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

0 votes

It depends a bit on what kind of models you're running and what your segments are. Assuming your segments are groups of respondents defined by some other variables in the study (e.g. demographics, usage variables) and assuming you're running hierarchical Bayes MNL, you can run traditional "frequentist" stat tests with the respondent level utilities, just like you might do for any other statistical test of any other survey research entity. More appropriate will be to run Bayesian tests. In either case you can contact me for more details if you like: keith@sawtoothsoftware.com.

Thanks for the reply. The model I'm aiming at is the latent-class analysis.

In this case you can use the standard errors of the utility estimates, and then just as you would for any other test (e.g. a t-test) you can use the difference in estimates and the standard errors to construct a statistical test. Your significance will be badly overstated, however, if you're testing the difference between segments that result from the latent class analysis.

Think about it, you're separating respondents into groups on the basis of their utilities and then you want to test if the utilities are different. In a different context, if you separated people into groups based on their eye color, would you be very surprised if the resulting groups differed in terms of their eye color?

Think about it, you're separating respondents into groups on the basis of their utilities and then you want to test if the utilities are different. In a different context, if you separated people into groups based on their eye color, would you be very surprised if the resulting groups differed in terms of their eye color?

...