@hengehold_da1 were you able investigate this further? We found the tech support track and this is what was shared with you:
JMP calculates the utilities according to the conditional logistic regression approach discussed in the below reference. I'd recommend contacting Sawtooth to ask about their methodology so you can compare the two.
McFadden, D. (1974), “Conditional Logit Analysis of Qualitative Choice Behavior,” in P. Zarembka, ed., Frontiers in Econometrics, pp. 105–142. See http://eml.berkeley.edu/reprints/mcfadden/zarembka.pdf . Retrieved 4/25/2016.
Note that the marginal utilities are the same as the parameter estimates for the analysis. In general, marginal utilities can be thought of as a measure of the worth of that choice to the respondents. And as you stated, the utilities can be exponentiated and expressed as ratios to calculate the probabilities:
Marginal Probability = exp(utility) / sum(exp(utility) all other choices)
Detail: in my first attempt to use the MaxDiff Firth-Bias utility estimation in JMP, I found that the raw utility scores were exactly half that of Sawtooth Software. Given that the next step is to rescale them through exponentiation for interpretability, the discrepancy (half vs. whole) impacts the range of results. I intend to try another project to see if the observation is reproducible when I get a chance.
