I read one article to identify effects of policies. The article was published from a high IF journal in my field.

According to the study,
they might use fixed effect term in xtgee like this.

Code:
tsset panel_id time
xtgee depvar time category indepvar i.panel_id,  link(identity) family(gaussian) corr(ar1)

I tried to do similar things in my research, but my senior econometrics expert advised me that the model was wrong.
He said that the panel_id has already been calculated as a random effect in xtgee, so using panel_id LSDV in GEE might lead a biased inference.

However, I would like to use fixed effect term in xtgee, because my data is unbalanced, and contains multi-level structured variables.
Here is an example of my data structure. Each Panel_ID has multiple same Time because of Category variable.
Panel_ID Time Category Indepvar Dep
1 1 1 ### ##
1 1 2 ### ##
1 2 1 ### ##
1 3 1 ### ##
2 1 1 ### ##
2 2 4 ### ##
2 2 5 ### ##
2 3 1 ### ##
…

One of my colleague recommended me to create a new panel id variable by combining Panel_ID and Category variable, then to use the new panel id variable as fixed effect term.
I agree with him and that's what I've done in other studies, but one of my purpose of research is to identify effects of Category variable on the Dependent variable.
One of my solution is to use just xtgee, but I had to consider unobserved heterogeneity.

Please recommend me what I should do in this situation.

Thank you.

(After all, fortunately, my results were robust whether or not the fixed effect term was included in the model.)