Have an idea?

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

Number of needed screens in a CBC/HB survey?

Dear all,

we are currently preparing a survey with one part dedicated to a CBC/HB conjoint analysis.

We want to test 3 possible offerings from one company vs. "none of these offerings" along 5 attributes with up to 5 variables.

My question is the following: given the fact that the number of expected respondent is between 2500 and 3000, what is the number of screens that each respondent should have in order to have a robust estimation of individual utilities for each respondent and each variable (max 5x5 variables)?

Thank you very much!!
asked Mar 7, 2012 by anonymous
edited Mar 7, 2012 by Walter Williams

1 Answer

0 votes
So, with 5 attributes each with 5 levels...  And, I assume you'll want to show at least three products on the screen plus the None...  And, assuming respondent count of 2500 to 3000...  And, assuming you don't need to examine results by different segments:

That's a very standard CBC design with 5(5-1)+1 = 21 parameters to estimate to account for Main Effects.

Based on results presented at our conference over the years, it appears that getting about 10 to 12 choice tasks will be sufficient to get good individual-level estimates.

But, if you told me that you were interested in just having good group-level predictions (and you were willing to live with some more noise at the individual level than was typical) your large sample size might lead me to suggest having just 4 to 6 CBC tasks.  If you did that, I would ask you to modify the default settings in CBC/HB software to use a lower Prior Variance (such as 0.5) and a much higher prior Degrees of Freedom (1/4 of your sample size).
answered Mar 7, 2012 by Bryan Orme Platinum Sawtooth Software, Inc. (198,315 points)
...