Have an idea?

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

Auto Calibrating Attribute Variability Multiplier for RFC

Hello Sawtooth team,

the error term E_A for the attributes is Gumbel distributed in SMRT. How do you determine the mean and the variance parameters from the variability multiplier for the Gumbel distribution (mu = 0, beta = variability multiplier?) and how do you auto calibrate the attribute variability multiplier for RFC? On which parameters depends the auto calibration? On the number of compared products in the simulation? I use CBC/HB data.

I would really appreciate an answer! Thank you in advance.

Kind regards,
asked Oct 6, 2014 by Frank

1 Answer

0 votes
We subtract off the mean for the Gumbel distribution to zero-center the error, but this really isn't necessary.

The attribute error multiplier depends on not only the number of attributes used in the simulation but in the number of products in the scenario.  We have conducted an extensive series of Monte Carlo simulation runs to find the right amount of error for every #attributes x #Products condition that that will return appropriate probabilities of choice under RFC (that mimics closely the logit rule simulation approach).
answered Oct 7, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (174,440 points)