Multi-State Convergence Issues

[Tyler Jessen]

Hi all –

I am attempting to use unmarked with a 3 state multi-state model to model habitat selection in relation to a number of covariates. The states correspond to individuals, groups of individuals, and groups of individuals with offspring, with the idea being that these different groups (states) select for different habitats. However, these 3 state models are understandably data-hungry, and I am having model convergence issues, e.g.:

Warning message:

In (new("standardGeneric", .Data = function (object)  :

  Model did not converge. Try providing starting values or increasing maxit control argment.

 

I am wondering if any of you might have suggestions for overcoming these issues, or if I am simply asking too much of the data. For reference, I am working with 1824 sites (from aerial surveys), 3 total surveys, and 1 year (in actuality 2 years using a ‘stacked’ design). Generally, p is quite low (0.05 – 0.15, depending on the model), which I know can affect model convergence. I have had some success with 2-state models and more simple occupancy models, but with the multi-state models, I am only able to fit one or two covariates before the models begin to breakdown. I have tried a number of strategies, such as scaling all covariates, successively increasing the complexity of each model, etc., but typically 3 covariates is where the models will not converge.

 

I have been looking through some of the unmarked chats as well as others, which has informed what I've been attempting above, e.g.:

https://groups.google.com/g/unmarked/c/pyTyiuPgaIY?pli=1

https://groups.google.com/g/unmarked/c/RWkCJsMnfvs

https://groups.google.com/g/unmarked/c/mYMpF7pwxa8

https://stats.stackexchange.com/questions/449132/warning-messages-in-occupancy-modelling-unmarked

 

…but I wanted to reach out to see if any of you might have any additional insight into this type of issue or whether my data are simply too sparse to fit too many covariates. Happy to provide some more information as needed.

On a slightly different note, I noticed that the M-B GoF test is not available for multi-state models. Might anyone have suggestions for how best to assess fit for multi-state models constructed with unmarked?

 

Any insight would be greatly appreciated!

Many thanks,

Tyler

[Marc Kery]

Dear Tyler

yours is a static multistate model, right ? And so have you used initial values in your attempts at get the model going ?

1800 sites would not count for me as a small data set at all, so I don't think it's that. If you have not tried to supply inits, then I'd definitely try that. Where should they come from ? Perhaps you can fit all your covariates first singly or in pairs and then use the resulting estimates as inits in more complex models, which, however, you would still build up progressively until you're at the model you're interested ?

Not sure about the GoF question.

Best regards  --- Marc

---

From: unma...@googlegroups.com <unma...@googlegroups.com> on behalf of Tyler Jessen <tjess...@gmail.com>
Sent: Wednesday, November 22, 2023 21:10
To: unma...@googlegroups.com <unma...@googlegroups.com>
Subject: [unmarked] Multi-State Convergence Issues

 

--
*** Three hierarchical modeling email lists ***
(1) unmarked (this list): for questions specific to the R package unmarked
(2) SCR: for design and Bayesian or non-bayesian analysis of spatial capture-recapture
(3) HMecology: for everything else, especially material covered in the books by Royle & Dorazio (2008), Kéry & Schaub (2012), Kéry & Royle (2016, 2021) and Schaub & Kéry (2022)
---
You received this message because you are subscribed to the Google Groups "unmarked" group.
To unsubscribe from this group and stop receiving emails from it, send an email to unmarked+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/unmarked/CACYCzUCK__aZBB-N-7Nj6Pqv8TRLNm0XK%2B5-qN2rPuCFiG4Mrw%40mail.gmail.com.