Have an idea?

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

What are best practices for Volumetric CBC design?

We are thinking of using Volumetric CBC for the first time, and have some questions.

- When conducting a Volumetric CBC, do the sample-size estimating guidelines for CBC still apply?  In other words, if a CBC is changed to Volumetric CBC would the the same sample size be required, or do we need more?

- Given how much more arduous it must be for a respondent to enter the allocations, is there any recommendation about how many loops to take a respondent through?  Normally for CBC, our projects range from 8-16 loops, but perhaps we ought to use less here?

- Finally, on the forum I have seen several comments that amount to "it is still being debated, but there is concern about whether this allocation data can be collected in a way that is realistic and valid".   I haven't seen any discussion recently, though, so is there any newer opinion on this subject?
asked Aug 3, 2017 by Andrew S (235 points)

1 Answer

+1 vote
Volumetric CBC is more demanding on the researcher than is allocation CBC because there is no constant amount that the responses must sum to.  Your questions seem to be about allocation CBC, however.  Which is it that you're wondering about?
answered Aug 3, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (97,375 points)
Thank you Keith.

I think I have used the wrong terms, but it must be "allocation" that was misplaced.  I was thinking of the kind of CBC where people can enter an amount that represents how many they would buy, and there is no particular sum required.

I'm aware of the Volumetric CBC .CHS Converter and was planning on using it.

This helps a lot:  there are a lot of things about volumetric that you have to  be careful about.  We might typically have respondents do slightly fewer volumetric questions than choice questions but the method is still pretty demanding in terms of sample size, so we don't want to have too few questions.  On the other hand, it's easy for respondents to get fatigued or bored with allocation questions, so you have to check their responses really carefully - often we'll see respondents buying multiple units of multiple SKUs early in the survey but after 8 or 10 questions some respondents start choosing just one unit of one SKU.

It helps in the questionnaire to add some volume checks and reminders.  For example, you might before the CBC questions ask the respondent what the maximum volume they'd ever buy is, and then prevent them from buying more units than that in their CBC questions (or at least pop up a question along the lines of "Hey, you said you'd never buy more than 7 and here you are buying 38 - is that right?").

Finally, you need to jump through some extra hoops when it comes to the analysis of volumetric choice data as there is no automated way to do it within the software like there is for discrete choices or allocations.   Check out Tom Eagle's paper in our 2010 Conference Proceedings for some good advice about modeling volumetric choices.
Hi Keith

Just to answer my first question, we've explored required sample sizes using the software and it seems that adding volumetric to a CBC doesn't demand a larger sample size than without it (everything else being equal).  However, as you suggested, it seems that we'll want to have fewer loops which then would obviously have a sample size requirement impact.

Thanks for the suggestion about checks to keep people focused on giving accurate answers.
I agree - and good luck!  I'd be interested to hear how this turns out for you - sometimes folks find volumetric studies tough, though I've usually had pretty good success with them.
Thanks.  I guess you've answered my last question, too, now.

If we are awarded the project by the client, I will come back here and let you know how we do!