Have an idea?

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

Recreating Sawtooth Latent Class output in Latent Gold


I have done a Maxdiff and have used the Latent Class analysis in Sawtooth to create segments.

I would now need to do another segmentation, in which I want to combine the Maxdiff output with additional variables in my dataset. I believe it is not possible to add these variables to the Latent Class analysis within Sawtooth.
Therefore, I would use the Latent Gold software.

However, I was wondering which input I should use, to get the same segments for the Maxdiff data only in Latent Gold compared to the Latent Class in Sawtooth. (just to make sure that the "original" segmentation based on Maxdiff only, is the same in both softwares)
Do I use the HB-utilities (probability scores/raw scores)? Or the original raw Maxdiff input?

Many thanks
asked Feb 22 by Tina Van Regenmortel (265 points)

1 Answer

0 votes

You are correct that you'll want to run your model mixing MaxDiff responses and other variables in Latent Gold.  The beauty of running this kind of latent class data is that you use the original MaxDiff choice data in Latent Gold:  were you to want to segment based on the HB utilities and other variables, you could run that in our CCEA cluster analysis software - but running this kind of model in Latent Gold and using the choices and the experimental design as an input in addition to whatever other variables you have is really the way to go.  

You may not get the same results for the MaxDiff only analysis in Latent Gold.  In fact, in Lighthouse Studio you might also not get the same results for your MaxDiff latent class analysis if you use different random starting seeds.  Latent class analysis often finds a local optimum, which is why it's valuable to use a larger-than-default number of starting points as you start homing in on your final model - and this is true in Lighthouse Studio or in Latent Gold.
answered Feb 22 by Keith Chrzan Platinum Sawtooth Software, Inc. (102,700 points)