Marginal effects of interacting variables in Multinomial Conditional Logit (asclogit)

[Lionel Conzett]

Hi there!

I'm interested in the marginal effect of the interaction of two continuous variables in an asclogit: 10 alternatives of which several can be chosen in any given case, dependent variable is a dummy which indicates whether or not an alternative has been chosen, several continuous explanatory variables (some alternative-specific, others case-specific) of which two might interact.

Unfortunately, inteff does not work for asclogit. According to Norton, Wang, and Ai (2004) one could use predictnl, however they state that "the user must be able to write correct formulas of the marginal effects in vector notation" (p. 156) which I am not.

Is there any other way how I can assess the magnitude and significance of the interaction term's marginal effect? Or is it possible to substitute the asclogit with a regular logit (i.e. to run one logit for each alternative)?

Help of any kind is highly appreciated! Also, please let me know if I was not precise enough in the description of my issue - I will clarify if necessary.

[Richard Williams]

Use estat mfx after asclogit. But first read Vince Wiggins' post about why Stata doesn't give you marginal effects for interaction terms:

http://www.stata.com/statalist/archi.../msg00293.html

[Lionel Conzett]

Thank you for your instant comment and the link, Richard. Also, Vince Wiggin´s post and the links he provided were quite helpful. I will use estat mfx as you suggested. But, is there any Stata command which allows me to plot marginal effects at several levels of the interacting variables (e.g. like marginsplot but for estat mfx)?

[daniel klein]

Or is it possible to substitute the asclogit with a regular logit (i.e. to run one logit for each alternative)?

I did not follow the rest closely, but in principle you could replicate your results using clogit (or xtlogit ,fe). Read more how to do this this here.

Best
Daniel