Test Significance Between Estimates- survival between areas

[jlaake]

There are too many model variants for me to keep in my head so I'd have to look up how f is handled but I know that Phi is an interval parameter so if you are setting p=0 for year y+1 you should set Phi=1 for year y. Right now you have them both set for the same years (times). Look at the design data for Phi and f. Times for interval parameters (action happens between occasions) is the beginning time of the interval whereas occasion parameters are labelled with the time of the occasion. I think f is interval but I can't remember for sure.

Now it looked like you simulated from a constant model and if you fit a constant model then you have a whole different story. Like I said, I've not done this with Jolly-Seber type models and I've never seen it done. Your original question was for a CJS style.

--jeff

[ctlamb]

Good call, Jeff. Yes they are both interval parameters, so y+1 for p and y for Phi and f.

Hmmm, okay, so perhaps this approach isn't appropriate for JS? Any thoughts on who I could ask? I can't seem to recreate the simulated estimates with this approach, even after the y+1 and y change so perhaps that is an indication. It would be much more elegant to do my analysis this was though.

Thanks for all your help on this, i really appreciate it.

CL

[ctlamb]

Folks, Thought I should close on this as Jeff helped me fix this problem, but it is not fully documented here.

I wanted to compute demographic parameters for two distinct populations of animals, sampled during similar time frames, with some non-overlap.

For Example, population A was sampled in 2005, 2010-2014

While population B was sampled in 2006, 2010-2014





I wanted to pool the data into one capture history for these populations because:

1) I wanted to test for differences in apparent recruitment and survival between the two populations using AIC,

2) I wanted to pool data to increase precision in detection probability, as detection is influenced by the same processes in both populations.

I used Pradel open models for this, specifically in RMArk language:

Code: Select all

 model="RDPdfHuggins"

I padded the capture histories for each population with 0's where the other population received sampling. For example, Pop A had 4 sessions in 2005, while Pop B was not sampled during this period, so I added 4 0's to the beginning of every capture history for Pop B. Likewise, Pop B had 4 sessions in 2006 while Pop A was not sampled, so I added 4 0's, at 5th position in the the Pop A capture history (5th position because the first 4 occasions are the Pop A 2005 sampling)



My processing call looked as such:

Code: Select all

NEWDF.proc=process.data(ch, model="RDPdfHuggins",groups=(c("Pop")), begin.time=c(2005),time.intervals=c(0,0,0,0,1,0,0,0,4,0,0,0,1,0,0,0,0,0,0,0,1,0,0,0,0,0,0,1,0,0,0,0,0,0,1,0,0,0,0,0))


Where the start time is 2005 for both studies, the Pop (A and B) are distinct groups, and the 4 year gap between 2006-2010 is accommodated.



FINALLY, to assess the lack of sampling Pop A in 2006 and Pop B in 2005, I did the following:

Code: Select all

###FIX PARAMETERS
##p
NEWDF.ddl$p$fix=NA
NEWDF.ddl$p$fix[NEWDF.ddl$p$Pop=="A"&NEWDF.ddl$p$session=="2006"]=0
NEWDF.ddl$p$fix[NEWDF.ddl$p$Pop=="B"&NEWDF.ddl$p$session=="2005"]=0

NEWDF.ddl$c$fix=NA
NEWDF.ddl$c$fix[NEWDF.ddl$c$Pop=="A"&NEWDF.ddl$c$session=="2006"]=0
NEWDF.ddl$c$fix[NEWDF.ddl$c$Pop=="B"&NEWDF.ddl$c$session=="2005"]=0


This worked in the case of estimating a fixed apparent survival and recruitment rate over the period of study, which was my goal. I confirmed the validity of this with simulated data. Jeff tells me that the approach would be slightly different if I wanted time-varying estimated of survival, but it is possible. I have not tested it though.

Hope this helps someone in the future, I have received much help from this forum and am very grateful!