Research Guide
Conjoint Analysis Survey Design
Conjoint quality depends less on the visual table than on the design behind it: attributes must map to hypotheses, levels must be plausible, randomization must be auditable, and the task burden must match the sample plan.
A practical workflow
- 1
Specify the estimand
Decide whether the study estimates marginal component effects, choice probabilities, subgroup differences, or prediction.
- 2
Author attributes and levels
Use clear, separable concepts and prohibit only genuinely impossible combinations so the design space remains informative.
- 3
Choose task structure
Set alternatives, repeated tasks, opt-outs, vignettes, and holdouts to match the real decision context.
- 4
Check exposure and sample size
Review expected level exposure and power assumptions before recruitment, especially for interactions and rare levels.
- 5
Export the realized profiles
Keep every displayed level, alternative order, task number, and selected alternative with the response data.
Key design decisions
| Decision | Guidance |
|---|---|
| Table or vignette? | Use a table when respondents compare alternatives directly; use a vignette when they should react to a single composed profile. |
| How many attributes? | Include only attributes needed for the research question. More attributes increase burden and dilute exposure to each level. |
| Independent draws or balanced combinations? | Independent draws are flexible; balanced combinations improve exposure balance when the valid design space is small enough. |
Common mistakes
- Adding attributes because they are easy to vary rather than theoretically necessary
- Removing unusual but plausible profiles
- Reporting only the chosen alternative without displayed levels
- Planning subgroup interactions with too little level exposure