# st: -margeff- question (interaction effects, -at- and -dummies- option)

 From Andrea Bennett To statalist@hsphsun2.harvard.edu Subject st: -margeff- question (interaction effects, -at- and -dummies- option) Date Tue, 22 Jul 2008 19:10:15 +0200

Dear all,

I wonder if anyone knows the following related issues regarding the - margeff- command:

1. standard interaction effects
Say, I have -(probit) Y = b0 + female + income + age_cat2 + age_cat3 + age_cat2*radio + age_cat3*radio + radio + controls + e-. After applying -margeff- with the dummies option -dummies(female age_cat2 + age_cat3 + age_cat2*radio + age_cat3*radio)-, how do I correctly state the interaction effect? Is it additive, e.g. if age_cat2=.3 and age_cat2*radio= -.1 and radio = -.5, is the marginal effect of age_cat2 0.3-0.1=0.2 at the mean of the variable radio when radio is present and 0.3 when there radio is zero (radio is 0-1 in percentages; e.g. 0.04, ..., 0.23)?

Now I wanted to check what happens when I apply a different value for the variable radio. But when using interaction terms it seems I cannot change the value of the variable radio. I also tried to include the interaction terms under the -at- option and set the values equal to each other (since age_cat* is (0/1) binary coded the interaction dummy should be the same as the value of the variable radio). I always get the error code r(111) with "Invalid at() option: .31179443is not found" or some other number depending on the number of control variables I have in my regression. Why can't I do that, or better how can I?

3. the dummies option
Is it correct to include the interactions, e.g. age_cat2*radio, as dummies in the dummies option?

If anyone could share his or her experience, that would help me tremendously!