Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Minimum Sample Size for Max Diff and CBC


Please let me know what is the minimum sample size recommended for CBC and Max Diff
asked Jul 26, 2012 by anonymous

1 Answer

+1 vote
It's question you can't answer without further information. Look at the SS paper:
They state how to calculate sample size for many applications.

Personally I never field neither CBC nor MD project with sample smaller then 200, though I often compute both for smaller subsamples. Minimum sample size is really a number which depends on many variables
answered Jul 26, 2012 by lkomenda Bronze (2,810 points)
Indeed, it's hard to know what to say on this.

Let's talk about an extreme case, for illustration.  Let's imagine there were only 10 people in the world who were in the market for a particular rare good or service.  And, let's imagine there were only 4 attributes with 3 levels each for a conjoint study of that good.  If you interviewed all 10 respondents, and each respondent completed 15 choice tasks, I suspect you'd get very good understanding and precision in predicting the choices of those 10 respondents.  There are only 9 parameters to estimate with this model, and with 15 choice tasks it would seem to provide adequate information.

Consider MaxDiff for the same problem, with just 12 items, and each respondent sees each item 3x across the MaxDiff tasks.  Resolution should be extremely good within each respondent in terms of measurement precision.  If you interview all 10 respondents who are in the population, then I would think you'd have excellent results.

In both cases, I'm assuming the use of HB estimation.  Though, purely individual-level estimation via individual-level logit could be possible.

So, with small designs in CBC and MaxDiff, and when your tiny sample sizes cover just about ALL potential respondents, then the information should be solid and these methods are feasible with what are usually considered tiny sample sizes.