random effects in E-SURGE

[roos]

Hi,

I have a question regarding the use of random effects in E-surge. I want to analyse dispersal in a very monogamous bird species, and thus pairs are not-independent of each other. However, not for every pair we have marked both sexes, only of 35% of the birds we know its partner. That means that if I use pair as a random effect, some have two individuals but most have one.

As a test, I tried to run the simplest CJS model with this random effect of pair, but I got an error even before the iteration process. If, as a test, I remove the ‘single’ individuals, then it seems to work though. Is it correct that a random effect can only be considered if the sample size per pair is more than one?

Another test I did is to run a multievent model with random effect for the dataset in which both sexes per pair are known, and of which the CJS worked. Of some birds we didn't knew their territory in every year, but often they were seen in the study area and thus alive, hence the multievent structure (4 events: not seen, seen territorial in A, seen territorial in B, seen alive, and 3 states: territorial in A, territorial in B, dead). The model works without the random effect. The iteration process did start, but writing the model results resulted in an error. What it says is this:

log10 of relative estimate error in integration: order16-15=-11.7216
Non available biological parameter
Index exceeds matrix dimensions.

It estimated 0 parameters.

Is this due to my particular dataset?

Hope you can help!

Roos Kentie.

[CHOQUET]

Hello Rosemary

In reply to:

>As a test, I tried to run the simplest CJS model with this random effect of pair, but I got an error even >before the iteration process. If, as a test, I remove the ‘single’ individuals, then it seems to work >though. Is it correct that a random effect can only be considered if the sample size per pair is more >than one?

Next version of E-SURGE will be able to manage this case

>Another test I did is to run a multievent model with random effect for the dataset in which both sexes >per pair are known, and of which the CJS worked. Of some birds we didn't knew their territory in >every year, but often they were seen in the study area and thus alive, hence the multievent structure >(4 events: not seen, seen territorial in A, seen territorial in B, seen alive, and 3 states: territorial in A, >territorial in B, dead). The model works without the random effect. The iteration process did start, >but writing the model results resulted in an error. What it says is this:

>log10 of relative estimate error in integration: order16-15=-11.7216
>Non available biological parameter
>Index exceeds matrix dimensions.

It is not an error. The -11 is an indication for the precision, i.e. the number of zero value after the comma for the error (relative), below 7, it is quite good. It tells you that the precision is fine. Otherwise, E-SURGE only computes the mathematical parameters, not the biological parameters.

TO BE IMPROVED !!

Sincerely,

Rémi

>It estimated 0 parameters.

[roos]

Hi Rémi,

Thank you for the answer. I'll will be eagerly awaiting the next version of E-SURGE! Do you have a prognoses when it is ready?

For now I'll select randomly one per pair of which I have two individuals, run the models multiple times and bootstrap the results.

Thanks,

Rosemarie.

[CHOQUET]

The new version of E-SURGE will be available in the beginning of april.