Groups in CJS with TSM

[CBardier]

Hi, I am fitting CJS models to a data set (of frogs) with 10 encounter occasions (seasonal surveys along 3 years). I have a problem with the definition of the groups, because I am not sure if depending on the definition of the groups, the parameter estimation would be biased.

I have adult males, adult females, juveniles recaptured as adult males, juveniles recaptured as adult females, a juvenile recaptured as juvenile, and juveniles never recaptured. Juveniles recaptured as adult males were 10, juveniles recaptured as females were four, one of which was recaptured as juvenile before being recaptured as adult female.

I made five groups: "adult males", "adult females", "juvenile males", "juvenile females", and "undetermined" (the juvenile recaptured as juvenile plus juveniles never recaptured). The GOF using UCARE supported the use of the CJS for the data set globally and separatelly for the first four groups, but could not compute any test for the "undetermined". So, I decided to exclude the undetermined from the data set (keeping the other four groups), fitting common CJS for the adults, and fitting TSM models for the juveniles recaptured as males or females (to obtain juvenile estimations).

My problem is that all animals within those juvenile groups were recaptured at least once, there are no "unrecaptured" juveniles (they were in the undetermined group which I deleted). When I fit the CJS, the juvenile survival and recapture probability result higher (when averaging) than the adult estimates, which I found suspicious.
Is there a bias acting in my juvenile groups?

If so, how should I group my data in order to avoid that bias without losing information?
I also found sex bias in captures and recaptures, so I would not like to pool males and females in an "adult" group, even the juveniles have that sex bias. And I would like to obtain a juvenile survival apart from the adults... am I pretending too much from my data?

Thanks in advance,

CB

[jlaake]

You should NOT define groups based on recapture outcomes which are data. Groups can be defined on animal attributes but being recaptured is not an attribute. If you have an uncertain group you should be thinking about multistate models with state uncertainty. MS models would also be useful to handle your stages.

[CBardier]

Thanks a lot jlaake!
My initial approach was a MS model, but the test WBWA in U-CARE could not be computed (the others and the global could and were not significant).
So I doubted if the data were suitable for a MS model, as I said, there are too little juvenile recaptures.
But I can retourn to that path.

Thanks again!

C