Have an idea?

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

Interpretation of Latent Class results

Bryan,

Thank you very much for your answer.

I analyzed my CBC data (n=120) running the Latent Class Analysis from Lighthouse Studio. I investigated the 2-group, 3-group, 4-group and 5-group solutions and here are my results:

       PctCert   CAIC    CHISquared     RelChiSquared
2    10.21        5097     564                     8.69
3    12.87        5015     711                     7.26
4    15.48        4937     856                     6.54
5    16.60        4939     920                     5.60

Segment Sizes

2 78%/21%
3 57%/19%/22%
4 32%/14%/15%/36%
5 12%/13%/29%/18%/25%

I cannot find any useful hint whether we should group respondents together or not. Can you help us?

Thank you in advance!
related to an answer for: CBC analysis based on groups
asked May 9, 2016 by Joanna

1 Answer

0 votes
Hi, Joanna,

CAIC minimizes at 4 segments (though we don't see what happens at 6+ segments ans we don't see other fit statistics like BIC) so this provides at least some evidence that 4 segments may be a better solution than 3 or 5.  

With an average of just 30 respondents per segment (and with one segment as small as just 17 respondents) it's not clear to me why you want to run LC:  your between-segment comparisons on any variables other than utilities will be extremely underpowered.  

Perhaps you have some theory about why LC will be of value?  For example, do you have a theory that would lead you to expect discrete latent segments rather than the continuous heterogeneity HB would provide?
answered May 10, 2016 by Keith Chrzan Platinum Sawtooth Software, Inc. (74,925 points)
Interpretation of LC results
...