What's the best way to generate segmentation off DCM data in two conditions

Question

Hi:
There is a chance this is a duplicate posting.  I apologize if it is and please feel free to delete one of the two.
We are conducting a DCM for a client and will be using HB for our standard utility modeling and preference share estimation.  No problem there.
We know we can build a segmentation off the same data using your LC program.  We were wondering if because we are using HB for our preference work whether we might be better off exporting the utilities from that to Latent Gold, which can handle continuous variables in LC segmentation?  What do you think?
Second, we may want to combine the DCM data with other questions to create a segmentation.  Are we better off using the raw utilities from the DCM for this or, first, building a segmentation off the DCM data and then using that segmentation as our way of combining DCM results with non-DCM questions.

Answer 1

It is possible to use linear (continuous) variables within our LC software.  For example, you can ask our LC system to model the Price or Speed attribute as a single coefficient, rather than using part-worth specification.

That said, Latent Gold does have more options for including non-conjoint variables within the run.  So, you may want to do it for that reason.

One of my favorite ways to use latent class is just as a segmentation tool, and for the purposes of placing people into segments.  You can then take those segmentation assignments, and merge those segmentation variables into our conjoint simulators (either the Online Simulator or SMRT) and use them as banner points and filters.  The underlying utilities you are using for predicting peoples' choices are still HB utilities, but the segments are broken out by those discovered using LC.

Generally, I think it's better to ask LC to develop segments not based upon HB utilities (I believe Latent Gold can do that, as it can run Latent Class on a series of basis variables, similar to the idea of running cluster on a bunch of basis variables), but on the original DCM data: the design and the choices.  HB involves some smoothing of the sample toward the upper-level model (often, the population).  So, to try to develop segments based on variables that have already been involved in smoothing out differences between people and segments seems like going one step forwards and two steps back.

But, please talk to the Latent Gold people about that, as LC is their emphasis.