Have an idea?

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

MBC using traditional tools

Hello everyone,

I scripted an MBC study in SSI Web 7 referring to the article "MENU-BASED CHOICE MODELING USING TRADITIONAL TOOLS", Orme, B., Proceedings of the Sawtooth Software Conference 2010, and examples found on the site.

Now as it is about to be launched, I'm thinking about ways of analyzing the data. I'm really eager to try running HB analysis according to the Exhaustive Alternatives Model (1- or 2-stage). When it comes to coding the alternatives, the paper suggests effects-coding for data matrix.

What I'm wondering about: Is it possible to create a .cho file(s) for such data using the same logic as in the paper?

I'm not very used to coding design matrices as effects or dummy, but have done a lot of .cho-file building and recoding in application to various CBC projects, so for me it would be more comfortable to build a .cho file based on my study design and data with a VBA macro.
I would also create an .att file(s) for the study.

I suppose if I do so, then the menu alternatives will be recognized as part-worths in CBC HB, and if I'm right, that means they'll be coded as effects internally in HB.

Please let me know what you think on this point.
Many thanks in advance!

Igor
asked Sep 9, 2014 by Igor Gusev (190 points)

1 Answer

0 votes
If you do something similar to the exhaustive alternatives example within that paper, then you will be using "user-specified coding" within the CBC/HB software.  The example on page 50 of the 2010 Proceedings (that you cite) has 48 alternatives--meaning that there are only 48 possible different ways respondents could fill out that menu.  As the menu grows in complexity, the possible combinations will grow exponentially until it isn't feasible to use this approach (unless splitting the menu into sub-sections that you model as exhaustive alternatives within each sub-section).

If you are coding up the parameters as "user-specified", then you really must follow the instructions in Appendix G of the CBC/HB manual to specify the proper prior covariance matrix.  With the exhaustive alternatives specification (which often involves a categorical attribute with a dozen or more mutually-exclusive outcomes), without the proper prior covariance matrix within the advanced settings menus of CBC/HB software your parameter will be very much biased and wrong for the effect of the "reference level" within your effects-coded categorical variable (relative to the other parameters).
answered Sep 9, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (133,765 points)
Thanks Bryan, I agree with your point regarding "user-specified" parameters, but I don't entirely understand why we can't pretend that we show 16 (if we simplify it to one car with 4 options) concepts on each screen, where the attributes of each concept would be:
- Opt. 1 (Yes/No)
- Alt-spec Price for "Opt.1, Yes" (4 levels)
- Opt.2 (Yes/No)
- Alt-spec Price for "Opt.2, Yes" (4 levels)
- etc.

Looks like an alternative-specific CBC setup, though the design might be the issue in this case.
...