[Jeff Wooldridge]

Your margins command in #15 isn't what you want. But if you're interested in whether the treatment effect varies across country, those differences are estimated by the coefficients on the interactions of the country dummies and patient_anticoagulant_med. So, for example, the effect is about 13.5% lower for country 2 compared with country 0 (using Poisson). The effect for country 1 is about 4.4% higher than for country 0. But neither is statistically different from zero.

[ahmed farhan]

thank you, that makes sense.
if i am looking to set level 0#yes as base and compare other interaction results relatively, then i presume my margin command should incorporate this in some way and set y=0 for 0#yes subgroup.. just not sure how to go about this? i have read the margins stata chapter and have tried command margins, at(country#patient_anticoagulant_med=_b[1._at#0.country]) however stata gives me an error saying "invalid at() specification;
levels of interactions not allowed". so i presume this is not possible?

here are the results for command margins country#patient_anticoagulant_med


Adjusted predictions Number of obs = 9,813
Model VCE : Robust

Expression : Predicted number of events, predict()


Delta-method
Margin Std. Err. z P>z [95% Conf. Interval]

country#patient_anticoagulant_med
0#Yes 48.64312 1.019292 47.72 0.000 46.64534 50.64089
0#No 41.02432 .4462822 91.92 0.000 40.14962 41.89902
1#Yes 49.398 3.636627 13.58 0.000 42.27034 56.52566
1#No 39.85951 1.056206 37.74 0.000 37.78938 41.92963
2#Yes 49.02358 4.131469 11.87 0.000 40.92605 57.12111
2#No 47.30465 2.168686 21.81 0.000 43.0541 51.5552
3#Yes 91.57066 6.618418 13.84 0.000 78.5988 104.5425
3#No 74.78761 2.94261 25.42 0.000 69.0202 80.55501