plotting interaction for xtpoisson

[John Aunna]

Hi Statalist fellow,
I have run an analysis but I don't trust my result. I would be extremely appreciated if you could help in this matter by checking my method.
I run an xtpoisson model with interaction term and I need to plot the interactions.

xtpoisson count a##b, fe vce(robust) irr

margins a#b
marginsplot, ytitle('IRR of count)


The issue is that there is no consistency between the xtpoisson IRR coefficients and the result of margins.
xtpoisson report:

IRR =1.38 (p=0.001) for a
IRR = 1.80 (p=0.001) for b
IRR =0.85 (p=0.001) for a#b

margins report:
Linear prediction, predict()
a#b
0 0 0.49
0 1 1.092
1 0 0.65
1 1 1.093


Can you please help to understand what's my mistake and how should I approach this issue?
Thanks,
John

[Phil Bromiley]

If you read the xtpoisson postestimation documentation for margins, you'll see that the default prediction is xb, not irr.