Different results between Mark and RMark in confounded param

[P. WALTER]

Hi there

I'm trying to learn to use RMark, and I've been fiddling a lot comparing Mark and RMark to check that I'm doing everything allright and that results match. I know that RMark doesn't estimate parameters, but I've been consistently getting different results for non-estimable values (such as confounded parameters (last p and last Phi in time-dependent models).

For example, for the next script, I consistently get for the last p a value of 0.16 in rMark while in Mark I get 0.4. I know is not really important, as it is not estimable and the value is not trustable, but, as RMark calls Mark for making the calculation, I would expect the same value. Is there an explanation behind this? Am I making any mistake?

Code: Select all

imported.pool <- convert.inp("Pop.inp", group.df=data.frame(sex=c("Female","Male")))
imported.pool$sex <- as.factor(imported.pool$sex)
pool.process <- process.data(imported.pool, model="CJS", time.intervals = c(4,1.429,2.143,0.714,0.857,1,1.286,1), groups = "sex")
pool.ddl = make.design.data(pool.process)
Phi.time = list(formula=~ time, link = "logit")
p.time = list(formula=~ time, link = "logit")

model.pool = mark(pool.process,pool.ddl, model.parameters = list(Phi=Phi.dot,p=p.dot,pent=pent.time,N=N.sex))

Thank you very much!

[jlaake]

Did you read the fourth sticky note at the top of the RMark list? My guess is that you didn't specify the logit link with MARK. Another possibility is the design matrix that was used. Please read all of the sticky notes.

[P. WALTER]

Hi there.
Thanks for the heads up, it was not the data that was different, nor the link function, it was just the DM.
I was using for the comparison only the PIM charts in Mark, as I was testing very simple models. It seems that the results in RMark match much better the parameters estimated with models designed with the design matrix. I didn't expect such a difference.

Sorry for the rookie post.

Cheers.