Conditional Occupancy versus Model Averaging

questions concerning analysis/theory using program PRESENCE

Conditional Occupancy versus Model Averaging

Postby jhubbell072786 » Sat Oct 19, 2019 10:50 pm

Hi Presence Users,

Question,

One of the things I've noticed throughout my time using this forum is that often in published studies, the researchers will either present model averaged estimates of psi and p or conditional estimates of occupancy (based on the output generated by PRESENCE for the best model among the candidate set). Is there a preference among reviewers, or does it matter? Obviously model average estimates reflect average detection and occupancy probabilities at a given site, whereas conditional occupancy is reflective of the PAO.

Sincerely,

Josh
jhubbell072786
 
Posts: 9
Joined: Sun Apr 09, 2017 9:04 pm

Re: Conditional Occupancy versus Model Averaging

Postby jhines » Mon Oct 21, 2019 7:45 am

I think there may be some confusion about the definition of "conditional" occupancy. In Presence, the "conditional" occupancy estimates are estimates of occupancy, given the detection history for each site. In a single-season model, any site with at least one detection will have a "conditional" occupancy estimate of 1.0 and sites with no detections will have a value < 1.0. These estimates can be computed for any/all models in the model-set, so you could use model-averaging on these just as you would for the "unconditional" estimates.

Estimates from any model are conditional on the model, so if you have a few models which are equally (nearly) likely, model-averaging the estimates allows you to present estimates which also include model uncertainty.

So, your question should really be 2 questions: 1) Should I use "conditional" occupancy estimates or "unconditional" estimates? 2) Should I use estimates from the top model or model-averaged estimates? The answer to the first question depends on what you want to convey. If it's information only about surveyed sites (eg., perhaps a map of occupancy of an area), then the conditional estimates may be more appropriate. If it's about any site with characteristics similar to covariates in the model, then the unconditional estimates are more appropriate. The answer to the 2nd question also depends on what you want to say. If it's about the relationship of occupancy and some covariate(s), you might give the beta estimates for the effect of that covariate. If it's about occupancy in general on a site-by-site basis, the model-averaged estimates may be best.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Conditional Occupancy versus Model Averaging

Postby jhubbell072786 » Wed Oct 23, 2019 9:00 pm

Jim,

Thanks for the response. And yes, in my definition of conditional occupancy, I meant it in reference to the individual detection histories for each site (my apologies if that was unclear).

I was unaware that you could perform model averaging on the conditional estimates. How would a user tell PRESENCE that he/she wants to perform model averaging on the conditional estimates specifically?

Thanks,

Josh
jhubbell072786
 
Posts: 9
Joined: Sun Apr 09, 2017 9:04 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 17 guests