"Jolly-Seber" model

[Eldad Elron]

I'm new to MARK so bear with me.
I'm working with my mark-recapture data on green toads that were captured during three trapping seasons (from November to April; first year – 15 events, second year - 12, third year - 15). I used the "Recaptures Only" model to estimate survival and recapture probabilities and it worked ok - I got the Phi and p results. But when I use the "Jolly-Seber" model with the exact same file (to get also the population size and growth) the process is stuck during the processing of the data in the DOS window and I get a cookie with the message:
"ERROR -- Type of input matrix ם, S, p, r, c, N, F, F', Gamma, Gamma', Gamma", R, R', f, M. d, Lambda, Psi, pent, or pi was not specified. Exit code 30 Exit Window?"

Below is an example of my lines of data :

/*4708*/ 000000010010000000001100000000000000000000 1 0;

I have individual records of 187 animals. The coding '0 1' indicates that the individual is a male, and '0 1' indicates that the individual is a female.

Does anyone have an idea why does it happen? Did I had to do anything differently compare to the procedure in the "Recaptures Only" model?
(As much as I know there isn't a section in the current MARK manual that explains how to do the classical Jolly-Seber model)

Any advise would be greatly appreciated. Thanks.

Eldad

[cschwarz@stat.sfu.ca]

You may wish to consult the new chapter that has been recently (within the last two weeks) posted to the Gentle Introduction to Mark book dealing with Jolly-Seber models and their fitting.

Are you fitting the Jolly-Seber model in MARK or POPAM or one of the variants? The Jolly-Seber option in MARK is known to have convergence problems and you may find that other formulations will work better - check the above mentioned chapter.

You mention though that there are 3 sessions (year) with 15, 12 and 15 events in each session. This sounds more like a robust design rather than a classical capture-recapture model. The GIM book has a nice section on the robust design as does Williams et al.

[cooch]

You may wish to consult the new chapter that has been recently (within the last two weeks) posted to the Gentle Introduction to Mark book dealing with Jolly-Seber models and their fitting.

Are you fitting the Jolly-Seber model in MARK or POPAM or one of the variants? The Jolly-Seber option in MARK is known to have convergence problems and you may find that other formulations will work better - check the above mentioned chapter.

You mention though that there are 3 sessions (year) with 15, 12 and 15 events in each session. This sounds more like a robust design rather than a classical capture-recapture model. The GIM book has a nice section on the robust design as does Williams et al.

Actually, there is no robust design chapter in the 'book' - yet. By June, hopefully...

I agree with Carl - sounds as if your design is a robust design problem, not a J-S problem. I'd consult the Williams et al. book for basic theory of robust design analysis, and then Bill Kendall's paper on RD in MARK, posted on Gary's website:

http://www.warnercnr.colostate.edu/~gwh ... umentation

[Eldad Elron]

Thanks for your replies.

I work with MARK and yes, I select the "Jolly-Seber" model as the data type (in the "Enter Specifications for MARK Analysis" window).

About the robust design:
I was told earlier that my data are not appropriate for use in robust design for an important reason: the 10-15 days of sampling within the larger 6-month period are not organized in short sections of consecutive days. they don't look like samples of succesional days with intervals between them like 3-3-3-3-3. I understand that this must happen in robust design because the shorter sessions form closed models.

Even when I tried to estimate these parameters (population size and growth) within-year through the "Jolly-Seber" model with a more "simple" .inp file, the process is stuck in the same place (during the processing of the data in the DOS window).

Below is an example of a line from the simplest file I tried to run:

/*5281*/ 111001100101 1; (I did not use here sex as a group)


I also tried to run it through "Run"->"Pre-defined models" and selected all possible parameters for each tab, and got the same problem for all models.

At the end I get a message that says:

ERROR

Error Code: 33 (DATA TYPE ERROR)
Subsystem: BASE
Function: -> LOGIC
Argument: {[0000000003]0x01F791DO}
Type: ARRAY
Requested type: LOGIC
CallStack:
MYTOPAPPWIN:RUNMULTMODELSNT (line: 23)
MYTOPAPPWIN:RUNMARKF (line: 833)
MYTOPAPPWIN:RUNMULTMODELSNT (line: 2)
APP:START (line 684)
START (line:

Any help will be appreciated very much.

Thanks,

Eldad

[cschwarz@stat.sfu.ca]

Have you tried the POPAN or Link-Barker models rather than the Jolly-Seber data type. As indicated previous, the Jolly-Seber data type in MARK has convergence problems. The chapter on Jolly-Seber models that is posted in the Gentle Introduction to MARK will walk you through the POPAN or Link-Barker or Pradel models that are EQUIVALENT to the classical jolly-seber models, but only the POPAN model (of the three formulations discussed in GIM) will give estimates of abundance directly.


In terms of the robust design, there is the classical robust design where secondary periods are taken from a closed design, and the newer robust design where even the secondary periods can be from open models. Check papers by Bill Kendal or Schwarz and Stobo in the last few years - do a key word search on robust design open populations temporary migration etc.


Returning to your problem at hand, as a first start, have you tried running the Jolly-Seber models (e.g. via POPAN, Link-Barker, or Pradel) on each year separately to see if there are any problems within years with sparse data etc. If these models don't fit very well for each separate year, you will have difficulty in fitting models to the entire dataset as well.


In general, I find starting from simpler models and working upwards to more complex models a useful exercise to indentify problems with the data etc.

As well, could you please post a little more detail on the sampling protocol. For example, 6 months of trapping in 12-15 intevals sounds like trapping every 2 weeks. Was the study area sampled in its entirely every 2 weeks or were sub-sections sampled every 2 weeks. Have you considered the necessary assumptions for any mark-recapture study such as tag loss, mixing, heterogeneity etc - see the GIM book for details.

[Eldad Elron]

Yes I did run the Burnham Jolly-Seber model on each year separately and it didn't worked as well.

I somewhat solved the problem and understand why the Burnham Jolly-Seber model I choose didn't work. As you suggested I downloaded the "Gentle Introduction to Mark book" and went through the Jolly-Seber models (chapter 14). I also saw that the authors mentioned that the Burnham JS model is problematic – "This model can be difficult to get numerical convergence of the parameter estimates…"

I decided to use the POPAN model that support estimates of abundance and it worked quite well. The N I received for the first site was quite close to the result I obtained in the past using Schnabel multiple-recapture method but for my opinion even more accurate.

Some more details on the sampling protocol: we have three years of mark-recapture data (PIT tags) of green toads (bufo viridis) from five study sites. Each year (trapping session), sampling lasted for about 6 months (until all individuals disappeared to their summer shelters), with unequal number of capture occasions (about 10-15 trapping events each year), and unequal time intervals between capture occasions (ca. 5 to 17 days). The study areas did not changed over time and were sampled in their entirely. From within year observations it seems that almost all tags were not lost (you can easily see that the insertion point was recovered and the tag stayed in the animal body).

Thanks,

Eldad