RMark - 1.9.3

[jlaake]

A new version of RMark with some bug fixes was posted today. This fixed some errant code wherein the confidence intervals computed in RMark (not in MARK) for real parameters were incorrect. The real estimates and standard errors were fine and so was model selection. However, if your previous analysis used confidence intervals for real parameters and you were either specifying a chat > 1 (intervals used c rather than sqrt(c)) or were using a non-logit link then you'll want to re-run the prediction code -- anything that used compute.real or covariate.predictions. Note that get.real and model.average use compute.real. There is no reason to re-run the models as the error only affected computation of the confidence intervals for real parameters which were computed in R. Also, if you were only getting intervals from summary without a c-hat adjustment then the reported values would have been obtained from MARK and would be ok.

My apologies for any problems this may have caused and my thanks to Carl Schwarz and Dave Hewitt for their reports and for Carl's detective work.

For those that may not know, Evan posted a pdf of some 60+ pages of notes that I put together for an RMark workshop that I gave last winter. You can find the link on the RMark page. I've had several reports that the notes were helpful.

For those of you using RMark on Macs/Linux, be aware that Bret Collier has reported that some compilers are tripping up on the FORTRAN code
that I added for the js and cjs functions. That code is not needed to interact with MARK, so prior to building the R package, delete all files under the src directory and remove the library.dynam line in R\zzz.r. Arpat reported that this worked to get the package functional on his Mac. Note that there is no Mac version of MARK and no plans for one.


regards --jeff

[jlaake]

My last posting implied that model.average could have been affected by the bug that was fixed in compute.real. It did not because model.average only used the real estimates and their var-cov matrix which were fine. Intervals computed from model.average will not change with this new version. The only intervals that could change would be those obtained from summary, compute.real, get.real or covariate.predictions.

--jeff