restricting model averaging to a subset of models

questions concerning analysis/theory using program MARK

restricting model averaging to a subset of models

Postby THOMO386 » Tue Dec 14, 2021 11:33 pm

Hello,

I'm trying to run a Burnham joint live encounter and dead recovery model estimating survival rates for a population of sea lions. All individuals are tagged as young and my aim is to identify survival rates for different ages/ age classes. Model selection indicates that the top model does not have high support so I am trying to model average survival rates.

My issue is that the candidate models have different structures for the survival parameter (S), therefore when I try to model average survival rates for a certain age (say 3 year olds), that parameter is individually identifiable in some models but not others (e.g. ages of 3 upwards have the same survival rate in certain models but not others).

What is the best approach to model averaging in this situation? Is there a way I can select a subset of models from the model selection results window on which to run model averaging?

Thanks very much,
Moss
THOMO386
 
Posts: 2
Joined: Sun Jun 13, 2021 10:38 pm

Re: restricting model averaging to a subset of models

Postby cooch » Wed Dec 15, 2021 8:25 am

There is no 'one button' approach to doing what you want -- MARK has the ability to average over models of a particular 'data type', but that's not what you're trying to do. There are approaches that involves re-indexing the parameters so that you only use certain parameter index numbers for certain 'subsets of models', but that can be a hassle for a large model set.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: restricting model averaging to a subset of models

Postby jlaake » Sun Dec 19, 2021 11:33 am

I assume by different structure between models you mean, different PIM structure. If that is the case, there is no good answer than to make the PIM structure constant across models and use design matrices to build the models. In RMark, I use the all-different PIMs for all models as the default because it makes it easy to both build models with formulas and to model average.
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: restricting model averaging to a subset of models

Postby cooch » Tue Dec 21, 2021 8:43 am

Jeff posted a better worded version of what I was driving at. In theory, an 'all different' PIM (or, in other words, a full 'age x time x cohort' PIM structure), you could build a set of models where you could achieve what you want. I would agree with Jeff that this is more efficiently implemented using RMark -- provided you have a pretty decent understanding of MARK and PIMs.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: restricting model averaging to a subset of models

Postby cooch » Tue Dec 21, 2021 2:44 pm

Having said that, if picking a subset of models from a larger candidate model set, (i) you'd need to re-normalize AIC values (easy enough), but (ii) you might be accused of violating some of the basic conceptual basis of making inference over an a priori specified set of models. Picking and choosing models to average over is mechanically possible (most straightforward way would be to export the table of model results as a spreadsheet, and then parse the ones you want using some scripting language or other -- I normally use a combination of grep and awk for such things), but perhaps not the most defensible approach.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University


Return to analysis help

Who is online

Users browsing this forum: No registered users and 11 guests