Conjoint sample-size estimator
How many people does your conjoint study need?
Build a starting budget for your study. Adjust the choice questions, reporting groups, and recruitment assumptions to see what drives the sample estimate.
Describe your study
For full-profile choice-based conjoint (CBC): respondents choose among products described by the same set of attributes. The example inputs below are editable. Adaptive, partial-profile, and menu-based studies need a different assessment.
Rule-of-thumb planning
A starting point for your budget
300usable respondents to plan for
Driven by: overall budgeting base.
Statistical power has not been assessed. This is a budgeting reference for full-profile choice-based conjoint.
What drives the number?
The estimate uses the largest of these requirements. Each applies to the same sample.
For your objective
Check the actual design and the precision of the feature comparisons that matter before setting the final sample.
Translate this into recruitment
- Completed interviews
- 334
- Eligible survey starts
- 418
- Screening starts
- 836
Expected yields at your entered rates, rounded up at each stage. Quotas may require additional screening.
Edit study inputsSee the formula and study summary
The Johnson–Orme planning rule
N ≥ exposure × c ÷ (t × a)
Here c = 4, t = 12 estimation questions, and a = 3 products. For main effects, c is the largest attribute level count.
- At 500 appearances
- 56
- At 1,000 appearances
- 112
The formula excludes validation questions and None from these counts. It does not inspect the actual experimental design or measure statistical power.
Your 5 attributes contain 9 main-effect parameters with categorical coding, before adding interactions or alternative-specific effects. The simple rule does not account for that full complexity.
3,600 estimation choices come from 300 people. Those choices are repeated observations, not independent respondents.
Explore a trade-off
Would more questions change your sample plan?
The exposure rule changes with question count. Your audience requirements may still determine the budget. More tasks also mean more work for each person; test that burden with your audience.
| Estimation questions | Exposure reference | Combined sample reference |
|---|---|---|
| 8 | 167 | 300 |
| 12 Current | 112 | 300 |
| 16 | 84 | 300 |
Validation questions are additional. These are formula comparisons, not tested designs or predictions of model performance.
Understanding your result
A useful budget starts with the decision.
A study can estimate overall preferences reasonably well and still struggle to distinguish two similar products or a small customer group. Tell us which comparison matters when you bring a plan to Russell.
Discuss your study designWhy does the headline differ from the formula?
The Johnson–Orme rule uses attribute levels, estimation questions, and product alternatives. A study with four levels, twelve estimation questions, and three products returns 56 people at the 500-appearance reference, or 112 at 1,000 appearances.
Orme describes 500 as a minimum exposure convention. The paper also offers practical budgeting guidance of 300 overall and about 200 per reporting group. This tool starts with those conventions, uses the higher exposure reference by default, and takes the largest applicable requirement. That combination is our transparent planning policy, not a validated statistical formula. [1]
Does meeting the reference mean the study has enough power?
No. Power is the chance of detecting a specified effect when it exists. It depends on the design, model, assumed preferences, effect size, and statistical criterion. Those inputs are not evaluated by this calculator. A formal assessment should address the differences your decisions depend on. [3]
Design-based standard errors can inform a conditional power calculation. They should come from the intended design and model; an aggregate model does not establish the precision of every person's preferences. [4]
Can more questions replace more people?
Additional questions provide more information from each respondent. Additional respondents broaden the people represented. The comparison table shows how the exposure rule changes, but it does not measure fatigue, representativeness, or design efficiency.
Questions held out for validation are excluded from estimation here. The example's two holdouts are not a recommendation for every study; validating subtle differences between models can require more. [5]
What can a larger sample fail to solve?
A large sample cannot separate features that the design always bundles together. More respondents also cannot repair an unrepresentative recruitment approach or unclear product descriptions. Weighting and unequal group sizes can change precision.
This planner does not inspect experimental designs, calculate confidence intervals, discover segments, or simulate willingness to pay. Those questions need assessment using your design and intended analysis.
Research behind the planning rules
- Orme, B. (2019). Sample Size Issues for Conjoint Analysis. Getting Started with Conjoint Analysis, 4th edition, chapter 7, particularly pp. 64–65. Basis for the exposure formula and practical sample conventions.
- Halversen, C. (2020). Sample Size Rule of Thumb for Choice-Based Conjoint. Sawtooth Software. Explains the limitations of exposure rules for individual-level models.
- de Bekker-Grob, E. W., et al. (2015). Sample Size Requirements for Discrete-Choice Experiments in Healthcare: a Practical Guide. The Patient, 8, 373–384. A framework for formal, model-specific sample planning.
- Chrzan, K. (2019). Quick and Easy Power Analysis for Choice Experiments. Sawtooth Software. Uses standard errors from an experimental design to examine detectable effects.
- Orme, B. Including Holdout Choice Tasks in Conjoint Studies. Sawtooth Software, 2015 archive edition. Explains the purpose and limitations of validation tasks.