment office defines anonymity as "the state in which the identity of individual responses is completely unknown and cannot be linked to their survey responses" (Penn State). Confidentiality means the researchers "have access to the participant's identity and individual responses, but no personally identifying information is shared or reported" in a way that could expose the person.

The University of Colorado Denver adds a warning: "Even if you are not collecting unique identifiers, your survey may not be anonymous if you are embedding or collecting demographic data" (CU Denver). A survey that asks for team, role and start year can point at one person even without a name.

Qualtrics, which sells survey software and tends to recommend confidential surveys, still lists the upsides of anonymous ones. Respondents "may be more comfortable saying what they really think", feedback on sensitive topics "will be more honest", and there is less social desirability bias, where people answer to look good (Qualtrics).

Bright Horizons, a childcare and employer services company, says anonymous surveys can generate more responses because people feel more comfortable. It also warns that fear of negative repercussions can lead to a survey pool that is "both less honest and less likely to respond" (Bright Horizons).

Both pages are vendor guidance, not experiments. Treat them as practitioners' advice.

You lose the ability to ask a follow-up. Qualtrics lists the costs of anonymity: "You can't contextualize negative comments or situations", and you can't link answers to other data (Qualtrics). If one person writes "the Tuesday shift is a mess", you cannot ask what happened.

You also lose comparisons. With no team or tenure attached, you cannot count how the answers differ between groups.

Qualtrics sums it up this way: for a rapid one-off poll, "an anonymous survey is sometimes all you need". If you plan to take detailed action on the feedback, it says a confidential survey is the better bet. Bright Horizons puts it as "the former can generate more responses ... the latter can be more useful."

A rule of thumb that fits those sources:

• Sensitive topic, people may fear blowback: anonymous.
• Small group where answers could be traced: anonymous, or no demographic questions.
• One-off poll, nothing to follow up: anonymous.
• You want to reply to individuals: confidential.
• You need to compare groups: confidential, with a minimum group size.

This list is our reading of the sources above, not a finding from a study.

1. Don't ask for a name or email. CU Denver's list of identifiers includes name, email address, student or employee ID and phone number.
2. Skip demographic questions you won't use. Each one narrows the pool of who could have answered.
3. Say what you collect and who sees it. Bright Horizons advises being clear about confidentiality and who will have access to the data, and explaining what the answers will be used for.
4. Don't promise more than you can keep. If your tool records anything that could identify people, say so rather than writing "anonymous".

If you do want to reply to people, ask for an email in a separate, optional last question, and say it is only for follow-up.

The open box is the weak spot. Penn State notes that open-ended responses "sometimes refer to individuals by name or position or include details that identify the respondent." Its suggestion is to plan for this, for example with a "sanitized summary" or by cleaning identifying content out of the answers (Penn State).

CU Denver recommends adding a line to open-response items. Its wording says the response may be seen by multiple people and that if you share identifiable information, such as names, anonymity or confidentiality can't be guaranteed (CU Denver).

Small groups are the other trap. Penn State's example is reporting on international faculty in a program with only one or two of them, where "it can become obvious who responded". It suggests a minimum number of responses before you report a group. CU Denver says it commonly uses 5 or 10 as that minimum.

A worked example

The numbers below are invented to show the method. They are not from a real survey.

A 30-person company runs an anonymous team survey with 24 replies and one open question. The owner wants to split the answers by department.

With a minimum of 5, only Sales and Support get their own counts. Finance and Design together add up to 4, still under 5, so the owner leaves them out of the split and reports them only in the overall total of 24 (worked out as 11 + 9 + 3 + 1).

There are a few honest ways to run an anonymous survey.

• By hand. Ask people to drop written answers in a shared box or an email alias set up for the purpose, then read them yourself. You control everything, but you can't promise anonymity if you can see who sent what.
• A general AI chat such as ChatGPT. Paste the answers in and ask for groups. It's quick, but you would be pasting people's words into a third-party tool, so check what you've told respondents. A summary can also merge two problems into one.
• A survey tool. Google Forms, Typeform and Tally can all collect answers. The Google Forms comparison shows where Google Forms and Enform differ.

Enform writes the survey from a description of what you want to ask, you share one link, and it reads every reply and counts what people said. Every count opens to the replies behind it, so you can check a group against the sentences. It is a better fit when written answers pile up. For a handful, reading them yourself is fine. See pricing, the pulse survey template, or the employee engagement survey template. Our post on reading counts at small sample sizes helps when few people reply.