RMark and MARK discrepancies

[jlaufenb]

I have constructed an a priori set of 7 models for a FullHet analysis of >1 data sets in RMark. Because I am more confident using the MARK design matrix, I check my RMark code with MARK results for a single data set. For this set of models, AICc, npar, deviance, etc match except for 1 model. This particular model is the only 1 that has group specific pi's. So far, I've found that npar is correct, but lnl, deviance and AICc do not match the results from MARK. Obviously, this changes model weights and subsequent model averaging results. Why does this only happen for a model with group specific mixtures? Which results table is correct?

Thanks
Jared

[jlaake]

Jared-

Did you compare the design matrices between the 2 models? Are you using same link function? There are only 3 ways to get disparities: 1) different data which is ruled out in this case, 2) different link functions, or 3) different or incorrect design matrices.

Check the output files from RMark run and the MARK run and compare the differences in the DMs. If you can't work it out send them to me offlist.

--jeff