Hi Statalisters,

My current project is using count data model to analyze panel data with fixed effects, such as -xtpoisson, fe-.

I intend to use -margins- to plot interaction effects as my previous projects. However, I have read other threads about how meaningless it is in using -margins- after -xtpoisson, fe-. This makes me confused about how to analyze interaction effects for count data model.

I did try -margins- after -xtpoisson, fe-, but as predicted, results are all insignificant.
For example,
Code:
xtpoisson depvar indvar1 indvar2 indvar3 c.indvar1#c.indvar3 c.indvar2#c.indvar3 i.SIC i.year, fe
margins, at(indvar1=(0(0.2)1) indvar3=(1(2)11)) vsquish
marginsplot
The above -margins- created a table of predicted number of response. However, one issue is all coefficients of margins are insignificant. The other issue is predicted number of response are all negative, which should not happen for count data (counts should be at least 0).

Appreciate if anybody can help me. Thanks in advance.