Trap dependence

questions concerning analysis/theory using program MARK

Trap dependence

Postby Mariana Mira » Sun Apr 27, 2008 10:09 pm

Using U-CARE to perform a Goodness of Fit test in my data, I found a trap dependence with trap-happyness in my data. I would like to know how do I perform a model accounting for this trap dependence. I did not find this in the Mark manual. Please, anybody can help me aswering the following questions? How would be the PIM for p? And the desing matrix?
thanks in advance
Mariana
Mariana Mira
 
Posts: 11
Joined: Thu Mar 01, 2007 10:25 am
Location: Brasília, Brasil

Re: Trap dependence

Postby Mariana Mira » Sun May 04, 2008 8:41 am

Please, can anybody help with the my question above? I just need to solve this problem to proceed with my data analyses.
Thank you!
Mariana
Mariana Mira
 
Posts: 11
Joined: Thu Mar 01, 2007 10:25 am
Location: Brasília, Brasil

trap dependence

Postby jlaake » Sun May 04, 2008 9:54 am

You didn't say anything about what model you are using or what your goals are. Some closed capture models can handle trap dependence. One way to handle trap dependence in some open models is to create a time dependent covaraiate. which is acturally a set of covariates. You use tdi to model capture probabiity p(i+1) and tdi is 0/1 depending on whether they were caught the previous occasion. Then you need to create a design matrix that includes a column with the values td1,td2,...tdk for the rows of p for p(2), p(3),....p(k+1)

Hope that helps.
jlaake
 
Posts: 1480
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: trap dependence

Postby Mariana Mira » Sun May 04, 2008 12:05 pm

Thanks for your help! But, I don't understand how exactly this set of covariate will be created because in the second capture event, for example, the data contains individuals that were caught on the previous event and that were not. How can I determine if the td2 will be 0 or 1? I am using the CJS model Phi(s*t) p(s*t) as a global model, where s is sex and t is time. And I found a trap dependence specially for males.
Mariana Mira
 
Posts: 11
Joined: Thu Mar 01, 2007 10:25 am
Location: Brasília, Brasil

Postby jlaake » Mon May 05, 2008 10:46 am

The td2 covariate can take on both values. Here is an example of what your data might look like if there were 3 occasions.
ch td2 td3
100 1 0
011 0 1
111 1 1

Simply use the first k(2 in this case) entries of the ch for td2,..tdk+1 values. A simple
design matrix might look like:

p2 1 td2
p3 1 td3

Sounds like you might need an interaction so it might be

Male p2 1 td2 0 0
Male p3 1 td3 0 0
Fem p2 0 0 1 td2
Fem p3 0 0 1 td3

--jeff
jlaake
 
Posts: 1480
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Postby CHOQUET » Mon May 26, 2008 9:23 am

The co-variates approach is not the classical and easiest way to deal with Trap-Dependance.
Two main approach exist:
1) The first one is described in
Pradel, 1993, Flexibility in survival analysis from recapture data: handling trap-dependence. Marked individuals in the study of bird population. P29-37

You have to transform your data set according to Pradel (93)
before any analysis and use two age classes to model capture.
The transformation of the data set can be done automatically
by U-CARE -> TRANSFORM DATA -> Split for trap-dep analysis

2) The second one is described in
Gimenez O., R. Choquet, and J.-D. Lebreton (2003). "Parameter redundancy in multistate capture-recapture models". Biometrical Journal 45:704–722

You have to consider two states, previously captured vs not previously
captured. Additional comments how how to implement this approach are given here

http://www.cefe.cnrs.fr/BIOM/pdf/Choque ... models.pdf

Sincerely,

Rémi
CHOQUET
 
Posts: 212
Joined: Thu Nov 24, 2005 4:58 am
Location: CEFE, Montpellier, FRANCE.

