Model-averaged confidence intervals

posts related to the RPresence library, which may not be of general interest to users of 'classic' PRESENCE.

Model-averaged confidence intervals

Postby jthomas » Wed Mar 21, 2018 3:34 pm

Hi there,

When using the modAvg function to predict model-averaged occupancy probabilities, how are confidence intervals calculated? I assume these are unconditional CI's that take model selection uncertainty into account, but are they calculated using the delta method? The same question applies for model-averaged real estimates (i.e. predict = FALSE).

Thanks very much!
jthomas
 
Posts: 4
Joined: Sun Feb 04, 2018 8:40 pm

Re: Model-averaged confidence intervals

Postby darryl » Wed Mar 21, 2018 3:54 pm

The CIs are calculated using the model-averaged estimates and their unconditional SE's (based on the revised SE calculation for MA estimates from B&A (2002)). The estimate and SE are transformed to the logit-scale using delta method, CI calculated on the logit-scale, and then the limits are back-transformed to the real (probability) scale.

Note that I'm not 100% convinced this is the best way of doing it, particularly for parts of the covariate space where different models give you quite different estimates. David Fletcher has done some work on MA CIs to try and account for these situations that I need to read up on one day.

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

Re: Model-averaged confidence intervals

Postby cooch » Wed Mar 21, 2018 4:41 pm

darryl wrote:The CIs are calculated using the model-averaged estimates and their unconditional SE's (based on the revised SE calculation for MA estimates from B&A (2002)). The estimate and SE are transformed to the logit-scale using delta method, CI calculated on the logit-scale, and then the limits are back-transformed to the real (probability) scale.

Note that I'm not 100% convinced this is the best way of doing it, particularly for parts of the covariate space where different models give you quite different estimates. David Fletcher has done some work on MA CIs to try and account for these situations that I need to read up on one day.

Cheers
Darryl


I'll need to read David's paper as well (reading Fletcher papers is almost always a smart thing to do), but it might be useful for you to read the verbiage in the MARK book on how the CI for model-averaged estimates is handled (and how it is either the same, or different, from what Darryl describes in his reply) -- see the -sidebar- on p. 55-56 of chapter 4:

http://www.phidot.org/software/mark/doc ... df#page=55
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Model-averaged confidence intervals

Postby darryl » Wed Mar 21, 2018 4:53 pm

Yep, the modAvg function uses the first of the 2 approaches outlined in that sidebar.
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Model-averaged confidence intervals

Postby jthomas » Thu Mar 22, 2018 12:07 am

Great, thanks to both of you for the quick replies!

-Julie
jthomas
 
Posts: 4
Joined: Sun Feb 04, 2018 8:40 pm


Return to RPresence

Who is online

Users browsing this forum: No registered users and 14 guests