Have an idea?

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

Overall Model Fit for Latent Class

I am running a latent class segmentation on a MaxDiff experiment.  How do you determine the fit statistic for the segmentation model?  I have seen other segmentation methods produce a statistic like 85% fit.  Does LC have a similar stat?
asked Feb 26, 2013 by anonymous

1 Answer

+1 vote
There are several measures of fit in LC, most of which you will find described on page 28 of the LC manual available here:  https://www.sawtoothsoftware.com/download/ssiweb/LClass_Manual.pdf.
answered Feb 26, 2013 by Keith Chrzan Platinum Sawtooth Software, Inc. (92,075 points)
In a max diff study the null log-likelihood is much larger than in the example given on page 28.  How is this calculated, also what is a good measure for percent certainty in a MaxDiff study.
Percent Certainty is probably what you are looking for, as it is scaled from 0 to 100 and acts like a pseudo-rsquared.
Regarding Null log-likelihood, it is computed by taking ln(1/C)*R*T*2


C = concepts per task
R = number of respondents
T = number of tasks per respondent

(Where T is the number of tasks that each responent saw, which get coded in the estimation file as two tasks per set...a "best" and a "worst" task...hence I'm multiplying by 2.)