Re: Trap dependence

Postby mekeys » Sat Feb 27, 2010 10:14 am

I am working on a single group study to ascertain survival estimates of a seabird population - with 32 phi's and 32p's - and also running various environmental/anthropogenic constraints. I have run a Ucare GOF and the data was found to be trap shy. I split for trap dependence. Now... I know I need to add an additional age class in my recapture PIM in order to account for this, using the transformed data obtained through the Ucare GOF test - split for trap dependence. How to I actually apply this to the matrix? Should I follow the mannual - chap. 7, page 6 onwards, ie: phi(t) p(a2-./.) for 2 age classes, constant through time and phi(t) p(a2-t/t) for 2 age classes time dependent through time? CAN ANYONE LET ME KNOW BECAUSE I'M A LITTLE CONFUSED BY THE MANNUAL OVER THE CORRECT PROCEDURE. Thanks
mekeys
 
Posts: 5
Joined: Sat Feb 27, 2010 8:43 am

Re: Trap dependence

Postby cooch » Sat Feb 27, 2010 12:41 pm

mekeys wrote:I am working on a single group study to ascertain survival estimates of a seabird population - with 32 phi's and 32p's - and also running various environmental/anthropogenic constraints. I have run a Ucare GOF and the data was found to be trap shy. I split for trap dependence. Now... I know I need to add an additional age class in my recapture PIM in order to account for this, using the transformed data obtained through the Ucare GOF test - split for trap dependence. How to I actually apply this to the matrix? Should I follow the manual - chap. 7, page 6 onwards, ie: phi(t) p(a2-./.) for 2 age classes, constant through time and phi(t) p(a2-t/t) for 2 age classes time dependent through time?


There are any number of ways to handle trap-dependence, that vary mostly in whether or not the dependence is strictly transient (i.e., lasts for some number of intervals after marking, or decays with time at some rate), or permanent, or some combination of the two . If it is transient (typically), the easiest way to handle it is to consider it as a form of capture heterogeneity. In that case, the most straightforward approach - which is the one you're hinting at - is to simply add 'age' structure (TSM - time since marking - is the preferred vernacular) to the encounter parameter. If *all* of your animals show the same degree of trap-dependence, then a simple two-class model with time-dependence for p is a good general starting model. You don't need to split the data for this (which is I suspect where the difficulty arises). So, just take your data, and fit a model with 2 TSM classes for p to it. If the duration of the trap aversive 'stage' is >1 year, then simply make the first TSM class 2 years in duration.

CAN ANYONE LET ME KNOW BECAUSE I'M A LITTLE CONFUSED BY THE MANUAL OVER THE CORRECT PROCEDURE. Thanks


Avoid using all caps - its the text equivalent of 'shouting', and is unlikely to have the desired effect of getting attention (in fact, it will often generate the opposite reaction).
cooch
 
Posts: 1654
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Trap dependence

Postby mekeys » Sat Feb 27, 2010 2:14 pm

Thanks very much for your help - I had previously been playinig with modifying the p PIMS as age/cohorts to build 2 age classes... I'll try the TSM approach. Apologies for the capitals... I'm a lowley undergraduate trying to get my dissertation finished under the duress of time pressure. Its the first time I've used this forum and wasn't familiar with convention - thanks for the tip, and the help.
mekeys
 
Posts: 5
Joined: Sat Feb 27, 2010 8:43 am

Re: Trap dependence

Postby cooch » Sat Feb 27, 2010 2:43 pm

mekeys wrote:Thanks very much for your help - I had previously been playing with modifying the p PIMS as age/cohorts to build 2 age classes... I'll try the TSM approach.


1. build the models you want with full time-dependence, and enough TSM classes as needed

2. build the corresponding design matrix

3. close the PIMs, and build all reduced parameter models by modifying the DM.

This is discussed at length in chapter 6 and chpater 7.

Apologies for the capitals... I'm a lowley undergraduate trying to get my dissertation finished under the duress of time pressure. Its the first time I've used this forum and wasn't familiar with convention - thanks for the tip, and the help.


No worries - although I might add that Fisher was 'a lowely' undergraduate when he came up with likelihood theory and the basis for much of classical population genetics. :wink: Never fear, we were all undergraduates at one time or another.

p.s. the admonition not to use caps is generic - whether its this or any other forum, a classical listserve or usenet group.
cooch
 
Posts: 1654
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Next

Return to analysis help

Who is online

Users browsing this forum: No registered users and 1 guest