www.phidot.org

I am working myself through the RMark manual and found some things that seem to be outdated:

on page C-30 the command

Code: Select all

merge.occasion.data

is used, but that has changed in the package into

Code: Select all

merge_design.covariates

. In the help file in R the correct example code is shown.

On page C-34 the command

Code: Select all

rm(list=collect.model.names(ls()))

is used, but R reports an error message could not find function "collect.model.names". However when typing

Code: Select all

?collect.model.names

, the help file appears as if nothing is wrong.

Yes it is unfortunately some what out of date and in need of my attention when I can get some time. You are correct about merge_design.covariates. The issue with collect.model.names is not a documentation issue. I see that I need to add that function to list of exported functions. When all packages moved to Namespaces, you have to specifically identify which functions are exported and I missed that one. I'll make that change here shortly. Just so you know you can access any non-exported function in a package by typing package:::function (eg, RMark:::collect.model.names).

Good as time as any to reiterate that RMark is on CRAN but I keep the most updated version at https://github.com/jlaake/RMark as well as an archive of old versions. Go to Downloads and select the package binary that you want. The current newest one there is 2.1.1 which is one version above the one on CRAN at 2.1.0. What I do is to accumulate changes for awhile and then post to CRAN to reduce the nuisance level with the CRAN folks. For example, the fix for RDFHet that Darryl reported is in 2.1.1 on github but not on CRAN. I'll make this change shortly and post it there. If many folks do download the most recent version from github, it does mean that there will be different versions floating around with the same number which may get confusing but at this point the changes in each version are quite small.

--jeff

Thanks, that works indeed.

I have another question that is slightly manual-related, so I'll post it here.
In the manual it is explained that RMark creats continuous variables for factors such as cohort, age and time (C-12). What is not really clear from the text is whether one should specify the discrete or continuous variable in the models. It matters greatly for the results. Somehow I have the feeling that is a great conceptual difference between the two and how they are used, but I can't figure out if there is some rule on which one to use.
Thanks in advance.

Use of the continuous version is a linear trend (in the link space) so it only uses a single parameter beyond the intercept. The factor version uses k-1 parameters if there are k levels in the factor variable. Usually the time,cohort factor variables are used but occasionally if a linear trend is appropriate. See the Cooch and White electronic book for a discussion of models.

--jeff