Model Convergence and #IND000 Errors

[hoarybat]

Hello,
I am having a little trouble understanding my outputs for my "global" occupancy model.

Some background first:
I sampled 71 sites for 2 nights each. I have 9 site covariates and 4 sampling covariates. My global model has 13 parameters to it. The number of detections I have is generally low. I believe there are 44 detections out of 142 sampling periods.

The trouble I am having is this: the standard error values for psi, p, and the covariates all contain #IND000 errors. With simpler models I do not get these errors. The global model also happens to be my top model. I assume these errors are from the models not converging. The 2nd best model is the null model [psi(.)p(covariate1, covariate2, covariate3)].

My question is, is there anyway to get this global model to run? I found this information on the online HELP file and thought it may be the solution:

"In some cases, poor starting values for the parameters can cause the problems noted above. This can be solved by giving better initial values to the program when running the model. For example, if detection probabilities are very small, and the default starting values of 0.5 are far away from the final expected parmaeter values, the optimization routine may fail. The solution would be to input small initial values (on the logit scale) for the model so the optimization routine does not have to search very far. Since simpler models converge more readily than complex ones, it is usually best to start with simple models, so you have starting values for complex ones if needed. "

I was wondering if there was a way, or what method one must take to insert initial values so that I can get this global model to converge? How do you know what initial values to insert?

Or is this a lost cause because I have too few detections for the amount of covariates?

Any help or insight would be greatly appreciated. I need to present this information at a conference next week.


Thanks phidot community.

[Alan Dextrase]

I suspect convergence/SE problems may be related to over-parameterization of the global model given the sample size. Including intercepts, k=15. With 44 detections, your data are probably not too sparse, but you may want to limit your models to include a core set of covariates that are most likely to affect occupancy and detection. There may also be collinearity with the covariates.

If you do want to pursue setting initial values, you can try inserting values for covariates from simpler models that did converge with reasonable SEs using the ‘Supply initial values’ option on the ‘Setup Numerical Estimation Run’ screen.

Alan