SE estimates of 0 or 10.00000

[nholoubek]

Hello,

Has anyone encountered standard error estimates of 0.00000 or 10.00000? There are no warning or error messages accompanying these, but my instinct is to not trust them. While it would be great if the estimates were actually good enough to warrant a standard error < 0.00000, I expect these models are not valid. The 10.00000 SE appears multiple times in my models, it seems odd to have a SE of exactly 10 for several models and several species.

I am modeling bird site occupancy across 2 seasons for 24 species. I am using 3 continuous vegetation variables. Some species model very well, while others model poorly, depending on detection history.

Any information is greatly appreciated!

Thank you,
Nathan

[darryl]

Hi Nathan,
Are these SE's associated with real or beta estimates in the output? What model and parametrization are you using?
Darryl

[nholoubek]

Hi Darryl,

These are for the beta estimates. I am using the multiseason model and for the most part the seasonal occupancy parameterizations (options 2 or 3). I have been trying different parameterizations because I have been getting invalid results for others (insufficient significant digits, SEs in the thousands, etc.). I read earlier in the forum that if I was having trouble with one parameterization to try others.

Thanks,
Nathan

[darryl]

It's actually these two parametrizations that tend to cause problems because there are naturally constraints that have to be enforced, particularly when you're trying to include covariates in the models. This has been discussed a number of times here on phidot. Generally the 1st parametrization is the most numerically stable. Note that the problems you mentioned with other parametrizations do not necessarily mean your results are invalid, check out the PRESENCE FAQ here on phidot. Furthermore, the problems may not stem from the model, but from your data in which case changing parametrization may not achieve much.

All that aside, the fact you're getting SE's of exactly 10.00000 using the 2nd or 3rd parametrization would lead me to suspect that you're banging up against some of the constraints that have to be enforced for these models, so the optimization routine used to obtain estimates may be getting 'stuck' and not converging properly. Check to see whether any of your colonization or extinction probabilities (one of which will be derived at towards the end of the output) are 0 and or 1. You could try to specify different initial values and see if the results come out the same.

Cheers
Darryl