Re: st: marginal effects with interaction term gologit2
Date
Wed, 01 May 2013 10:25:55 -0500
This is an annoying limitation of gologit2. Unfortunately every time
I try to change it I run into problems. Factor variables and the
techniques gologit2 uses to impose proportionality constraints don't
get along too well together.
You might check out the "do it yourself marginal effects" approach
outlined on slides 28-30 of
I haven't tried it, but I think you would just have to add a few more
lines to also compute the interaction terms when dummy = 0 and dummy = 1.
You may also be able to do things more directly with the adjust or
margins command. It is pretty easy when the dummy = 0. Just set both
the dummy and the interaction to 0. When dummy = 1, it is probably
harder, although perhaps not so hard if you are willing to use MEMs
rather than AMEs.
Whatever you do -- to make sure you did it right, try running a model
with ologit and then run the same model in gologit2 using the -pl-
option. If the marginal effects are the same then hopefully you have
done it right.
At 08:12 AM 5/1/2013, cstrobach@aol.com wrote:
Dear Statalister,
I am using the command gologit2 to estimate a generalized ordered
logit model. My dependent variable has three outcomes. My
independent variables are a continuous variable(c), a dummy (D), an
interaction term of both and some other control variables. As
gologit2 does not allow the use of factor variables, I created the
interaction term as c*D. Now I would like to obtain marginal effects
using margeff, however this will not provide meaningful results with
respect to the interaction term.
