Have an idea?

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

How does the market simulator calculate share of preference?

No matter which way I try to calculate share of preferences with the raw utilities, I never get the same values as shown in the market simulator. I do it for each respondent individually, I have tried unweighted and weighted (with importance of attribute) calculation of total utilities, I have tried with exp and without exp. I also have tried rescaling the utilities so that they are positive values but I'm not sure what should be the base (lowest value) in that case.
Can anyone help and explain? Or post a link to where the formula is explained?

Thanks a lot!
asked Dec 12, 2017 by Ann1212

1 Answer

0 votes
Make sure you are using the raw HB utilities (not the rescaled, zero-centered diffs).

Sum the total utility for each of the alternatives in the simulator.  Exponentiate the sums.  (That means use =exp(U1), where U1 is the total utility for alternative 1, etc.

Normalize those exponentiated sums to sum to 100% within respondent.

Average those results across people.

If you are still having problems, you can email your spreadsheet to our technical support group and let them have a look to see where your mistake is.  

My best guess is you are trying to use zero-centered diffs rescaled utilities rather than the raw utilities.
answered Dec 12, 2017 by Bryan Orme Platinum Sawtooth Software, Inc. (187,915 points)
Oh, another possibility is that our simulator uses Randomized First Choice as the default market simulation method.  If you change that to be "Share of Preference", this is what you are trying to replicate.

If you are using Discover-CBC software, the market simulator is set to Randomized First Choice and you cannot change it.  Thus, you would not be able to reproduce the Randomized First Choice results using the logit (Share of Preference) math steps.
Hello Bryan, thanks for the quick answer.
Yes, it must be the Randomized First Choice logic which is responsible for the different values. And I guess this simulation method would be a bit too complex to reproduce outside of the simulator, correct?
It's possible to mimic RFC within the Excel environment, but due to the differences in random number generators, you'll never get exactly the same answer.  Glad you figured out the differences.