Page 1 of 1

Random effects or clustered occupancy in Rpresence?

PostPosted: Tue Dec 19, 2017 4:28 pm
by Chris Smith
My name is Chris Smith, a biologist currently working with some occupancy data on neotropical river otters. I have data from 10 river systems from Costa Rica, each with 4-6 segments per river; each segment is 450 m long. The segments are back to back, which creates problems with spatial autocorrelation (they are basically clustered in a linear format). I have been posting on the unmarked forum on google, and have had two recommendations: using river as a random effect or using Jim Hines' "Tigers on Trails: occupancy models for cluster sampling". I have been poking around package Rpresence for either of these and haven't found any sign of them. I was curious if
1) anyone knows if random effects can be used in Rpresence (and if so is it simply (1|river)? or
2) if any function for the clustered occupancy work had been introduced into Rpresence

Thanks in advance for any help,
Chris

Re: Random effects or clustered occupancy in Rpresence?

PostPosted: Tue Dec 19, 2017 4:34 pm
by darryl
Hi Chris,
1) No, there's no random effect capability in RPresence (or PRESENCE) at this stage. Not sure about other packages, but easy to do in OpenBUGS (or similar), if you know how.
2) It's in there. Referred to as the 'correlated detections' model. Do you have repeat surveys of your segments, or are they your repeat surveys for otter presence in the river?
Cheers
Darryl

Re: Random effects or clustered occupancy in Rpresence?

PostPosted: Tue Dec 19, 2017 5:24 pm
by Chris Smith
Hey Darryl,
Thanks by the way for creating this package! The way other several recent otter papers have done "repeat visits" is by subsegmenting each segment and counting each "subsegment" as a different visit. For me, I have 10 rivers with several 450 m long "sub-segments" within a single river, with each segment visited once. It seems like an exact copy of the "Tigers on Trails" proticol to me.

I read through the single season correlated detection occupancy model example and "Tigers on Trails" paper and had a quick question if you have more time. Can you explain the theta or "secondary-scale occupancy probability" and th0pi "probability of local-use before 1st segment"? From what I gather this is the full Markov model from the paper, and they are decomposing capture probability into animal is present (th0pi?) and animal is detected given it is present (theta?). Is there a way to run the Trap Response model? (my model has 10 sites and 6 visits, so I am worried about the amount of data).

Another, more general question as well: Can survey-scale variables be used in the psi part of the model? I think they should, but I have loaded a pao object successfully, and can run models with psi~1, p~var1, where var1 is a survey-level variable, but not psi~var1, p~1.
Error in eval(predvars, data, env) : object 'var1' not found


Also just an FYI for other users. I had my working drive set to an external hard drive, and it kept crashing R when RPresence ran any model. Saving it to my computer instead seems to have alleviated that problem.

Thanks again for all your help,
Chris

Re: Random effects or clustered occupancy in Rpresence?

PostPosted: Wed Jan 24, 2018 8:45 pm
by cooch
darryl wrote:Hi Chris,
1) No, there's no random effect capability in RPresence (or PRESENCE) at this stage. Not sure about other packages, but easy to do in OpenBUGS (or similar), if you know how.
2) It's in there. Referred to as the 'correlated detections' model. Do you have repeat surveys of your segments, or are they your repeat surveys for otter presence in the river?
Cheers
Darryl


You can do random effects in MARK, for single- and multi-season occupancy designs, using either a methods of moments approach (Appendix D in the MARK book), or the MCMC capabilities in MARK (Appendix E in the same book). But, in order for this to work, you need enough groups (say, 10 or more territories, or some such) for the estimation of \{\mu, \sigma\} to have much meaning.