Have an idea?

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

None option part-worth calculation


I'm trying to reproduce my CBC Sawtooth Logit results in R with the mlogit function. I get the exact same part-worth utilities, standard errors etc for all attributes except for the No-choice option part-worth.

So my question is: Is the part-worth of the no-choice option calculated differently than the other part-worths?

Thank you very much!
asked May 19, 2020 by alinalefebre (225 points)

1 Answer

0 votes
The utility estimation is the same for the None; the concern would be how you are coding it in the design matrix.  Different codings can lead to the same maximum likelihood solution and same predictions in the market simulator, but that the utilities between None and the other attributes shifting around with respect to one another.

We code the None as its own product alternative, with zeros across the design vector, except for a 1 in the None column position.  In other words, we are dummy-coding the None alternative.
answered May 19, 2020 by Bryan Orme Platinum Sawtooth Software, Inc. (203,065 points)
Thank you for your response. If I'm understanding it correctly, I'm doing the same coding.
I have a row for each alternative (4 incl. None Option) and a column for each attribute plus separate column for the None option.
Is there anything else which could lead to different results? The difference is not that big: -1,79 vs. -1,49

Thank you again.