st: why Margins gives "not estimable" ?

[Francesco <cariboupad@gmx.fr>]

Dear Statalist,

I have a simple, yet puzzling question (at least for me) about -margins.

I run a fixed effect regression and I would like to understand the
effect of a dummy variable (DUM) interacted with a categorical
variable (CAT)
CAT can take 5 values (1, 2, 3, 4 and 5) lets say.

When I run

-xtreg Y 1b.CAT##DUM, vce(cluster id)

I get the results
and if I type

-margins CAT, dydx(DUM)
I get by how much I increase the predicted values of the output Y when
DUM=1 and according to the values of CAT.
So far, so good.

However, if in the original regression I include another categorical
variable : CAT2 (7 possible values)

-xtreg Y 1b.CAT##DUM i.CAT2, vce(cluster id)

then
-margins CAT, dydx(DUM)
or
-margins CAT, dydx(DUM) at(CAT2="a particular value of CAT2)
or even -margins CAT, dydx(DUM) over(CAT2)

they all return "not estimable"...

Do you have an idea ? I am probably missing some important option or
syntax in the margins command ?

Many thanks
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/