What is social desirability bias?

Mode matters more than wording

The strongest single factor is whether a human being is present. Self administered surveys, whether online or on paper, produce more candid answers on sensitive subjects than interviews conducted by phone or face to face, and the gap is large and well documented. That has a practical consequence for anybody choosing a method: if the subject is sensitive, a form somebody fills in alone will beat a conversation, even though the conversation feels more thorough. Within online surveys the same logic applies to perceived observation. A workplace survey that collects a name, or that collects enough demographic detail to identify somebody, will read as observed no matter what the introduction promises.

Wording that makes every answer acceptable

The technique is to write the question so that the less flattering answer sounds ordinary. Assume the behaviour rather than asking whether it happened: how many times in the last month, rather than have you ever. Offer a permissive preamble, such as noting that many people find it difficult to keep up with something, before asking whether the respondent did. Use broad frequency bands rather than exact counts, since a band feels less like a confession. Avoid words that carry judgement, including failed, neglected and should. None of this is manipulation; it is removing an artificial reason for somebody to misreport.

Anonymity has to be real and visible

Promising anonymity and then collecting a team, a role and a tenure band is not anonymity, and respondents work that out quickly. If you say a survey is anonymous, make it so: no identifier, no email capture, no demographic combination that isolates an individual, and a stated rule that no group below a threshold will be reported separately. Say all of that on the form rather than in a separate policy, because a classic field's description is not shown to the respondent and an assurance buried elsewhere does not do the work. Where follow up is needed, offer an optional contact field and be explicit that providing it ends the anonymity.

Indirect techniques, and when they are worth it

Research methods exist for particularly sensitive measurement: asking what people like the respondent would do rather than what they do, list experiments where the sensitive item is hidden in a count, and randomised response techniques that add deliberate noise to protect individuals. They work, and they cost sample size, analysis complexity and explanation. For most business surveys they are more machinery than the question warrants, and the same gain is available from a self administered mode, genuine anonymity and careful wording. Keep them in mind for research where the estimate has to withstand scrutiny and the topic is one people will not answer straight.

Building a candid form

Access settings decide how anonymous a form actually is, and a public link with no login is the honest choice when anonymity has been promised. Keep sensitive questions optional and put them late. Use radio fields with a prefer not to say option rather than forcing a choice, remembering that dropdown, radio and checkbox fields have no working other option, so a free text alternative is a separate short text field. A conversational form renders a field's description under the question, which is useful for the reassurance that precedes a sensitive item; on a classic form that reassurance has to be a paragraph block above the field.

Asking about behaviour rather than intention

Intentions are the most flattering thing a person can report, which makes them the least reliable. Would you recommend, would you attend, would you pay more for a sustainable option all collect a version of the respondent they would like to be. Asking what actually happened narrows the gap considerably. Did you recommend us to anybody in the last three months is answerable from memory and is far harder to embellish than a hypothetical. The same applies to frequency questions: a specific recent window such as the last seven days produces better data than a question about what somebody usually does, because usually invites an estimate assembled from the person's self image rather than from events. Where an intention question is genuinely needed, pair it with a behavioural one and read the two together rather than reporting the intention on its own.

Questions people ask