Robust Design Gamma Parameters

[benvw]

I am trying to model movement between observable and unobservable states using the gamma parameters in a robust design model. I understand the basic concepts behind the alternative models ("no movement", "random emigration", "Markovian movement", "even flow") as explained in Cooch and White, but I'm not confident that I'm coding them correctly in RMark. The code I've included below is based on the ?robust example. In particular, I'm not sure how to set gamma''=(1-gamma') in the "even flow" model. Any help on how to code these alternative models would be appreciated.

Ben VanderWeide
Division of Biology
Kansas State University

[code]
# "No Emigration" - set Gamma'' = Gamma' = zero
GammaDoublePrime.GammaPrime.nomove=list(GammaDoublePrime=list(formula=~1,fixed=0,share=TRUE), GammaPrime=list(formula=~1,fixed=0,share=TRUE))

#"Random Emigration" model - set Gamma'' = Gamma'
GammaDoublePrime.random=list(formula=~1,share=TRUE)

# Markov constant emigration rate
GammaDoublePrime.GammaPrime.markov=list(GammaDoublePrime=list(formula=~1),GammaPrime=list(formula=~1))
[\code]

[jlaake]

Ben-

On the first one it should work as either one shown below. The share= is only used on the GammaDoublePrime.

[code]
# "No Emigration" - set Gamma'' = Gamma' = zero
GammaDoublePrime.GammaPrime.nomove=list(GammaDoublePrime=list(formula=~1,fixed=0,share=TRUE))
GammaDoublePrime.GammaPrime.nomove=list(GammaDoublePrime=list(formula=~1,fixed=0), GammaPrime=list(formula=~1,fixed=0))
[\code]

Your random and constant ones are fine as they are. I'm not familiar with the even flow model. Show me how you would code that in a DM.

--jeff

[cooch]

Ben-

On the first one it should work as either one shown below. The share= is only used on the GammaDoublePrime.

[code]
# "No Emigration" - set Gamma'' = Gamma' = zero
GammaDoublePrime.GammaPrime.nomove=list(GammaDoublePrime=list(formula=~1,fixed=0,share=TRUE))
GammaDoublePrime.GammaPrime.nomove=list(GammaDoublePrime=list(formula=~1,fixed=0), GammaPrime=list(formula=~1,fixed=0))
[\code]

Your random and constant ones are fine as they are. I'm not familiar with the even flow model. Show me how you would code that in a DM.

--jeff

Chapter 15 -- sidebar beginning bottom of p. 35. Tricky DM -- requires a 'trick' (meaning, something a bit out of the ordinary).

I should note that there are only very specific circumstances where an even flow model makes any biological sense. If there is no a priori biological motivation for fitting the model, doing so might be evidence of some degree of 'data dredging'.

[jlaake]

If I understand the material Evan referenced, then you need to switch to the MSCRD model with an observable and unobservable state and set the substract.stratum values for the Psi parameters. That makes sense. I couldn't work out how it could be done with a DM but what is suggested would work in RMark.

--jeff