Large SE's for Psi after detection is modeled

[BDWilliams]

I'm currently in the process of modeling some data for a striped bass survey. I'm using the second parameterization of the multi-season models (seasonal occupancy and colonization, detection). I began by focusing my efforts on modeling p. Once the 'best' model for detection is identified, it is then used to model the occupancy parameter in order to determine the best overall model. However, I keep encountering an issue with very large SE's for the beta estimates for occupancy and colonization. Is anyone familiar with this?

This is a portion of the output...

Untransformed Estimates of coefficients for covariates (Beta's)
==============================================================================
estimate std.error
A1 ccupancy psi1 20.404409 (31622776601.683792)
B1 :colonization gam1 41.759362 (31622776601.683792)
D1 :detection P[1-1] 5.229681 (1.677108)
D2 :detection P[2-1] 2.810364 (1.490426)
D3 :detection P[1-1]DIST 0.023930 (0.010652)
D4 :detection P[1-1]TEMP -0.171684 (0.059870)

[jhines]

The first thing to check is if the estimates make sense for your data. Since the beta for occupancy is 20, it's saying that occupancy is 1.0. Is this reasonable for your data?

When parameters are estimated near the boudary (0 or 1), often the standard error cannot be estimated. You might say the standard error is undefined in this case.

Since occupancy is 1.0, then there are no unoccupied sites. So, the colonization rate (prob that an unoccupied site becomes occupied) cannot be estimated. That's why the standard error for gamma is high, which also might be the reason the first standard error is high.

I'd suggest trying the default parameterization (psi,gam,eps), or the other parameterization (psi(season),eps).

Jim

[darryl]

In addition to Jim's comments:

What does you data look like? Is it very sparse or do you tend to have a reasonable number of detections? Do you have at least 1 detection at almost all of your sites? Are you estimates of p very small? Did you standardize your covariates? Have you tried different starting values?

Cheers
Darryl

[BDWilliams]

Thanks for your replies. Jim, occupancy in the first season is 1.0 but it's less than 1.0 in the second season (however it's still quite high). This was our reasoning behind using the multiseason model parameterization.

Darryl, I have quite a few detections so sparse data shouldn't be an issue. Most occupied sites have multiple detections. I tried standardizing some covariates (mean length was an issue) to no avail. Could this be another issue with very high occupancy rates clouding the true effect of some of my covariates?

[darryl]

The model results you presented only had 1 occupancy parameter indicating that the model you were trying to fit had occupancy as the same in both years (=1.0). If that's the case then it's not possible to observe colonizations or extinctions. As Jim suggest, try a different parameterization and see if that helps. Note that if occupancy really is 1, in the first year you may have to fix some parameters (eg gamma=0).