Three options, three different meanings
A respondent may leave a question unanswered because they lack information, because the question does not fit their situation, or because they would rather not disclose the answer. “Don’t know,” “Not applicable,” and “Prefer not to say” give people ways to express those different reasons. Treating them as the same kind of missing answer can hide a problem with the question or lead to a misleading summary of the results.
“Don’t know” means the question applies, but the respondent cannot give a reliable answer. “Not applicable” means the question’s premise does not fit their experience. “Prefer not to say” means they are choosing not to share. Before adding any of these options, ask what a respondent would need to know, experience, or reveal to answer the question. The answer will help you decide which alternatives belong on the list.
Use “Don’t know” when an informed answer may be unavailable
“Don’t know” is useful when people may reasonably lack the requested information. Someone might be able to rate a product they use but not know which subscription tier their household pays for. Without an uncertainty option, they may guess, select a neutral-sounding answer, or abandon the question. If an estimate would serve your purpose, say so in the question rather than assuming respondents will provide one.
The option is less helpful when the question asks for a straightforward personal preference, such as which of two features someone would choose. In that case, “No preference” may describe the response more accurately. Avoid using “Don’t know” as a catch-all for unclear wording: if many people cannot tell what a term means, define it or rewrite the question. A high share of “Don’t know” responses can itself be useful feedback about what the audience understands.
Use “Not applicable” when the situation does not fit
Offer “Not applicable” when some respondents have no relevant experience or circumstances on which to base an answer. For example, a question asking people to rate a recent customer-support interaction does not apply to someone who has never contacted support. Asking that person to choose a satisfaction rating would produce a number, but not a meaningful measure of satisfaction.
Where possible, use a screening question or survey routing to avoid showing irrelevant questions at all. Ask whether the respondent has contacted support, then show the rating question only to those who have. If routing is not available, a clearly labeled “I have not contacted support” option may be more informative than a generic “Not applicable.” Do not confuse inapplicability with a low or zero value: “I spent nothing” and “This question does not apply to me” may represent different experiences.
Use “Prefer not to say” to allow a deliberate refusal
“Prefer not to say” gives respondents a clear way to decline a question that may feel personal or sensitive. It can be appropriate for questions about income, household circumstances, identity, or other information people may not want to disclose. An unanswered field does not tell you whether someone skipped it by accident or chose not to answer; an explicit option makes that choice clearer.
This option is not a substitute for explaining why you are asking or limiting questions to information you genuinely need. Consider whether broad ranges would meet your research goal better than a precise answer. Also consider making the question optional. Even a familiar topic can feel sensitive in a particular context, so review the survey from the respondent’s point of view rather than relying on a fixed list of sensitive subjects.
Choose deliberately, then analyze each response separately
You do not need to add all three options to every question. A support-rating question might need “Not applicable,” while a question about awareness of a competing product might need “Don’t know.” A personal-information question might need “Prefer not to say.” Some questions may reasonably need more than one, provided the labels describe distinct choices. Place them where respondents can find them, and avoid wording that suggests one answer is more acceptable than another.
In a pilot test, ask a few people what each option means to them and watch for cases where none fits. After collecting responses, keep the three categories distinct from one another and from blanks caused by skipped or incomplete surveys. Check both the number of usable answers and the pattern of alternative responses before interpreting a result. Many “Not applicable” answers may point to a routing issue; many “Don’t know” answers may signal an unfamiliar topic or unclear wording. Those patterns can improve the next survey as well as the current analysis.
