GoF testing and trap-happiness

questions concerning analysis/theory using program PRESENCE

GoF testing and trap-happiness

Postby mparren » Mon Feb 18, 2019 3:52 pm

Hi all,

I am working on single-species single-season occupancy models where I have camera trap data from over 300 sites with 27 re-samples (24 hr period = sample). However, I am getting monster c-hat values for my global model when I assess model fit.

I had this issue earlier and realized that detection histories with many 1's seemed to be throwing my goodness of fit testing. Guessing this may have been a trap-happiness response that my covariates were not accounting for, I divided my detection histories into 3 day survey periods and got acceptable c-hat estimates. However, I ran into some later convergence issues I thought using my full detection histories might help resolve and am now trying to find a way to use all 27 re-samples.

I created a series of Markov-dependency/memory covariates to represent trap-response where the value=0 previous to a detection and then =1 following a detection for x # of days (up to all days following detection). Models with these covariates always performed better than models without them. Unfortunately, my c hat estimates are still enormous.

Does anyone have any suggestions as to what could be going on and what I need to do to move forward? I am pretty certain my covariates account for most variation and that trap-response is my issue given the c-hat values near 1 when I used 3 day samples.

Thanks!!

-Woefully-stuck grad student
mparren
 
Posts: 3
Joined: Fri Jan 25, 2019 7:28 pm

Re: GoF testing and trap-happiness

Postby darryl » Mon Feb 18, 2019 4:26 pm

Hi Molly,
When you say
I divided my detection histories into 3 day survey periods

do you mean that you aggregated your data down from 27 surveys to 9 surveys? Or did you split your single season of 27 surveys, into 9 'seasons' with 3 surveys each (so still 27 surveys in total).

What exactly are your "convergence issues". PRESENCE gives a warning about possible convergence issues, although sometimes the results are probably fine. This is discussed in a FAQ that use to be a sticky post on this forum (haven't looked recently).

It could be trap response, or could also be general heterogeneity. There's also the correlated detection models that you might consider.

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

Re: GoF testing and trap-happiness

Postby mparren » Mon Feb 18, 2019 5:02 pm

Hi Darryl,

Thanks so much for the speedy response! I apologize for my vagueness: I aggregated my data from 27 surveys to 9 surveys so if a species was detected at any point in a 3 day period = present.

My "convergence" issues actually arose later when I was doing conditional two-species modeling, which is my ultimate goal. I had some real estimates and standard errors of 0 that I hoped might be improved by more samples- or so had been suggested to me by people far more experienced than I. Additionally, I'm actually using MARK for the majority of my modeling, just because I am more familiar with the program and its outputs. I'm primarily using PRESENCE for the goodness of fit testing.

Thanks for the suggestions, hopefully I can figure something out.

Best,
Molly
mparren
 
Posts: 3
Joined: Fri Jan 25, 2019 7:28 pm

Re: GoF testing and trap-happiness

Postby jhines » Fri Feb 22, 2019 9:16 am

Hi Molly,

How are you doing GOF? Are you doing each species separately in a single-season model with the aggregated data? I'd be happy to take a look at the GOF results if you'd like.

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