# How do you stratify a Poisson regression in GLM?

I would like to obtain a stratified baseline hazard in a Poisson regression model. What is the correct way to do it ?

Let A (=0/1) be the binary covariate on which I wish to stratify my baseline hazard, is it sufficient to write (in lme4, for example) :

glmer(event ~ interval*A + (1 + A|C), data=data,family=poisson,
offset=log(time))


In this case, I got an estimate for a fixed coefficient for A0, but that does not interest me. It suggests to me that something is wrong with my formulation.

Any help ?

Thanks :-)

• If i'm understanding this correctly, you want to remove the main effect of your variable $A$. You can simply add $-A$ in your formula or write $\text{intercept}:A$ if you just want the interaction. The notation $A * B$ is equivalent to $A + B + A:B$ in R. – winperikle Mar 15 at 13:58
• It's exact, I want to remove the main effect of my variable $A$. I tried to add $-A$ in the formula but it did not change anything. As well as writing $interval:A$ which gave me only the estimation in one level of A. – Flora Grappelli Mar 15 at 14:12
• If $A$ is truly zero-one valued, have you included it as a factor? Also add $0$ in the glm formula to prevent overparametrization. – winperikle Mar 15 at 14:30
• $A$ is not included as a fixed factor in the model. It is used as an interaction term with a factor C. When I add $0$ and $-A$ in the model the estimated coefficient are the same than with interval*A + (1 + A|C) but the labels of two coefficients have changed ! (Intercept) -7.82887 in interval0 -7.668470 and A0 -0.59597 in interval0:A -0.598916 – Flora Grappelli Mar 15 at 14:45