st: margins after xtnbreg, does it works?

[JUAN CRESPO CEPAS <Juan.Crespo-cepas@ut-capitole.fr>]

Dear all,

I have some problems when using and interpreting the results of the  
-margins- command after a negative binomial regression with fixed  
effects (-xtnbreg, fe-). I run the following model in Stata 11  
(windows):

.xtnbreg y er ir dd c.dd#c.dd, i(nb) fe

Then, I calculate the marginal effect of the variable dd when all the  
other (continuous) variables are at their mean value:

.margins, dydx(dd) atmeans

However, the value of the marginal effect is the same than the  
coefficient dd of the regression. I find this puzzling, and when I try  
to do the same with non panel data (-nbreg- and -margins-) the  
coefficient and the marginal effect are not the same. Moreover, if I  
give a particular value (different to the mean) to the other  
variables, for instance -er=1-:

.margins, dydx(dd) at(er=(1)) atmeans

I get the same result again. I find this puzzling too. This is not  
what I would expect since I am using a non linear model. Any insights  
about what I am doing wrong?

Thanks for your consideration,
Joan

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