Multiple Comparison of Predictive Margins

[Makoto Okawara]

Dear all Statalists,

I am using the following model in Stata 16.1:

nbreg y i.x1 i.x2 (other covariates)

x1 is 3 categorical variables (A,B,C).
x2 is 12 categorical variables (1-12).
I am not interested in the interaction between x1 and x2. Therefore, I will not include an interaction term in the model.
What I want to know is whether there is a difference in the Predictive Margins of x1, at each cases of x2.

First, I used the margins command to calculate the Predictive Margins.
I calculated the Predictive Margins at each levels of x1 and x2 (3*12=36 different Predictive Margins).
Substitute values as observed for all but x1 and x2.

margins, over(x1 x2) mcompare(bonferroni)
marginsplot, plotdimention(x1)

Looking at the plot, there seems to be a significant difference between A vs B and A vs C of x1 in some cases of x2.

Q: What is the command to test if there is actually a difference in A vs B and A vs C of x1 for each x2 (12*2=24 ways)?

Following commands indicated the different results to my thoughts looking marginsplot.

margins r.x1,over(x2) contrast(effects) mcompare(bonferroni)

Sincerely,

[Makoto Okawara]

Please let me know if there is any information missing to answer my question.
English is not my first language, so I apologize if any of this sounds strange.

[Joseph Coveney]

I am not interested in the interaction between x1 and x2. Therefore, I will not include an interaction term in the model.
What I want to know is whether there is a difference in the Predictive Margins of x1, at each cases of x2.

Those two statements seem mutually incompatible to me.

If you want to be able to test for a difference between levels of the first predictor separately at each level of the second predictor, wouldn't you need to include the interaction term?

[Makoto Okawara]

I appreciate your comments, which have helped me organize my thoughts and lead to better analysis.