Page 1 of 1

Mark-resight model with varying effort?

PostPosted: Wed May 23, 2018 9:45 pm
by darryl
Hi All,
I'm wanting a fit a mark-resight model where lamda_j = E_j*lamda, where E_j is the effort expended for either group j or primary period j (if using a robust design). For example, E_j might be the number of days of searching in different areas, or days searching each primary period. In a GLM framework, E_j would often be referred to as an 'offset'.

I've had a poke around various websites and documentation and can't find a way to do this easily. Which now that I've said that, someone is going to post a link and tell me to RTFM ;-)

I could use E_j as a covariate, which kind of does what I want but not exactly. Only other solution I can think of is breaking data down into periods of consistently-sized effort, and use robust design with appropriate constraints and missing data.

Ideally looking for a solution in RMark/Mark, but open to suggestions of other R-based solutions.

Cheers
Darryl

Re: Mark-resight model with varying effort?

PostPosted: Thu May 24, 2018 11:16 am
by jlaake
I don't know mark-resight models well and Brett is probably the best person to answer this but I don't follow why using effort as a covariate doesn't do what you want. Also, this isn't really an RMark question.
--jeff

Re: Mark-resight model with varying effort?

PostPosted: Thu May 24, 2018 4:33 pm
by darryl
Thanks Jeff. Posted this in RMark forum as using RMark for the analysis currently and wanted to check with the experts that there wasn't an RMark/MARK option/trick I was missing for fitting such a model, before exploring other options or settling for using effort as a covariate.

Using effort as a covariate doesn't do it, exactly, because that gives the model (using log-link):
lambda_j = exp(beta_E*E_j)*lambda

and I want to use effort as a fixed offset rather than as a covariate with an estimated effect size. I could fit the model I'm after by using log(E_j) as a covariate if I could fix the associated beta parameter to 1, but I don't think you can fix beta parameters, only real parameters.

Cheers
Darryl

Re: Mark-resight model with varying effort?

PostPosted: Thu May 24, 2018 4:49 pm
by jlaake
I think I'm confused because none of the mark-resight models that are in RMark have a lambda parameter so it appeared to me that lambda was the parameter to be estimated. But there is no way to include an offset in MARK that I'm aware of because you are correct that you cannot fix a beta parameter.

--jeff

Re: Mark-resight model with varying effort?

PostPosted: Thu May 24, 2018 4:58 pm
by darryl
Sorry Jeff, my bad. lambda is the derived parameter. Thanks for the confirmation.

Re: Mark-resight model with varying effort?

PostPosted: Fri May 25, 2018 9:00 am
by cooch
jlaake wrote:But there is no way to include an offset in MARK that I'm aware of because you are correct that you cannot fix a beta parameter.

--jeff


There *is* code in MARK to fix betas, but it hasn't been 'turned on', so to speak.

Re: Mark-resight model with varying effort?

PostPosted: Fri May 25, 2018 9:04 am
by cooch
darryl wrote:Sorry Jeff, my bad. lambda is the derived parameter. Thanks for the confirmation.


You can fit models to derived parameters, in a fashion, using the moment-based random effects models. For example, you can build a DM, apply it to the derived parms, and generate shrinkage estimates. But, you can't actually fit the model to the data.