Significance of interactions in multinomial logit

[Hendrik Bruns]

Dear Stata-Users,

I have a question concerning the interpretation of interaction effects in multinomial (and ultimately also binary) logit models.
I read the Stata Tip 87 by Maarten Buis, so I use relative risk ratios for interpretation and generate interaction effects via "mlogit y x1##x2 ui, robust rrr base(2)" (where ui are other variables). However, I know of no tip or post by Mr. Buis that he states that the z-values for the interaction effect posted by Stata can be interpreted. I am under the impression that z-values for interaction terms in nonlinear models cannot be interpreted (therefore, one would need to use -inteff-).
I assume that reporting relative risk ratios (or odds ratios) wouldn't really make a difference because that obviously doesn't affect the z-values.

Is it possible to "transform" the multinomial variable into two binary variables, with the respective base category and check via -inteff- if the interaction effects are significant? If they are significant I could just interpret them in the model with the risk rate ratios?

Thank you for input,
uk013

[Maarten Buis]

You now have a post by me that states you can use the z-values to test your hypotheses. The dificult part of interactions in these models the correct interpretation of the results. Once you have done that, you can formulate the correct null hypothesis, and look at the z-values.

We have a long tradition on this list that we post under our real name. We believe that that has contributed to the generally friendly and profession atmosphere on this list. You can help maintain that tradition by changing your login name by going to the "contact us" button at the bottom right of this page.

[daniel klein]

Maarten might want to answer this one again, but in the meanwhile see this discussion.

Best
Daniel

[Hendrik Bruns]

Thank you both for your quick answer.
It seems I did not search well enough for similar questions that had already been asked, sorry for that.

I contacted the administrator of the forum. I got into the habit of never using my real name on forums, that is why I did not think about that when I registered.

Best regards,
Hendrik