Occupancy rate of 1, CI [0,1]

questions concerning analysis/theory using program PRESENCE

Occupancy rate of 1, CI [0,1]

Postby Jaranlanae » Tue Mar 29, 2022 5:35 pm

Hi,

I have encountered a problem with my occupancy analysis that I am hoping to get some guidance on. I am doing an occupancy analysis of five small mammal species across five plots. I am performing this analysis in R using the unmarked package and the AICcmodavg package.

The issue I am encountering is in regards to the occupancy estimate confidence intervals. This issue only applies to one species. For this species, I get reasonable occupancy estimates across the five plots. Plot was included as a site covariate in the top competing models so the predicted occupancy rate varies across the plots. For two plots (Sand_sheet_east & Sand_sheet_west), this species was detected very often. The detection histories for these two plots are almost all 1's, with only a few 0's. So the occupancy probability is basically one, with a very small standard error. This seems to make sense. What I do not understand is why the confidence interval goes from 0 to 1, and not ~ 0.98 to 1. I posted the predicted occupancy output below.

mod.avg.pred uncond.se lower.CL upper.CL. Plot
1 0.726145 0.083266 0.538517 0.857657 Cottonwood
2 0.550194 0.089675 0.375419 0.713412 Acacia
3 0.768903 0.077462 0.586047 0.886617 Heather
4 0.999989 0.000780 0.000000 1.000000 Sand_sheet_east
5 0.999993 0.000685 0.000000 1.000000 Sand_sheet_west


Thanks in advance for any help,

Jaran
Jaranlanae
 
Posts: 4
Joined: Tue Mar 29, 2022 3:16 pm

Re: Occupancy rate of 1, CI [0,1]

Postby darryl » Tue Mar 29, 2022 5:59 pm

I'm guessing that unmarked is calculating confidence intervals on the logit-scale, then back transforming to the real scale. This tends to work fine, except when you have real estimates that are very close to 0 or 1 because on the logit-scale, the SE tends to be really, really big, which leads to really wide intervals on the logit-scale and how you end up with real-scale CI or 0-1.

It's a case where you have to ignore the output, and/or try other CI methods (eg bootstrapping, profile likelihoods, etc) if you're not satisfied by saying it's a range that is very close to 1 (in this case)

Cheers
Darryl
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Occupancy rate of 1, CI [0,1]

Postby egc » Tue Mar 29, 2022 6:05 pm

Jaranlanae wrote:Hi,
I am performing this analysis in R using the unmarked package and the AICcmodavg package...


Darryl is too polite to ask, but I will -- why aren't you asking this question on the unmarked Google group? https://groups.google.com/g/unmarked

The group is very active, and there are any number of unmarked users (including most of the developers) who routinely reply to posted questions.
egc
Site Admin
 
Posts: 201
Joined: Thu May 15, 2003 3:25 pm

Re: Occupancy rate of 1, CI [0,1]

Postby Jaranlanae » Thu Mar 31, 2022 2:35 pm

Darryl,

Thank you so much for your response. Your knowledge and work is inspiring.

Egc,

I apologize for posting on the wrong group. I am new to these group threads and getting help through them. I did not realize there was a more appropriate option. Thank you for passing on that info.

-Jaran
Jaranlanae
 
Posts: 4
Joined: Tue Mar 29, 2022 3:16 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 2 guests

cron