Have an idea?

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

How do I know which utility model and which probability of choice model I used?


I used Sawtooth Software Lighthouse studio 9.5.3 for a CBC with traditional full-profile CBC-Design plus None-option and with Balanced overlap. As response type I used discrete choice and I randomized the tasks for each respondent.
For my analysis I used just the HB-utilities in SPSS.
Now I have to describe the model. How do I figure out what kind of utility model and which probability of choice model the software used?

utility model: vector model, ideal point model or partworth-model?

probability of choice model: max-utility-model, random-choice-model, attraction-model or logit-choice-model?
(source: Backhaus, Erichson, Weiber (2015))

Thanks for your help!
asked Mar 7 by Hannah

1 Answer

0 votes

For utility estimation, if you used the default settings in the software you used the part-worth model (i.e. you used a particular categorical coding of the variables called effects coding).  Settings in the software allow you to run a linear (or a transformed-to-linear) utility model if that's what you would prefer.

The software uses some variant of the conditional multinomial logit (MNL) model and it sounds like you used the hierarchical Bayesian version of that in your analysis.  conditional MNL is by far the most widely used model both by practitioners and academics (even many of the other choice models can be estimated in the conditional MNL framework with minor adjustments that are easy to do if you export your data files to Excel, make the adjustments and then analyze the data in our standalone CBC/HB software.
answered Mar 7 by Keith Chrzan Platinum Sawtooth Software, Inc. (53,775 points)
Hi Keith,

thanks for your answer!
I got the first step: I used the default settings in the software, that means I used the part-worth model. (So I use partworth utilities; my HB utilities are individual partworth utilities right?)

What ist the probability of choice model for?
I read that MNL is to aggregate data. Did I used then both MNL and Hierarchical Bayesian (HB) or is HB a version of MNL?
What does MNL do?


A good way to think of it is that MNL is a particular way of mapping utilities to choice probabilities.  It is the mathematical basis of a family of models, including models at different levels of respondent aggregation.  So an aggregate MNL is a model that describes all the respondents.  An HB MNL model provides MNL utilities for individual respondents; and a latent class MNL model produces utilities for segments of respondents.  All three are MNL models.  I hope this helps.
Hi Keith,
I think I got it! Thanks a lot for your help!
Hi Keith,
one more question to this topic:
Does the partworth-model then works additively rather than multiplicatively in my case to get the total utility model?
With our software you can build additive utility models but also models with multiplicative elements (e.g. interactions).   If you use our CBC/HB software you can customize your design and its coding even more flexibly.
Thank you!