Error runing multi-state models

questions concerning analysis/theory using programs M-SURGE, E-SURGE and U-CARE

Error runing multi-state models

Postby anacjesus » Fri Apr 03, 2015 5:38 am

Hi,

I am analyzing a 28 years of data for grey seal females (297 individuals) sighted in 5 locations of an island (regions A, B, C, D and E). I want to see if regions where animals are sighted have an influence on their site-fidelity.

I am using a multi-state model for recaptures only. After inserting data into MARK a thousand of parameters is generated. Because it is impossible to run all selected models (too many parameters), I decided to run a model for each region's comparation fidelity, where I would have, for example, psi (A-B) time dependent, while making constant all the other variables (survival, recapture probability and all the other psi's). Then I would do this to psi (B-A), psi (C-A), psi (B-C), etc.

It is the only way I am thinking for testing this hypothesis.
Anyways, when I try to run this an error pops up: "Numerical convergence never reached, with maximum G=..."

Does any one have other ideas? Or know how to solve this error? I looked into MARK book but their advice didn't help much.

Thank you!!

Cheers
anacjesus
 
Posts: 8
Joined: Wed Apr 01, 2015 9:18 am

Re: Error runing multi-state models

Postby cooch » Fri Apr 03, 2015 7:32 am

anacjesus wrote:Hi,

I am analyzing a 28 years of data for grey seal females (297 individuals) sighted in 5 locations of an island (regions A, B, C, D and E). I want to see if regions where animals are sighted have an influence on their site-fidelity.

I am using a multi-state model for recaptures only. After inserting data into MARK a thousand of parameters is generated. Because it is impossible to run all selected models (too many parameters), I decided to run a model for each region's comparation fidelity, where I would have, for example, psi (A-B) time dependent, while making constant all the other variables (survival, recapture probability and all the other psi's). Then I would do this to psi (B-A), psi (C-A), psi (B-C), etc.


Thats because the default time-dependent model is over-parameterized for your data. If you have 5 states, and (say) 20 occasions, then you have *lots* of parameters to estimate, and unless you have relatively decent encounter probabilities (at the least), you're going to have trouble fitting a fully time-dependent model. Rather than assume that MARK is the problem, try running a starting model where all parameters are fixed over time. If you don't know how to do this, or why you might try this, then...(see last comment, below).

It is the only way I am thinking for testing this hypothesis.
Anyways, when I try to run this an error pops up: "Numerical convergence never reached, with maximum G=..."

Does any one have other ideas? Or know how to solve this error? I looked into MARK book but their advice didn't help much.


[So, why would a book about MARK tell you something about problems you have running M-SURGE?]

It seems that you have a number of basic misconceptions about fitting these sorts of models to your data. Whether you use MARK or M-SURGE to fit multi-state models is not the issue (for many of the models you're likely to be interested in, they'll do the same thing for you). I would suggest you step away from your data, and work through the first 7-8 chapters of the MARK book. Then, once you have a better understanding about models, model specification, estimation, and model selection, you can turn to your own data. Even if you ultimately use M-SURGE, the basic concepts introduced in the MARK book still apply.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Error runing multi-state models

Postby anacjesus » Mon Apr 06, 2015 4:37 am

Hi,

Thank you very much!

I am following you advice and have been reading some more chapters on the Mark book. I actually realized later that I shoudl have posted that question in the MARK section, as I am working with MARK and not M-SURGE. I thought I had posted there, so sorry for the confusion!

I decided to work with RMark because I believe is easier to change the design matrix, so I am now playing around with that.

However, I have a parallel question. I used MARK for testing barker's model with the same dataset, as I wanted to look at site fidelity among animals that migrate from the study year within their encounter history. I managed to do that in MARK and the results were what I was expecting. Although, I wish to fit the model to the data by changing the c-hat. I was going to this by runing GOF test in U-CARE and calculate c-hat, but whenever I try to run it U-CARE says I cannot run multistate models in a normal GOF test.

I don't understand why does it consider this dataset multistate, as it is actually a Barker model, with LDLD data format (10; 01 and 02 only -> no states considered!)

Anyways I try to run the GOF for multistates and U-CARE crashes all the time. Any ideias?

Thanks again!
anacjesus
 
Posts: 8
Joined: Wed Apr 01, 2015 9:18 am

Re: Error runing multi-state models

Postby Guillaume Souchay » Tue Apr 07, 2015 3:37 am

Although, I wish to fit the model to the data by changing the c-hat. I was going to this by runing GOF test in U-CARE and calculate c-hat, but whenever I try to run it U-CARE says I cannot run multistate models in a normal GOF test.

I don't understand why does it consider this dataset multistate, as it is actually a Barker model, with LDLD data format (10; 01 and 02 only -> no states considered!)

Anyways I try to run the GOF for multistates and U-CARE crashes all the time. Any ideias?


This is basically because the data format is not the one required by U-CARE ...
Have a look at U-CARE manual.
The format in U-CARE is SSSSSSS where S is the state of the individual at the sampling occasion. Thus, it can be either L or D.
Reading manuals can help to understand how software works and what they require.

Cheers.
Guillaume Souchay
 
Posts: 64
Joined: Wed Feb 27, 2008 5:20 am
Location: Nantes, FRANCE

Re: Error runing multi-state models

Postby anacjesus » Tue Apr 07, 2015 8:54 am

Thank you for that!

I actually didn't have the update version of UCARE manual and the one I had didn't account for multistate models.

However, I am still trying with my data in LLLL format and 2 states (1 and 2) and U-CARE crashes in the middle of GOF analysis.

Any ideas on that?

Cheers
anacjesus
 
Posts: 8
Joined: Wed Apr 01, 2015 9:18 am

Re: Error runing multi-state models

Postby Guillaume Souchay » Tue Apr 07, 2015 9:12 am

Hi,

The last version is old, from 2005 and can be found here: http://www.cefe.cnrs.fr/images/stories/DPTEEvolution/biostatistiques/Chercheurs/remi_choquet/Choquet-USER%20MANUAL%20U-CARE%202.2.pdf

How did you code your data? Only with 1 and 2?
Including a dead state, you need to have 0 (not encounter), 1 (detected alive) and 2 (detected dead). Then, U-CARE should be able to automaticaly detect that the state 2 is an absorbing state.

Also: http://www.cefe.cnrs.fr/images/stories/DPTEEvolution/biostatistiques/Chercheurs/remi_choquet/Choquet-U-CARE-Ecography-Dec-2009.PDF
more recently, McCrea et al 2014 on GoF Test with lvie and dead data: http://link.springer.com/article/10.1007/s13253-014-0174-1
Guillaume Souchay
 
Posts: 64
Joined: Wed Feb 27, 2008 5:20 am
Location: Nantes, FRANCE


Return to analysis help

Who is online

Users browsing this forum: No registered users and 17 guests

cron