Have an idea?

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

Reporting 2LL Interaction Effects

Hi there,

could you please explain how to report significant interaction effects?
I am looking for something like e.g. F(2, 100)=146.000, p = 0.0004), but I can only find Parameters in Model, Log-Likelihood Fit, Chi Square Value, 2LL P-Value for Interaction Effect and Gain in Pct. Cert. over Main Effects in the report. So how would you report a significant interaction effect drawing on the 2LL test?
χ² (?) = ?, p = 2LL P-Value for Interaction Effect

Furthermore, I am wondering if the values in the worksheet "charts" are p-values or part-worth utilities? They look like PWUs to me, since some of the values have negative-signs.
So how do I know between which groups the significant differences exist or can I just make a statement about the significantly greater likelihood of the model with the interaction effects based on the 2LL-test?

Thanks very much for your support!

asked Apr 24, 2019 by Chris Berlin Bronze (570 points)
edited Apr 24, 2019 by Chris Berlin

1 Answer

+1 vote

The -2LL test is a chi-squared test, so you can report the chi-squared statistic.  The number of d.f. is the same as the number of added parameters in the model from your interaction.  The easy way to do this if your two interacting attributes have a and b levels, for example is that the number of added parameters is (a-1)(b-1).  So if you had a 5 and a 3 level attribute, you'd have (5-1)(3-1) = 4 x 2 = 8 additional parameters, the d.f. for your chi-squared statistic.

I'n not sure what values in charts you're talking about, so please clarify and ask as a new question so someone doesn't think from my response that all of your questions have been answered.
answered Apr 24, 2019 by Keith Chrzan Platinum Sawtooth Software, Inc. (85,325 points)
Thanks very much Keith.