global model underdispersion and gamma

questions concerning analysis/theory using program PRESENCE

global model underdispersion and gamma

Postby Artax » Thu Dec 31, 2020 12:38 pm

I created a global model in Presence for a single species multi season analysis. I have 4 seasons, 46 sites and 44 surveys total (11,13,13,7) . I ran the MacKenzie Bailey goodness of fit test using 10,000 parametric bootstraps. My results show a very underdispersed c-hat value of .08 but a P value of .5. According to chapter 4 of Occupancy Estimation and Modeling(2nd Edition) MacKenzie et al 2018 (in reference to Burnham and Anderson 2002) , underdispersion can be ignored. Burnham and Anderson 2002 seem to put an emphasis on the p value even if c hat is not close to 1. However I have read elsewhere that a c hat value this low should indicate that this model should not be used to make inferences. Could I still use this global model to continue fitting models?

My second question is about my real parameter gamma being 0 with a SE of 0. Would this also indicate that a model is a poor fit?

Thanks
Artax
 
Posts: 3
Joined: Wed Nov 18, 2020 4:07 pm

Re: global model underdispersion and gamma

Postby jhines » Thu Dec 31, 2020 3:40 pm

The GOF test for multi-season data has not been very well tested at this point. I've been working on it and accidentally enabled the option in Presence. The results you mentioned don't make much sense to me. If you
don't mind sending me the data, I'd be happy to see what's going on. I think the best option at the moment for GOF test would be to do the GOF test on the individual seasons.

An estimate of gamma=0 with SE=0 doesn't indicate poor fit. It indicates that colonization is very low. One thing to remember about colonization is that it is estimated only for un-occupied sites. So, if all sites are occupied in a season, there is no information available to estimate gamma for that season. (Same goes for epsilon, except that it is estimated for occupied sites). Any estimate of 0 or 1 will usually give a SE=0. If epsilon=0, or is very low, then a single-season model might fit the data fairly well, which would allow you to do a GOF test on the entire dataset at once.

Jim (jhines@usgs.gov)
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA


Return to analysis help

Who is online

Users browsing this forum: No registered users and 8 guests