Interpreting GOF results for MSCR model with repro states

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

Interpreting GOF results for MSCR model with repro states

Postby Charlotte » Fri Mar 24, 2017 7:43 am

Hello all,

I’d be super appreciative of any advice regarding how to proceed with my analysis given the results I’m getting from U-CARE.

I want to use multi-state CR models in E-SURGE to investigate the effect of environmental conditions on female costs of reproduction. To do this I have constructed recapture histories for 1156 individuals across 32 years (so individuals enter and exit the study at different points), with each occasion corresponding to a Spring-Summer period when breeding occurs. I’ve used 7 states, with 6 corresponding to differing degrees of reproductive investment and the 7th corresponding to an individual being found dead. Here are a couple of examples of recapture histories:

06167000000000000000000000000000
00000000000000006131313112170000

As recommended, I wanted to conduct goodness of fit tests in U-CARE before proceeding with the E-SURGE analyses, but I’m finding that I’m a bit stuck with their interpretation. The results are:
WBWA (X2 = 447.810, df = 191, p = 0.000)
3G.SR (X2 = 28.719, df = 46, p = 0.98)
3G. Sm (X2 = 290.593, df =190, p = 0.000)
M.ITEC (X2 = 23.7, df =13, p = 0.03)
M.LTEC (X2 = 0.000, df = 0.000, p = NaN)

So, this seems to suggest that transience isn’t a problem, but that there is evidence for memory and trap dependence (however, there isn’t enough data to assess long-term trap dependence). Is my interpretation reasonable?

I’m struggling most to understand what I should do given the apparent memory. This seems important to the question I’m wanting to answer given that I might expect females experiencing certain environmental conditions to be more likely to breed successfully in consecutive years etc. I’m also not sure if the trap dependence results are somewhat driven by the nature of the data, particularly because each occasion corresponds to a single year? I’d be hugely grateful for any recommendations for how to proceed!

Thanks in advance,

Charlotte
Charlotte
 
Posts: 2
Joined: Fri Mar 24, 2017 5:31 am

Re: Interpreting GOF results for MSCR model with repro state

Postby Guillaume Souchay » Thu Mar 30, 2017 3:21 am

Dear Charlotte,

can I guess you're working with Flamingo's dataset? ^^

Well, you have memory in your dataset (not so surprising for this question on your sp), one possibility is to constrain your transition with both former and current state. Another solution (and maybe more robust) is to build your state with information and both previous and current state. As an example, an individual breeding at time t can have code B and was breeding the year before with code A, then its code for time t would be "AB".

How to model memory in CR model is given by Rouan et al 2009 :
Rouan, L., Choquet, R. & Pradel, R. 2009. A general framework for modeling memory in capture—Recapture data. Journal of Agricultural, Biological, and Environmental Statistics, Springer-Verlag, 14, 338-355 http://dx.doi.org/10.1198/jabes.2009.06108. Such solutions may need fewer classification of breeding investments.
You can also found an example in Lagrange et al 2014
Lagrange, P., Pradel, R., Bélisle, M. & Gimenez, O. 2014. Estimating dispersal among numerous sites using capture–recapture data. Ecology, Wiley-Blackwell, 95, 2316-2323 http://dx.doi.org/10.1890/13-1564.1. A similar approach for your case would be to build new state based on whether the investment in time t is higher/same/lower than in time t-1 to avoid too many states ...

Do not hesitate to contact me if you have further questions (ask Roger).

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

Re: Interpreting GOF results for MSCR model with repro state

Postby Charlotte » Tue Apr 04, 2017 7:05 am

Dear Guillame,

Thank you very much for your reply and your helpful suggestions! Alas, I'm not working on Flamingos, I actually work on something far less exotic - sheep. I'm hopeful that your suggestions will help me to incorporate memory into my analysis, so thank you for that.

I'm still slightly struggling to establish whether I should be worried about the trap dependence results as I'm wondering whether it's somewhat driven by the data structure. Given the results of the analysis and the data set-up do you think it is sensible to try to model trap dependence as in the literature (e.g. http://dx.doi.org/10.1371/journal.pone.0032666)?

Thanks again,

Charlotte
Charlotte
 
Posts: 2
Joined: Fri Mar 24, 2017 5:31 am

Re: Interpreting GOF results for MSCR model with repro state

Postby Guillaume Souchay » Tue Apr 04, 2017 7:27 am

Dear Charlotte,

Yes, it could be good to model both trap-awareness and memory, however, it can become quite difficult, with lots of states.
You could perhaps start by computing chi-squarred without the memory effect and see if modelling trap-awareness is really needed, or if you can just account for it using a coefficient of overdispersion (c-hat).

Also, it could help to understand why some trap-awareness is present in your dataset: is it related to the method of capture, age of individuals, ...
If you are able to catch every breeding individuals in your study area and that there is no skipping once an individual starts to breed, then you will have continuous captures of breeding females. This could thus explain why you can find a trap-dependence effect (is it a positive or negative trap-dependence?). So, before incorporating everything (memory + trap-dep) in your model, it could be good to think about it.

If you can give more details, it could be helpful to understand the reason why you find trap-dep in your dataset.

Cheers,

Guillaume

ps: feel free to contact me in private if you don't want to give too much information here.
Guillaume Souchay
 
Posts: 64
Joined: Wed Feb 27, 2008 5:20 am
Location: Nantes, FRANCE

Re: Interpreting GOF results for MSCR model with repro state

Postby Laura Touzot » Fri Jun 08, 2018 5:46 am

Dear Mr. Souchay,

I allow myself to directly ask a new question in this existing post.
Once you have estimated the coefficient of overdispersion with U-CARE, how do you specify this coefficient in E-SURGE?

Thanks a lot,
Laura
Laura Touzot
 
Posts: 2
Joined: Wed Mar 14, 2018 10:35 am

Re: Interpreting GOF results for MSCR model with repro state

Postby simone77 » Sun Jun 10, 2018 12:19 pm

You can change the c-hat going to:
Run & See / Change overdispersion coefficient.
simone77
 
Posts: 197
Joined: Mon Aug 10, 2009 2:52 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 14 guests