“Error 0 from VA09AD” not solved with simulated anneling

[anagalantinho]

Hello,

We are analyzing capture-recapture data of the wood mouse using Robust design closed captures parameterization on MARK. We have 104 encounter occasions (26 primary x 4 secondary occasions), 353 animals and we wish to include non- individual covariates (e.g precipitation, bush cover) in the models.

I started by one model per covariate (logit link) and I noticed that AICc and estimates were the same despite using different covariates. Plus a warning “Error number 0 from VA09AD optimization routine” appeared in the output.

When I ran the same model with simulated anneling there was still a warning” Numerical convergence suspect” and the AICc and estimates where identical for different covariate models.

Without continuos covariates using PIM (sin link) I do not get this warnings when I ran models with constant phi and gama’’ and gama’’ nor when I run models with time varying phi and gama’’ and gama’ (setting the last 2 gamas equal). But I do get warnings again when grouping these parameters by precipitation levels (dry seson, rainy season) even with logit link.

How can I overcome this optimization problem?

Thanks,
Ana

[anagalantinho]

Hello again!

Today I have checked again the outputs and noticed that this problem only occurred after installing the newest version available at phidot on 18 july 2012 (MARK setup file was last modified on: Wed, 18 Jul 2012 12:18:44 GMT).

There is no single smily for what I am feeling right now.

Best wishes!

Ana

Hello,

We are analyzing capture-recapture data of the wood mouse using Robust design closed captures parameterization on MARK. We have 104 encounter occasions (26 primary x 4 secondary occasions), 353 animals and we wish to include non- individual covariates (e.g precipitation, bush cover) in the models.

I started by one model per covariate (logit link) and I noticed that AICc and estimates were the same despite using different covariates. Plus a warning “Error number 0 from VA09AD optimization routine” appeared in the output.

When I ran the same model with simulated anneling there was still a warning” Numerical convergence suspect” and the AICc and estimates where identical for different covariate models.

Without continuos covariates using PIM (sin link) I do not get this warnings when I ran models with constant phi and gama’’ and gama’’ nor when I run models with time varying phi and gama’’ and gama’ (setting the last 2 gamas equal). But I do get warnings again when grouping these parameters by precipitation levels (dry seson, rainy season) even with logit link.

How can I overcome this optimization problem?

Thanks,
Ana

[bacollier]

Hello again!

Today I have checked again the outputs and noticed that this problem only occurred after installing the newest version available at phidot on 18 july 2012 (MARK setup file was last modified on: Wed, 18 Jul 2012 12:18:44 GMT).

There is no single smily for what I am feeling right now.

Best wishes!

Ana

Ana,
The fact that the AIC/parameter estimates are all the same is the typical flashing warning sign that something is not correct in the model structure, so check that. Working under the assumption that you have your DM correctly structured for the individual covariate (which I assume is constant over time, is your individual covariate for bush cover time-specific? ) you have done what I would have done as a first step.

For convergence, perhaps you can try using the parameter estimates from your constant model that does run and use those as initial estimates for the parameters in your more complex model. If that does not work and if you are sure your model's are structured correctly, then we will try something else.

Bret

[anagalantinho]

Dear Bret,

I had checked before, but after your help I checked again! Before I constrained g’’k-1=g’’k and g’k-1=g’k’ but after your suggestion I remembered reading somewhere on the mark book that if we introduce covariates we don’t need to force the final gamas to be the same Thank you!

So now I have different AICc. But they are only slightly different such as the estimates, which also have large confidence intervals … The problem might be that we have too many parameters for our data. I am going to introduced covariates on the models which group parameters and see if I can get better estimates.

These covariates are time-specific but I assumed that the relationship with the parameters does not change over time. So I introduced the covariate values in just one column and not right next to each of the time-varying survivals coding as is suggested on MARK help:

1 1 0 0 0 var1
1 0 1 0 0 var2
1 0 0 1 0 var3
1 0 0 0 1 var4
1 0 0 0 0 var5

Thank you very much for your answer!
Best wishes!
Ana