SE estimates of 0.0000 / 31622776601.683792

[ReneeLorica]

Hi,

I've been getting SE values of 0.00000 from Individual site estimates of psi, gamma, eps and p, while I get SE's like this:
estimate std.error
A1 : a1 37.936569 (31622776601.683792)
B1 : b1 -16.471556 (31622776601.683792)
C1 : c1 -38.221279 (8029694.720004)
D1 : d1 -36.051331 (4703592.478897)
D2 : d2Obs 76.471443 (31622776601.683792)

for Untransformed Estimates of coefficients for covariates (Beta's). Can I trust these values? Or did I make a big mistake somewhere along the way? Thank you.

[darryl]

You should search the forum. Your real estimates are essentially 0 or 1, right on the boundary of allowable values which causes problems when estimating SE's. You may have sparse data, small sample size problems, or the estimates could be right.
Darryl

[ReneeLorica]

Darryl,

Thank you for your response. I do have a small sample size so that could be the problem. Are my results still useable though?

Renee

[darryl]

How small a sample? Results seem to be suggesting you have 100% occupancy, no colonization (which you can't have with 100% occupancy anyway) and no extinction. Detection depends on how you've coded up your observer covariate.

[ReneeLorica]

I only have 26 sites, surveyed over 3 seasons by 3 observers. Here's how I coded my Observer covariate:

1-1 1-2 1-3 2-1 2-2 2-3 3-1 3-2 3-3
1 0 0 - 0 0 - 0 0 -
2 0 0 - 0 1 - 0 0 -

Did I do it wrong? Thanks for your response.

[darryl]

What do the 0's and 1's represent in your Observer covariate? How did you conducted the 3 surveys within each season?

Just looking at your detection data, in the first year did you have a detection at each of your 26 sites? Or did you actually have very few detections of your species?

[ReneeLorica]

The 0's and 1's represent the observer which detected the sign. Did I do this wrong?

Each observer walks different transect each until all transects have been surveyed for that season and records their observations for each transect so I have 3 observer data per transect per season.

In the first sampling season, only 8 sites have positive detections; second season has 7; third season has 5.

[darryl]

Ok, I'm going to try and pull the various threads together here.

Presumingly that that you want to include observer as a covariate because you think some people are better at finding leopard cat sign that others, you need to set up the covariate so that it indicates which survey was conducted by each observer, not which detection was due to each observer. Have you set up your data such that each column represents a particular observer in each year? ie column 1 is detection data from observer 1, column 2 for observer 2, etc? If so, then you don't need to define a covariate for observer to include an observer effect in a model as, in your case, setting up a 'survey-specific p' model is saying that the detection probability is different for each column of data hence different for each observer. Of course if you haven't set your data up like that, then that won't work. If you've defined your observer covariate to indicate which observer detected the cats, that will go a long way to explaining your weird results as that covariate will only =1 when you had a detection, and will always =0 for a nondetection, so will perfectly correlate with your detection data. Do estimates look more reasonable for a model without the observer covariate?

How many detection in total were there for each season?

On your spatial correlation question, no-one has put together the model that you're looking for (that I'm away of), a multi-season, multi-method (or multi-observer) model with correlated replicate surveys (from the segments).

[ReneeLorica]

Darryl,

I think I got it the covariates fixed. I'm getting much better SEs and CIs.

Total detections are:
Season1 9
Season2 7
Season3 6

Not very much, actually. Thanks again!

[ReneeLorica]

Hi,

I've fixed my observer covariate already though from time to time I still get huge SE values for some models like this one:

psi(cane,althab),gamma(.),eps(.),p(.)

estimate std.error
A1 : a1 -95.791818 (2.520458)
A2 : a2Cane 97.002338 (2.517953)
A3 : a3AltHab -1.636560 (1.495930)
B1 : b1 -62.635125 (31622776601.683792)
C1 : c1 -0.949948 (0.627018)
D1 : d1 -0.435813 (0.318086)

What does this mean? Can I still use this model?

Thank you!