Dear S-Morten,
Quick question regarding your link-function:
According to your problem description, and the model’s premises, it is assumed that your dependent variable has a Poisson distribution. Nevertheless, you have applied a Inverse-Logit linking-function to your Poisson distribution.
The Poisson distribution usually has an Exponent (and its inverse, a Log) linking-function for effectively connecting the linear terms (B*X + A) to the E(Y).
Is there any particular reason for why you are using this other linking-function in your model?
Sincerely,