no convergence when using survey-specific covariates

[Diego.Pavon]

Hello,

I am running the 3rd parameterization (psi, eps, p) and I am having problems when constraining PSI with survey-specific covariates. There is no problem when constraining PSI with site covs. or any other type of covariates. I only found convergence issues with survey-specific covariates.
I have tried to run other parameterizations (e.g. psi, gamma, p), but I have found the same problem. I have also tried to put initial values (according to the "bad" estimates obtained from the other parameterizations) but the models do not converge.

I would like to model PSI as function of density (survey specific) and then P and Eps with some other covariates (site covs, survey covs,etc..).
Is the problem perhaps because of the nature of the covariate (survey-specific)? Even the simplest model, i.e. PSI(density), Eps(), p(), does not converge.

Does anybody have any suggestion or idea why this happens and how to solve it?

Thank you very much in advance.

Diego

[jhines]

The most common problem with the parameterizations where occupancy is season-specific (psi,eps,p or psi,gam,p) is when the optimization process tries parameters for psi and eps (or psi and gam) which are impossible. Although it will limit the psi's and epsilon's to be between zero and one, certain combinations of psi and epsilon lead to values of gamma which are not between zero and one. This leads to problems with convergence. If you aren't trying to model psi as a function of covariates, I usually recommend just using the 1st parameterization (psi,gam,eps,p). Since you wish to model psi as a function of a seasonal covariate, you can't do that. My next suggestion would be to try different sets of random initial values.

Other reasons for convergence failure are usually sparse data or a mis-specified model. If you send me the backup file (presence_backup5.zip), I'd be happy to take a look at it.

Jim