Have an idea?

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

Share of Preference Model with Correction for Product Similarity


I have queries regarding the Sawtooth article that talks about how to build the Share of Preference model in a conjoint simulator while accounting for product similarity.  In the first step of the procedure after we calculate the dissimilarity matrix, it says:
"Next, total dissimilarities are rescaled by a constant so the maximum possible is 3.0, rather than 10 times the number of attributes."
I have two questions about this:
1) What is the reason of choosing 3 as the maximum?  Is it a subjective assumption?  Can we tweak this?
2) The procedure does not take attribute part worths into account - which means similarity in any attribute will reduce the preference share by the same ratio.  Is this a correct hypothesis?

Please correct me if I have misunderstood.  Thanks in advance.
asked Sep 5, 2017 by Kshitij

1 Answer

0 votes
What a blast from the past.  The old Model 3 correction for product similarity was initially devised by Richard Smallwood and was implemented in software code for Sawtooth Software by Rich Johnson sometime, I think, during the 1980s.  Starting in the late 1990s, we no longer recommended its use (after seeing anomalies and problematic simulation situations); instead recommending Randomized First Choice (RFC).  We dropped the old Model 3 correction many years ago from our software products.  None of us remaining at the company know the algorithm, so we don't know the answers to your questions.

The best first step to reducing IIA problems and correcting for product similarity is to use individual-level utilities, such as from HB estimation.
 We strongly recommend you take another market simulation approach such as RFC or simulating using either first choice or the logit equation on the HB beta draws.  Depending on your sample size, your needs for individual-level precision, and the speed of your simulator, you can use 50 to 500 beta draws per respondent and do an excellent job.
answered Sep 5, 2017 by Bryan Orme Platinum Sawtooth Software, Inc. (131,390 points)