U-CARE GOF with multiple sites

[moses502]

I will be conducting a survivorship analysis in MARK using the CJS model structure. I have 11 study sites where mark-recap data has been collected over a range of 5-2 yrs. As well, each site has had a varying number of capture occassions. I have 4 groups representing sex (F/M) and age (A/SA) combinations. In MARK I will use the dot notation to allow for the missing data and to permit site comparisons by including all data.

I have run GOF in U-Care and have found only one site to have a significant Global TEST. I have examined that sites component tests to reveal the "trap dependence" test to be non-significant and the 2.Ct to be significant. Moreover, when examined per group, only the Male-Adult group appears to be significant and seems to be causing the lack of fit. Any suggestions on how to use this knowledge to derive the most adequate "umbrella model"?

[CHOQUET]

Hello,

If only Test2.Ct is significant then there is trap-depedence and
you have to modelize it for group Male/Adult.
If the structure of age is not used then the best option is to decompose
the data according to

Pradel, R. 1993. Flexibility in Survival analysis from recapture data: Handling trap-dependence. Pages 29-37 in J.-D. Lebreton and P. M. North, editors. Marked individuals in the study of bird population. Birkhaüser Verlag, Basel, Switzerland.

An option in U-CARE does this job for you. Then you have to apply two classes of age in capture for group Male/Adult.

Otherwise, I don't really understand the notation dot for missing data
used in Mark. Apparently, it is new but I don't see the purpose!!!!! .
By definition the Capture-Recapture models handle missing data.
For notation on models, refer to Lebreton(1992) for singlestate and Choquet(2004) for multistate. Softwares like M-SURGE and E-SURGE do sometimes a better job for multistate models especially with a large number of states.

Sincerely,

Rémi Choquet

[moses502]

Rémi Choquet,

Thanks for the advice. Additional question, should I model trap dependency for all Male/Adult groups across all 11 study sites, or just the site where it has proved significant?

Regarding the "Dot Notation", this is my understanding...MARK requires that number of occassions as well as the interval length be specified at the onset. If, as in my case, a variety of sites were sampled and A) the sites were not sampled for the same number of occassions and B) each site has a unique sampling interval, the user is left with three options: 1) Run each analysis seperately for each site, straight forward but then making across site comparisons are difficult, 2) run them all together and set the probability of capture equal to zero for sites where sampling did not occur, or 3) include a (.) on the input file for occassions when sampling did not occur. The last option does, in effect, set probability of capture equal to zero, its just faster for modelling. Thats my take on the "Dot Notation".

[CHOQUET]

> Thanks for the advice. Additional question, should I model trap
> dependency for all Male/Adult groups across all 11 study sites, or just > the site where it has proved significant?

YES

> Regarding the "Dot Notation", this is my understanding...MARK
> requires that number of occassions as well as the interval length be
> specified at the onset.

This is not related to MARK but to Capture-Recapture data in general

> If, as in my case, a variety of sites were sampled and A) the sites
> were not sampled for the same number of occassions and B) each site has a unique sampling interval, the user is left with three options:

Run each analysis seperately for each site if there is really a lot of difference in the sampling.

> 2) run them all together and set the probability of capture equal to
> zero for sites where sampling did not occur, or 3) include a (.) on the
> input file for occassions when sampling did not occur. The last option
> does, in effect, set probability of capture equal to zero, its just faster
> for modelling. Thats my take on the "Dot Notation".

If the only difference is that at some site there is sampling or not at some occasion then 2) because I don't understand the meaning of 3)

However it is a difficult subject and it is very difficult to talk
of this kind of subject by the forum ore e-mail. I only have a partial view of the way indivuals are captured in each site.