model averaging of N in model="Closed"

[mwegan]

I have built some very simplistic closed capture models in RMark to work towards some more complex analyses. In checking my work, I found that I am having issues with model averaging.

I ran three closed capture models to estimate p, c, and N in R and then used the same data set in Program MARK. All of the model averaged estimates are equal between the two methods, except for N. The model averaged estimate in R is less than half of the true model averaged estimate. I have RTFM and then RTFM, again. I can't figure out why every other estimate is equal between the two methods. N is estimated as 58, 46, and 46 in the individual models in both R and MARK, so why would R return a model averaged estimate of 23, which is lower than any of the individual model estimates?

If this should happen to catch anyone's interest to the point that they'd like to see the .inp and my MARK files, I'd be glad to share them.


Here's the code:

Code: Select all

library(RMark)
test2<-convert.inp("C:/Documents and Settings/weganm/My Documents/tests/test2.inp")

#p.c.N
constant=mark(test2,model="Closed")
#estimate of p for final interval of time dependent models
p=constant$results$real$estimate[1]

cdot=list(formula=~1)
ctime=list(formula=~time)
pdot=list(formula=~1)
ptime.fixed=list(formula=~time,fixed=list(time=7,value=p))
Nhat=list(formula=~1)

c.ptN=mark(test2,model="Closed",model.parameters=list(c=cdot,p=ptime.fixed,N=Nhat))
ctptN=mark(test2,model="Closed",model.parameters=list(c=ctime,p=ptime.fixed,N=Nhat))

m.r.models<-collect.models()
m.r.models

p.avg<-model.average(m.r.models,"p",vcv=T)
p.avg$estimates$estimate

c.avg<-model.average(m.r.models,"c",vcv=T)
c.avg$estimates$estimate

n.avg<-model.average(m.r.models,"N",vcv=T)
n.avg$estimates$estimate



[jlaake]

While Phidot was down I sent an email offlist that explained this difference to Mike. The parameter N is really f0 the number not caught and that is what RMark is averaging. By adding Mt+1 to that estimate you'll get what MARK produces. The code in RMark is generic for any parameter and there is no specific code to handle handing Mt+1 for N.

--jeff