CJS RMark salmonid survival analysis help

[john.lyssenko]

Hi all,

This is my first analysis in RMARK, or any mark-recapture analysis for that matter, and I would like to consult with someone here to help me understand how to construct and choose my best model.

I am interested in finding reach-specific survival for juvenile chinook released from a hatchery to the mouth of the river. I am working with three group variables, "Tagger," "Release Group," and "Release Time," and two individual covariates, "Weight" and "Length."

For my CJS model construction, I chose to use the "nuisance" approach, which I didn't realize was a dangerous way to go about model construction. However, I have kept all model formulas I considered while using the "nuisance" approach and then used the mark.wrapper function to test all combinations, so question number one: is that a sensible way to go about model construction? Considering White et al. 2012 suggested an "all combination" approach.

Here is what it looks like:

Code: Select all

Survival_analysis_aggedCOL=function()
{
#Isolate Phi and try all combination for p
  Phi.constant= list(formula=~1)
  p.constant= list(formula=~1)
  p.time=list(formula=~time)
  p.Stock=list(formula=~Stock)
  p.Weight=list(formula=~Weight)
  p.Length=list(formula=~Length)
  p.releasegroup= list(formula =~ReleaseGroup)
   
  p.time.plus.Stock= list(formula=~time+Stock)
  p.Tagger.plus.time=list(formula=~Tagger+time)
  p.Weight.plus.time=list(formula=~Weight+time)
  p.Length.plus.time=list(formula=~Length+time)
  p.ReleaseGroup.plus.time=list(formula=~ReleaseGroup+time)
   
  p.Stock.x.time=list(formula=~Stock*time)
  p.Stock.x.Tagger=list(formula=~Stock*Tagger)
  p.Stock.x.Weight=list(formula=~Stock*Weight)
  p.Stock.x.Length=list(formula=~Stock*Length)
  p.Stock.x.ReleaseGroup=list(formula=~Stock*ReleaseGroup)
  p.Tagger.x.time=list(formula=~Tagger*time)
  p.Tagger.x.Weight=list(formula=~Tagger*Weight)
  p.Tagger.x.Length=list(formula=~Tagger*Length)
  p.Tagger.x.ReleaseGroup=list(formula=~Tagger*ReleaseGroup)
  p.Weight.x.time=list(formula=~Weight*time)
  p.Weight.x.Length=list(formula=~Weight*Length)
  p.Weight.x.ReleaseGroup=list(formula=~Weight*ReleaseGroup)
  p.Length.x.time=list(formula=~Length*time)
  p.Length.x.ReleaseGroup=list(formula=~Length*ReleaseGroup)
  p.ReleaseGroup.x.time=list(formula=~ReleaseGroup*time)

# Try all combinations for Phi to find overall lowest model for Phi and P
  Phi.Stock=list(formula=~Stock)
  Phi.Tagger=list(formula=~Tagger)
  Phi.Weight=list(formula=~Weight)
  Phi.Length=list(formula=~Length)
  Phi.ReleaseGroup=list(formula=~ReleaseGroup)
  Phi.time=list(formula=~time)

  Phi.Stock.plus.time=list(formula=~Stock+time)
  Phi.Stock.plus.Tagger=list(formula=~Stock+Tagger)
  Phi.Stock.plus.Weight=list(formula=~Stock+Weight)
  Phi.Stock.plus.Length=list(formula=~Stock+Length)
  Phi.Stock.plus.ReleaseGroup=list(formula=~Stock+ReleaseGroup)
  Phi.Tagger.plus.time=list(formula=~Tagger+time)
  Phi.Tagger.plus.Weight=list(formula=~Tagger+Weight)
  Phi.Tagger.plus.Length=list(formula=~Tagger+Length)
  Phi.Tagger.plus.ReleaseGroup=list(formula=~Tagger+ReleaseGroup)
  Phi.Weight.plus.time=list(formula=~Weight+time)
  Phi.Weight.plus.Length=list(formula=~Weight+Length)
  Phi.Weight.plus.ReleaseGroup=list(formula=~Weight+ReleaseGroup)
  Phi.Length.plus.time=list(formula=~Length+time)
  Phi.Length.plus.ReleaseGroup=list(formula=~Length+ReleaseGroup)
  Phi.ReleaseGroup.plus.time=list(formula=~ReleaseGroup+time)

 
  Phi.Stock.x.time=list(formula=~Stock*time)
  Phi.Stock.x.Tagger=list(formula=~Stock*Tagger)
  Phi.Stock.x.Weight=list(formula=~Stock*Weight)
  Phi.Stock.x.Length=list(formula=~Stock*Length)
  Phi.Stock.x.ReleaseGroup=list(formula=~Stock*ReleaseGroup)
  Phi.Tagger.x.time=list(formula=~Tagger*time)
  Phi.Tagger.x.Weight=list(formula=~Tagger*Weight)
  Phi.Tagger.x.Length=list(formula=~Tagger*Length)
  Phi.Tagger.x.ReleaseGroup=list(formula=~Tagger*ReleaseGroup)
  Phi.Weight.x.time=list(formula=~Weight*time)
  Phi.Weight.x.Length=list(formula=~Weight*Length)
  Phi.Weight.x.ReleaseGroup=list(formula=~Weight*ReleaseGroup)
  Phi.Length.x.time=list(formula=~Length*time)
  Phi.Length.x.ReleaseGroup=list(formula=~Length*ReleaseGroup)
  Phi.ReleaseGroup.x.time=list(formula=~ReleaseGroup*time)
 
 
cml=create.model.list("CJS")

mark.wrapper(cml,data=detections_processed_aggedCOL,ddl=detections_ddl_aggedCOL, output =FALSE)
}


Secondly, I have several models within one AIC value. How should I go about interpreting these neighborhood models? I understand model averaging is one way to assess these top models, but I have been stumped on implementing model.average because of difficulties with creating additional design data using find/fill. covariates. In any case, are there statistics I can consider from the model.table and results that can point me to the best model? Any help would be much appreciated! Thank you for your time

[john.lyssenko]

Answered my own question on deviance..

If someone else is curious here is where I found it http://www.phidot.org/software/mark/rmark/RMarkWorkshopNotes.pdf

page 71

[jlaake]

This isn't exactly an RMark question except where you mention a potential problem with model average. You didn't really describe in detail but I'll try to address. You have both design covariates like Tagger and release group and individual covariates like weight. Model average will use the average values for individual covariates and will use the values for the design covariates in the design data, so you should not need to use find and fill like you mentioned. If you want to specify values of individual covariates to use you can do that. See the help file and the workshop notes.

With regard to your model selection issues you may want to consult the books on model selection by Burnham and Anderson and materials in Cooch and White MARK online book