How to compute utility after a choice survey ? / What formula does JMP use to do so ?

Question

Good morning,

I am trying to compute Willingness to Pay based on the responses of a survey. I want to know based on which articles, or where I can find the formula used by JMP to compute the utility of the different products I evaluate.

In JMP Help platform (Consumer Research>Choice models>Statistical details>Utility and probabilities), there is a formula yielding the probability that a given subject makes a given choice. I do not really understand what the "subject attribute design row" or the "choice attribute design row" are, and how to evaluate them. And I think that the calculation of utility might be based on that probability, but I don't know how to perform it.

The bottom idea is that the Choice platform in JMP needs a very special formatting of data, and a special type of questions to get the data in shape and analyze it. I think I would be able to better understand which questions I can ask to get WTP if I understand the whole calculation behind it.

Thank you !

AB

txnelson · Answer

You should be able to find the information you need in the documentation for the Consumer Response area in JMP
     Help==>Books==>Consumer Response

Search on "Choice Models"

akb3 · Answer

Good morning cwillden,

Thank you for the answer ! Do you think you could give me an example with two subjects and two choice sets for example, very basic, so I could understand better how it works ? Or redirect me towards a (not too advanced) article that describes it ?

I have trouble visualizing the shape of the matrices and what values they contain.

Basically then, the utility of a feature j for a given subject k would be the beta(X[k] x Z[j]) ? And to get the average utility, just average it over all subjects k ?

Thank you for your time !

AB

cwillden · Answer

You're right, I read through your question a little too hastily.  Sorry about that.  The measures of "utility" are the model parameter estimates (the betas) of a logistic regression model.  You have 2 matrices describing the covariates; X identifies the subject in the study Z identifies the choice set.  You should have parameter estimates for each feature in your choice sets.  The larger the number, the greater the average utility across all subjects.  Sorry, I only have a minute to respond.  I'll try to add more later if needed.