Have an idea?

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

Test significance total attribute (not level)

Hi everyone,

I ran a LC analysis and thus have access to the t-test ratios for all attribute levels via sawtooth. My supervisor asked me to do another test to test if a whole attribute is also significant (can not conclude if one or more levels are significant, the attribute is significant)

 She said I have to do a F-test but I really don't get it.
I have the segmentmembership and individual utilities per respondent, but not a mean per attribute or something like that, only attribute importance scores.
So my question is how can I do this (in SPSS)? I think I have to do an one-way ANOVA, but what is the dependent and the factor variable?

Thanks a lot!
asked Dec 30, 2017 by anonymous

1 Answer

0 votes
MNL enables you to run a very simple test called the -2 log likelihood test.

First run the logit model both with and without the attribute in question, and you'll get two log likelihood statistics.  To perform the test subtract the one log likelihood from the other and then double the difference.  Compare this against a critical chi-squared value from a chi-squared distribution with degrees of freedom equal to the number of levels in your attribute minus 1.  

If twice the difference in log likelihoods exceeds the critical chi-squared value then you've a significant result and if not then not.

The above is for an aggregate logit analysis; if you have more than one latent class you'll need to adjust your degrees of freedom accordingly.
answered Dec 30, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (73,500 points)