Replicated Poin t Patterns and Designed Experiments 701
16.10.3.2 Complete ly different interactions for different cases
Suppose we now want to fit a model to th e simba2 data which is Poisson in the control group ,
and Strauss in the treatment group. Th e Poisson and Strauss interactions can be given as separate
columns in a hyperframe of interactions:
interaction=hyperframe(po=Poisson(), str=Strauss(0.07))
What do we write for the iformula? The fo llowing will not work:
iformula=
~
ifelse(group=="control", po, str)
This does not work because the Poisson and Strauss models are ‘incompatible’ inside such expres-
sions. The ca nonical sufficient statistics for the Poisson and Strauss processes do not have the same
dimension. Internally in mppm we translate ...