Plotting interaction effects in FE panel models: margins command returns "not estimable"

[Tatevik Harutyunyan]

Dear All,

I intend to plot interaction effects using marginsplot. I am using StataSE 18.
Here are some specificties of my variables and models:

• the dataset is longitudinal
• observations are on firm-year level
• in the dataset, I have indicators of the firm growth, variables of the organization environemnt (capital intensity, high-tech industry, volatility of the industry etc), dummies on education and experience of firm's human resources (e.g. the firm has human reosurces with graduate degree, with prior industry experience etc), as well as financial indicators of the firm, among other variables.
• the purpose of the study is to look into how the experiences of human resources can affect firm growth under dufferent environmental conditions.
• the two interacting variables are float.
• The denotation of x2 stands for combination of two experiences of human resources (comboexp; 4 different combinations taking the values of 1 to 4 of the variable).
• m2 is a moderator, and the denotation of m2 stands for the volatility of environment where the firm operates (vol; two categories: 0-not volatile, 1-volatile).
• this moderator m2 is measured once for each firm (at the year of the firm registration), so is omitted from the FE panel regression results.

Here is what code I ran:

tsset firm_id year
xtreg $y1 $controls_1 i.$x2 i.$x4 i.$m2 $x2#i.$m2 $x4#i.$m2, fe robust
margins, at (comboexp=(1 2 3 4) vol=(0 1))


After running these codes, I get a table where every line of interaction values (from 1 to 8) states:
.(not estimable)


So, I do not even reach the point of running marginsplot to plot the interaction effects of experiences and volatility in the environment.

I wonder why marghins are "not estimable" and how i can resolve the issue and ultimately, plot the interaction effects.
I would appreciate your guidance in this regard!

Thank you,
Best,
Tatevik

[George Ford]

if the coef on vol is not estimated, then how can you get a margin on it?

[Tatevik Harutyunyan]

Hi George,

That is what I was trying to understand.
I am not familiar with the technicalities of estimating margins.

While vol is omitted from FE panel models, the interaction terms yield results and I have their coefficients.

Best,
Tatevik

[daniel klein]

See this thread. You must not include the (omitted) main effects in the xtreg specification.

[Tatevik Harutyunyan]

I appreciate your help, Daniel!
The thread was informative.