Have an idea?

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

Clustering choice data to fit HB ?


I have a synthetic choice data that includes pre-defined individual preferences.
The individual preferences are intentionally defined to generate segments.
Now, I am thinking two different HB fitting scenarios.

1. HB fit for population. Then, compute the segment-level utilities
2. Divide the data into a few segments. Then, fit HB for each segment.

In the result of the scenario 1, I observed some over-estimated utilities near zero . I guess this is due to the normal distribution assumption for prior.
The scenario 2 resulted in plausible results that meet my expectation (?) about the pre-defined preferences.
Is is possible to use segmented choice data?
asked Jun 13, 2014 by jkshin933 (185 points)
reshown Oct 30, 2017 by Walter Williams

1 Answer

0 votes
In your approach #2 did you break the data into the segments you knew from your data generation process would be different?  If so, I think you might be better off running a single HB model but using segment membership as a covariate.

I guess the answer really depends on your objectives for this exercise and how realistic you want to be.  It occurs to me that if this were a real study, the segmentation variable might be latent so that neither the covariate approach nor your approach #2 would really be viable.
answered Jun 13, 2014 by Keith Chrzan Platinum Sawtooth Software, Inc. (92,075 points)