Have an idea?

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

Max Diff Latent class analysis

Hi ,

I am currently trying to achieve a segmentation of max diff results on the Latent class software.

Uploaded the data through the export function of lighthouse (CHC).

I have two queries regarding it, running a default latent class analysis, only 23 of the 24 attributes appear in showing the relative importances between segments, could you indicate me why?

Secondly, the results themselves look very odd as in when I ran HB analysis, aggregrately item 1 for example is one of the highest at importance (e.g.8%) but in my latent class, in all of the solutions and the segments that item is coming up as one the lowest importance (1%).

Similarly an item that was of low importance  in the HB aggregrate, seems to one of the most relative importances in the latent class segment solutions.

Would you be able to indicate me perhaps where i could have gone wrong or let me know if you need more details.

asked Oct 13, 2016 by anonymous
The missing item is fixed at a utility value of zero, all other items' utilities are estimated relative to that fixed value.


Thanks Dan!

Any suggestions of the other issue? why it would not generate any segments with an item being important that was in the aggregrate HB analysis?


1 Answer

0 votes
Dan's right about the missing item (the reference item) in one implementation of our software just meaning that the reference it (typically the last item) is a 0 utility.

But, the rank-order of preference of the weighted average across the latent class vector solutions (where the weights are the segment sizes) should align pretty closely with the average weights from HB analysis.  If those two aren't highly correlated, then you should look for data processing errors of some sort.
answered Oct 14, 2016 by Bryan Orme Platinum Sawtooth Software, Inc. (201,565 points)