Melanargia and multistates model

[Isaël]

Hello,
I post this message here because it is urgent to me to realise analisis with my data. My training period take end the 31/08.
I'm begginer with mark and I understand gradually a lot of things, but too slowly for the end of my study. And here, in this natural reserve, the are no persons abble to help me.
I'm using mark in this time with my data. I try to use the Multistates model as I was advised by sbonner.
In my inp file, I've 10 states (10 Capture-release sites) and 2 groups (males and females). Here is a part of my .txt (the total of line is 1138).

/*ID historique M F */
/*001*/ AB0000000000 0 1 ;
/*002*/ A00000000000 1 0 ;
/*003*/ AJ0JA0000000 1 0 ;
/*004*/ A00000000000 1 0 ;
/*005*/ A00000A00000 1 0 ;
/*006*/ A0A0A0A0A000 1 0 ;
/*007*/ AA0A0AA0B000 1 0 ;

I have read a big part of the markbook but I've just 2 questions that I am not sure:

-In the matrice index chart, the number of parameters is very big (220), it is logic (Phi site 1 to site 2, 1 to 3...10 to 1...) but do you think if it is pertinent to use my data like this ?
-Is it possible with this sort of data and Multistates model to obtain another estimations that recapture probability and apparent survival? (for example size of population, movements...).
Is Popan ok for this type of data?

Another thing, when I start the analisis, a error message appear after a few time : I have read a post in this forum and the solution was to use the alternative optional method.
But mark is running since 3 days and I don't know when the analisis will finish (I can't see the % already realised). I have reverified several time my data in the .txt file.
For intance I I'm working about phytosociologic tab so I don't have stop mark but today or wednesday I re-work with mark. Do you have an idea for resolve this problem?


I thank you in advance,

Isaël

[sbonner]

Hi Isael,

If you have 10 different capture-release sites and are interested in studying the movements between the sites then the multi-state model is the way to go.

Regarding the size of the model, it is easy to generate multi-state models with many parameters -- especially with multiple groups. My suggestion (at the risk of being accused of data snooping) would to fit some restricted models and then see where the models fail, but it seems like you have already done this. I'm guessing from the number of parameters that you have restricted the movement probabilities so that they are the same on all occasions. Unfortunately, even this might be too much to get precise estimates for your data set, but the only way to find this out is to run the model. You might consider some models with fewer parameters -- e.g., a random movement type model assuming that the probabilities of moving from any one state to any of the other 9 states are all equal -- but only if such a model is biologically plausible.

Movement probabilities can be estimated if you are willing to make some further assumptions about the dependence of survival on the different states. This is covered in section 8.1 of the GIM "Separating survival and movement" (Hmmm.... have you read the entire chapter?).

To my knowledge, population size estimation for multi-state models is not implemented in MARK. Schwarz and Dupuis (Biometrics, 2007) did propose a Bayesian approach, but this would be more complicated.

Finally, do you know what the error message actually said? If you think that the alternative optimization routine is the solution, then I'm guessing that it's a convergence error. To be honest, it's not too surprising with a large number of parameters. The likelihood surface is likely to be very complicated with ridges and local maxima. Simulated annealing is less prone to these errors, but not guaranteed to find the global maxima, and it also takes longer to run. This is simply one of the pitfalls of these complicated models -- it takes a lot of data to estimate the parameters.

Hope this helps.

Simon