Model table in MARK with exported models doesn't match RMark

[jlaake]

I'm posting a question that was sent directly to me because it may come up again. The individual ran a set of models in RMark and then exported them and imported into the MARK interface. They noticed that the AIC and ordering of the models did not match. What they didn't notice what the number of parameters for each model changed. Because the number if parameters differed, the AIC changed and the model order changed.

The default for RMark is to assume the model does not have any confounded parameters and the counted parameters is the number of columns in the design matrix. That is the default setting of adjust=TRUE. If you set adjust=FALSE then it will use the parameter count that MARK gets. Often that parameter count is low because it will not count parameters at boundaries which are poorly estimated due to sparse data. I have done that on purpose because I have seen many situations in which folks fit overly complex models with poorly estimated parameters. I believe it is safer to underfit rather than overfit. This is all laid out in the documentation. When MARK reads in models that RMark exports, it uses the parameter count from the MARK output file.