How to Run a Conjoint Experiment
A practical path from a conjoint research question to a fielded study, analysis, and documented result.
In Brief
A conjoint experiment starts with a real choice that respondents can understand. Once that choice is clear, define the profiles, pilot the task on the devices people will use, field it by URL, and carry the profile-level records into AMCE or ATE analysis.
Begin with the Choice
Start by asking what decision respondents should actually make. Once that is clear, define the thing being compared, the attributes that can vary, their levels, the choice prompt, the number of tasks, and whether a neither option belongs in the design. A clear choice keeps the rest of the study grounded.
Pilot Before Launch
Run the complete survey with a small pilot before paying for the full sample. Check whether the task makes sense, how long it takes, how profiles fit on a phone, whether randomization behaves as intended, and whether the exported data have the structure you expected.
Analyze and Document
After fielding, keep every displayed profile in the analysis file and account for repeated tasks from the same respondent. Then preserve the codebook, design summary, methods text, exclusions, and appendix materials alongside the results. That record will save time when the paper is revised or reviewed.
FAQ
How many tasks should respondents complete?
There is no single correct number. More tasks can improve precision, but they also increase fatigue. Choose a defensible starting point, then use the pilot to see how long the survey takes and whether response quality weakens near the end.
Should I use impossible-profile restrictions?
Use restrictions for combinations that truly cannot exist, not merely combinations that seem unlikely. Document every restriction and remember that the resulting estimates describe the profile space respondents were allowed to see.