Have an idea?

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

CBC with partial profile

Hi Keith,

First of all, thanks a lot for your ideas.

A few more questions arised with reference to the CBC partial profile approach. Let us assume, we would produce the information for the comparison of the different areas with an alternative approach.

So, performing the CBC partial profile then could focus on the attributes. The different areas then would only be relevant for grouping the attributes. But now my question:

1) I guess it would be possible to only show attributes from one area within each CBC task by using contraints - Is this also possible in case we have a different number of attributes within each area? So sometimes we would show e.g. 5 attributes dealing with the seat, but showing only 3 attributes for the steering wheel area? (By the way: Can the question text be adapted depending on which area is shown?)

2) In case we would show e.g. 15 or 20 CBC tasks to each respondent, each respondent would see about 2 CBC tasks for each area (we got about 10 areas) - would this produce enough information for the CBC/HB analysis when we conduct about 400 interviews in total?

3) And last but not least: Would it also be possible to tell the Sawtooth software that e.g. it should first show 5 CBC tasks only from one area (so 5 tasks e.g. only dealing with the seat), and the next 5 for another area, and so on. In this case, each respondent would see 5 tasks for each area, but only attributes from 4 areas in total (in case of 20 tasks)?
But I am not sure if the last idea would procude good data quality for HB estimation at all.

Thanks a lot in advance
related to an answer for: Some hint for an appropriate approach
asked Jan 24, 2013 by TomHenning (140 points)

1 Answer

0 votes

In a partial profile CBC all attributes appear with one another, without respect to area, as you have here.  So in this case you're not looking at a partial profile model in so much as a separate CBC for each area.  So if you have 10 areas, this is 10 separate, if smaller, CBC experiments.   

You still won't be able to get by with just 2 tasks per area unless your sample sizes are huge, so what I would recommend is to have each respondent receive CBC questions for 3 of the 10 areas perhaps.  Decisions about adequate sample size depend a lot on the number of attributes and levels you have; if the numbers are reasonable you can usually get by with 200 respondents, so if each respondent sees 3 areas (and a dozen CBC questions per area, this means you'd need over 600 respondents.  

For your last point, because this is essentially 10 separate CBC tasks, there's no need to put in a lot of constraints and try to program this as one large experiment - that would needlessly complicate things.

answered Jan 24, 2013 by Keith Chrzan Platinum Sawtooth Software, Inc. (62,700 points)