Trap happiness, TSM, MARK methods

[clare_wormald]

Question regarding trap happiness methods in MARK

I am using MARK to generate survival estimates of live recaptures of a coral reef fish.

Component tests using UCARE indicate trap happiness (it appears that an individual's probability of capture at time t+1 is influence by whether or not the individual was captured at time t).

I understand that a suitable model for trap happiness incorporates the TSM model structure in p (recapture).

I wish to verify that the method I will use to deal with trap happiness is correct, specifically these two steps:


1. Transform the data by splitting the encounter histories for trap dependent analysis

2. Use a 2 age class structure in p (recapture)

The encounter history splitting technique is not mentioned in the MARK manual so I am concerned that I may be confusing two methods of dealing with trap dependence. If these techniques are not correct, I would appreciate any guidance on dealing with trap happiness you can give me.


Supplementary information:

Fish are recaptured, not resighted -
Handling methods of inital capture and tagging and subsequent capture occasions are almost identical, therefore trap response is unlikely to be limited only to the interval after initial capture. Fish likely use the traps as shelter which may account for the "happiness" in entering an unbaited trap.

[Christian Ramp]

Transients:

If you have transients (animals seen only once) in your data set you can model two age classes for phi. Transients are modeled in the first age class. The second age class is then not biased by this ‘age’ group. Since the actual age is not known, these models are referred as to TSM -time since marking - models.

Trap response:

You can use the option ‘splitting for trap dependence’ in U-CARE.

You can use the multi state approach and define one state as the departure state (animal captured/not captured at t) and the second state as the arrival state (animal captured/not captured at t+1).

Using the single state approach you can implement trap response as individual covariates in just adding the encounter history for each individual as covariate, without the last capture occasion. That means for each p you take into account if the animal was seen in the occasion before. There was a posting by Gary describing that in the old forum, I do not know any other source for the way.

I think both approaches are called immediate trap response models (Pradel, 1993) and are marked phi (t) p (t*m). I do not know how you handle ‘permanent’ trap response, as it seems that your fishes once trapped come back to the traps over and over again.

Further explanations you find in the MARK (GOF and age chapters) and U-CARE handbooks.


Hope that helps.
Chris

You might want to read:

Pradel, R., J.E. Hines, JD. Lebereton and J.D. Nichols 1997
Capture-Recapture Survival Models. Taking Account of Transients. Biometrics 53: 60-72

Pradel, R. 1993 Flexibility in Survival Analysis from recapture data : Handling Trap-Dependence. In : Marked Individuals in the Study of Bird Population. J.-D. Lebreton and P.M. North (eds.) Birkhaeuser Verlag Basel Switzerland pp 29-37

[clare_wormald]

Chris, thanks for the response. I very much appreciate the dialogue.

Fortunately, there is no evidence of transience in this population.


My information for the two step method I mentioned in my inital post - splitting, followed by treating trap dependence as age dependence in the split data set was derived from the Pradel 1993 paper.

The method outlined in Pradel 1993 makes logical sense to me in the treatment of the trap happiness in my fish as it deals with the short term trap response seen in my data.

I think that the data DO demonstrate "immediate trap response" (the positive influence of capture at one occasion on future capture does not persist beyond the next occasion). The approach Pradel advocates then in data splitting and the use of the phi (t) p (t*m) I think is appropriate.


My uncertainty is the compatibility of this method with the MARK software and protocols. I am looking for validation!


The single state approach you mentioned also looks promising, I will explore that option more if I cannot verify the Pradel approach is appropriate in MARK. I am less familiar with multistate models so that would be my third choice at this point.

Many thanks.