May MARK count more parameters than are really identifiable?

[simone77]

I have a CMR dataset with two groups, nine occasions (eight intervals).
I ran a fully time varying CJS with sin link and I get 31 parameters estimated as identifiable by MARK, while I would expect to be 30 (15 for each group: s+7ps+1beta).
In my experience, this is the first time MARK estimates more parameters than are really (believe so!) identifiable.
I had a look at the beta and real estimates and the final phis and ps had st.error=0 as it could be expected if they are confounded parameters.

[abreton]

Sounds like you've already discovered this section, but if not have a look at Addendum: counting parameters on page 4-64 in the MARK book. Also, you did not mention whether your model was 'additive'? See page 6-62: "Counting parameters for the additive model..."

andre

[simone77]

Sorry, I should have been more explicit.
The model is {phi(gender*t) p(gender*t)} that for 9 occasions, 8 intervals should have, for each gender:
7 phis + 7 ps + 1 beta
estimable parameters.
Given there are two genders, 30 parameters.
But when I run the model with the sin link the result browser tells me there are 31...
Even though, phi8, phi16, p24 and p32 are clearly non estimable (st error=0).
I have read the part of the GI on the way MARK counts parameters and I don't understand how this situation is possible, that's why I am asking here (sometimes you make something wrong but cannot see it).

Simone