Vignette Questions in Research: How to Write and Use Them

Short answer: a vignette question describes a short, concrete situation involving a hypothetical person or account, then asks the participant what that person should do, would do, or how they judge the outcome. Use vignettes when the thing you want to learn is a judgement or a decision rule, especially one that is sensitive, rare, or hard for people to describe in the abstract. Keep each vignette short and plausible, change one or two details at a time between versions, show each person only a handful, and always ask why they answered as they did.

Most interview advice tells you to ask about real past behaviour, and that advice is right. But some questions cannot be answered from memory. A participant may never have faced the exact decision you care about. The topic may be awkward to admit about oneself. Or the decision depends on a combination of factors that nobody can list on request. Vignettes fill that gap. They give every participant the same concrete case to react to, so you can compare answers, and they move the spotlight off the participant, so people say what they actually think.

What a vignette is

The sociologist Janet Finch, writing in Sociology in 1987, described vignettes as "short stories about hypothetical characters in specified circumstances, to whose situation the interviewee is invited to respond." That definition still holds. A vignette has three parts:

A character. A person, team or company the participant is not.
A situation. Specific details: what happened, when, what it cost, who was involved.
A prompt. What should they do? What would you advise? How fair is this? How likely is it that they switch?

Here is a simple product example:

Priya runs a five-person design agency. Her team's project tool raised its price by 30% at renewal, with two weeks' notice. Switching would take about a week of setup, and two of her clients have shared boards in the current tool. What do you think Priya should do, and why?

Compare that with the direct version: "How would you react to a 30% price increase?" The direct question asks people to predict their own behaviour in the abstract, which is exactly where hypothetical questions go wrong. The vignette supplies the details that real decisions depend on, and it asks about Priya, not about the participant.

When vignettes are the right tool

Vignettes earn their place in four situations.

Sensitive topics. Asking someone whether they have ever shared a login, ignored a security policy or bent an expense rule invites social desirability bias. Asking whether a fictional colleague in a described situation was wrong to do it is less threatening. People reveal their norms through their judgement of others.

Rare or future decisions. If you are designing for an event most people have not faced yet (a data breach notice, a contract renegotiation, a first hire), there is no past behaviour to ask about. A vignette lets you study the reasoning before the event.

Decision rules with several inputs. Buyers weigh price, switching cost, risk and who else is affected at the same time. Ask them to rank those factors and you get a tidy list that does not match what they do. Vary the factors inside a vignette and you can see which ones actually move the answer.

Comparability across participants. Open questions about past experiences produce stories that differ in every detail. When everyone responds to the same case, differences in answers reflect differences in people, not in the situations they happened to describe.

When not to use them

Vignettes measure judgement about a described situation, not behaviour. They are a poor choice when you can observe or ask about the real thing directly, when the decision depends on emotion or context a paragraph cannot carry, or when the participant has no relevant experience to judge from. If your team needs to know what customers actually did last quarter, ask about last quarter. If you need to know how people weigh trade-offs in a situation they have not met yet, a vignette is often the best evidence you can get before launch.

Do vignette answers predict real behaviour?

This is the right question to ask of any stated-preference method, and there is direct evidence on it. Jens Hainmueller, Dominik Hangartner and Teppei Yamamoto, in a 2015 paper in PNAS, compared vignette and conjoint survey designs against a real-world benchmark: Swiss municipalities that decided citizenship applications by referendum, where voters received official descriptions of each applicant. The survey estimates of how applicant traits affected support matched the referendum results closely. The best design, a paired conjoint in which respondents compared two profiles side by side, came within about 2 percentage points of the behavioural benchmark on average. The authors also found that design choices mattered: some formats tracked the benchmark noticeably better than others.

Two practical lessons follow. First, well-designed vignettes can recover the relative weight people give to different factors. Second, presenting two cases side by side and asking which one deserves the outcome tends to perform better than rating one case at a time. If you only need the ranking of factors and you have many respondents, a conjoint analysis is the structured, quantitative cousin of this method.

How to write a vignette

1. Decide what you are varying

Write down the two to four factors you believe drive the decision. For the agency example: size of price rise, notice period, switching effort, and whether clients are affected. These become your dimensions. Everything else in the story stays fixed.

2. Keep it short and concrete

Aim for three to five sentences. Use names, numbers and specific events. "A large increase" means different things to different people; "30% at renewal" does not.

3. Make it plausible

Implausible combinations damage the data. In research on factorial surveys, Katrin Auspurg, Thomas Hinz and Stefan Liebig found that highly complex vignettes and implausible cases led respondents to take fewer of the described factors into account when making their judgements. If a combination would never occur in real life (a two-person startup with a 200-seat contract), drop it.

4. Write a neutral prompt

Ask what the character should do, would do, or how fair the outcome is, and decide which one you need. "Should" captures norms. "Would" captures expected behaviour. Mixing them across versions makes answers impossible to compare. Avoid evaluative words in the story itself ("unfortunately", "outrageous").

5. Choose a response format, then ask why

A vignette can end with an open question, a rating scale or a forced choice between options. Use a scale or choice when you want to compare versions numerically, and follow it with an open "why". The number tells you whether a detail changed the judgement; the explanation tells you how people reasoned about it.

6. Limit how many each person sees

Vignettes are cognitively demanding. A 2011 study in Survey Research Methods by Carsten Sauer, Katrin Auspurg, Thomas Hinz and Stefan Liebig found that respondents with lower levels of education gave less consistent evaluations when they had to rate a large number of vignettes, and that the number of dimensions only hurt consistency when respondents rated many vignettes. The authors concluded that the method works in general population samples, but with a limited number of vignettes and dimensions per respondent. Academic factorial surveys often show each respondent 10 to 20 descriptions; in a product interview, two to four vignettes with an open follow-up on each is a realistic ceiling.

