Hi Statalisters,

When I use count data model to analyze my panel data with fixed effects, like -xtpoisson, fe-, I have questions about how to address my interaction effects.

In my project, The dependent variable is the number of patents (npatent) and independent variables include customer concentration (CC), customer dependence (CD), and supplier-customer years of linkage (Linkage).

The full xtpoisson model is like: npatent = t*exp(β0 + β1*CC + β2*CD + β3*Linkage + β4*CC*Linkage + β5*CD*LInkage).

I could use
Code:
xtpoisson npatent CC CD Linkage c.CC#c.Linkage c.CD#c.Linkage $controls i.SIC i.fyear, fe
to have main regression results.

However, I want to plot interaction effects that manifest how Linkage moderates CC and CD. Therefore, I encounter two problems:

1. I have read previous threads saying that margins after -xtpoisson, fe- produces meaningless results. This makes me confused about how to properly express interaction effects via plots.

2. As a trial, I use, for example, the below code to plot interaction term CC*Linkage.
Code:
xtpoisson npatent CC CD Linkage c.CC#c.Linkage c.CD#c.Linkage $controls i.SIC i.fyear, fe
quietly margins, at(CC=(0(0.05)0.3) Linkage=(1(2)11)) vsquish
marginsplot
The generate plot is kind of weird because y axis for the number of patents is negative. I suppose at least it should indicate positive numbers. I am not sure if this problem arises from the -xtpoisson, fe- -margins- codes or my data.

Array

Appreciate guys for any suggestion. Thanks