Does Choice Design platform produces odds ratio?

[#1]

Hi,

I'm new to JMP and statistics, and I design a discrete choice experiment using Choice Design platform and use the platform to analyze the survey data, following instructions in this case study  ( Case Studies on Designing and Analysing Discrete Choice Experiments Using JMP® ).

According to this case study, choice model being estimated in choice design platform is a multinomial logit model. From what I've read, for MNL we would want to know odds ratios instead of coefficients (or parameter estimates as called in JMP) of each attribute level to compare their effects. Does JMP do this for you? If so, where can I find it?

[#2 · Accepted solution]

The Choice analysis in JMP does not provide the odds ratios. It provides a Utility Profiler and a Probability Profiler to predict the utility or the probability for any combination of feature values. You can also save the utility model as a new column formula.

The odds ratio is based on the odds, and the odds are based on the probabilities. They have interpretative value in other domains that use logistic regression models. You might see if the profilers provide the information that you want from your model.

[#3]

Thanks for your reply! 

I'm also confused about the choice model being estimated in JMP. It produces a parameter estimate report of the model, but what's the dependent variable of choice model? Is it the probability of one alternative being selected or its utility? 

I read JMP's Consumer Research Guide and it says:" The Parameter Estimates report gives estimates and standard errors of the coefficients of utility associated with the effects listed in the Term column."

However, in the video in the case study the presenter says its a multinomial logit model that's being estimated, and if so the dependent variable should be probability (if I understand MNL right). 

[#4 · Accepted solution]

The parameter estimates are the estimated utility values. In the presentation you referenced, the model under consideration is on the slide 19 (Multinomial logit model), and those utilities are the beta coefficients. One subtle difference from what you might be used to is that it is conditional on a given choice set, so that probability changes depending on the choice profiles in a set. I've attached a paper if you want to look into some of the details.

So, those utilities can be used to predict probabilities in a new choice set, such as in the probability profiler or multiple choice profiler.

As Mark noted, you won't directly get the odds ratios, but you can use the "Comparisons" option from the red triangle menu, to get the odds for two different choice profiles that you specify, if you are just wanting to compare one profile to another.