Designing variations

The power of vignettes comes from changing details systematically and seeing whether answers move.

Between-subjects designs show each participant one version. Group A reads about a 30% rise with two weeks' notice; group B reads about a 30% rise with three months' notice. Compare the groups. This avoids participants spotting the manipulation, but needs more people.

Within-subjects designs show each participant several versions. You need fewer people and you can see how the same person's judgement shifts, but participants may notice what is changing and answer strategically.

Christiane Atzmüller and Peter Steiner, in a 2010 paper in Methodology, point out that because a full set of combinations is usually too large for one person, each respondent sees a subset, and how you choose that subset decides what you can analyse. Randomly drawn subsets mix effects together and rely on assuming no interactions between factors. A planned experimental design keeps the main effects and the most important interactions interpretable. For most product teams the practical version is simple: vary no more than two factors, cross them (2 x 2 gives four versions), and give each version to a similar number of people.

Common mistakes

Too many details. Every extra detail is another thing participants may react to instead of the one you care about.
A story that is really about the participant. If the vignette describes the participant's own company with the names changed, you have lost the distancing benefit.
Leading context. "Despite the team's loyalty, the vendor raised prices" tells the reader what to think.
No "why". A rating without an explanation tells you that a detail mattered, not how it was interpreted.
Treating answers as behaviour. Report vignette results as judgements and stated intentions, and say so in the readout.

Analysing vignette responses

For structured answers, compare the distribution of responses across versions. If moving the notice period from two weeks to three months shifts the share who say "switch" substantially, that factor matters. With small samples, treat differences as directional.

For the open explanations, code the reasons people give, then compare reason frequencies across versions. Often the most useful finding is not that the answer changed, but that the reasoning did: with short notice, people talk about feeling disrespected; with long notice, they talk about cost. That difference tells product and pricing teams what to change, not just what happened.

How Koji helps

Koji runs AI-moderated interviews, by text or voice, from a study brief you write once. That suits vignette research in a few specific ways.

The same vignette, delivered the same way. Your vignette goes into the brief as a question, and every participant hears or reads it with the same wording. Compare that with a human moderator retelling a story slightly differently in each session.
A structured answer plus a probed explanation. Koji supports six structured question types: open-ended, scale, single choice, multiple choice, ranking and yes/no. Put a vignette in a scale or single-choice question to get a comparable number, and let the AI interviewer ask follow-up questions about the reasoning. For scale questions, the interviewer can anchor the follow-up on the rating the participant just gave. You set how deep the follow-ups go per question.
Versions as separate studies. For a between-subjects design, write version A, then duplicate the study from its card menu and edit the vignette to create version B. The copy keeps the brief and settings, and each version gets its own link and its own results, so you can send each link to a matched group. Koji does not randomly assign vignette versions inside a single study, so plan the split in your recruitment.
Pilot before you field. The Preview tab on a study's Interviews page lets you take your own interview by text or voice, exactly as a participant would. Previews are free and never appear in the report. Read a vignette aloud in voice mode to check it is short enough to follow by ear.
Analysis that keeps the reasons. Koji's reports chart the structured answers for each question automatically and run thematic analysis on the open explanations, with quotes linked back to transcripts. Compare the two versions' reports side by side to see whether the rating moved and whether the reasoning changed.

Compared with a traditional survey tool, the difference is the follow-up: a form can collect the rating, but it cannot ask "what made you say Priya should stay?" and then ask again when the answer is vague. Compared with moderated sessions, you get consistent delivery and you are not limited by how many sessions a researcher can run in a week. The judgement about what to vary and how to interpret it stays with you.

A worked example

Here is an illustrative example (a composite, not a client case). Imagine a team selling scheduling software that wants to know what triggers a clinic to leave a vendor after an outage. Asking clinic managers "would you switch after an outage?" would produce confident, inconsistent answers. Instead the team writes one vignette about a fictional clinic manager whose booking system went down for a morning, and varies two details: whether the vendor contacted the clinic first, and whether patients were turned away. Four versions, each sent to a separate group. Suppose the structured answer (stay, look at alternatives, switch) barely moves with patient impact but moves sharply with whether the vendor called first, and the explanations show managers reading silence as a sign the vendor does not understand clinics. The roadmap item that follows is proactive outage notification, not uptime marketing. That is the kind of finding a direct question cannot produce.

Key takeaways

Vignettes are short, concrete stories about someone else, used to study judgements and decision rules.
They work best for sensitive topics, rare or future decisions, and multi-factor trade-offs.
Evidence from a real-world benchmark shows well-designed vignette and conjoint designs can track actual behaviour closely, with paired comparisons doing best.
Keep them plausible, vary few factors, show each person a handful, and always ask why.
Report results as judgements, not as observed behaviour.

Related Resources

Sources

Finch, J. (1987). The Vignette Technique in Survey Research. Sociology, 21(1), 105–114.
Hainmueller, J., Hangartner, D., & Yamamoto, T. (2015). Validating vignette and conjoint survey experiments against real-world behavior. PNAS, 112(8), 2395–2400.
Atzmüller, C., & Steiner, P. M. (2010). Experimental Vignette Studies in Survey Research. Methodology, 6(3), 128–138.
Sauer, C., Auspurg, K., Hinz, T., & Liebig, S. (2011). The application of factorial surveys in general population samples: The effects of respondent age and education on response times and response consistency. Survey Research Methods, 5(3).
Auspurg, K., Hinz, T., & Liebig, S. (2009). Complexity, Learning Effects and Plausibility of Vignettes in the Factorial Survey Design. Methods, Data, Analyses, 3(1).

Vignette Questions: How to Research Decisions People Will Not Describe Directly