marked package now on phidot

questions concerning analysis/theory using the R package 'marked'

marked package now on phidot

Postby jlaake » Fri Oct 03, 2014 2:42 pm

I have started getting a few questions about the marked R package. Also, I and others have a paper that will be published soon describing a CJS model that estimates survival in the presence of tag loss. Thus I thought it would be wise to create a forum for questions and Evan kindly did so on phidot. I've created an FAQ that answers some introductory questions, so please review it.

Marked is NOT replacement for MARK. My intention is provide a package as an alternative to explore other models and ways of fitting c-r data with source code that the user can explore and use as a learning tool. In particular I have become a huge fan of hidden Markov models, so much so that I wrote this tutorial on fitting c-r models with HMMs. (https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0CCUQFjAA&url=http%3A%2F%2Fwww.afsc.noaa.gov%2FPublications%2FProcRpt%2FPR2013-04.pdf&ei=cukuVOquKsT9yQTb8IDIDA&usg=AFQjCNHdrthQ3uEtU8Q4ARIEx7LP-aj6sA&bvm=bv.76802529,d.aWw).

As part of that exercise I learned ADMB (automatic differentiation model builder) (https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0CB4QFjAA&url=http%3A%2F%2Fadmb-project.org%2F&ei=6OkuVICCD4-KyASbgIGoBQ&usg=AFQjCNGN3hdXEOPeLJpRNYa3R9VmutpvqQ&bvm=bv.76802529,d.aWw) and implemented some models with ADMB so mixed-effect models can be fitted for CJS. Note that I recently had a question about this which was part of the prompting to create this forum. The question was in regard to the dipper data and why the results did not match the results of Maunder et al in the Modelling Demographic Processes in Marked Populations. The reason was that currently you need to specify crossed=TRUE for designs other than simply individual random effect models (eg CJSRandom in MARK) and the model fitted by Maunder et al had a time random effect. Once crossed=TRUE was set the results matched. I'm working on automating that decision. When crossed=FALSE it uses guass-hermite integration and when it is TRUE it uses the Laplace approximation with a sparse hessian.

One issue that I'm currently facing is that with large CJS problems (10000+ capture histories with 27 occasions) and 250+ parameters, ADMB is failing to converge. This is something that I'm currently exploring. So if you are in that situation, don't use use.admb=TRUE and it will use the R/FORTRAN code for model="CJS" or you can use model="hmmCJS" to use the hidden markov algorithm.

The marked package is relatively young and under development. I do have help files and a vignette but I'm way behind on documentation. If you decide to use it, please be patient. As I state in the FAQ, if you are new to mark-recapture I strongly recommend that you use MARK. Also, the number of models in marked is tiny in comparison to what you'll find in MARK. Most are models that I need for my own analysis. I hope others may find them useful as well.

regards --jeff
jlaake
 
Posts: 1008
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Return to analysis help

Who is online

Users browsing this forum: No registered users and 1 guest