Fixing p for survey period with no detections

questions concerning analysis/theory using program PRESENCE

Fixing p for survey period with no detections

Postby Daisy » Sun Nov 22, 2015 1:46 am

Hello,

I am attempting a single-season occupancy model with 3 survey periods. In the third survey, there were no detections (I have a column of only 0s). The first two columns have a combination of 1s and 0s.

I found a previous post (http://www.phidot.org/forum/viewtopic.php?f=11&t=2299&p=7125&hilit=no+detections+in+survey#p7125) about this issue, but need a bit of clarification on this quote:
jhines wrote:If you have covariates, you won't be able to model the surveys with no detections as a function of the covariates as there is no variation in detection to model. If there is very little variation in the covariates, or sparse data, models with covariates might not converge.


I would like to model survey-specific covariate effects on p for this single-season model, and obviously this will also be an issue when I attempt multiseason models using this data set, although all other survey periods for other years have detections.

So in order to do models of the form psi(.),p(covariate), with what value should I fix the parameter p[3]? And will the design matrix be as follows?

psi: a1=1
p1: b1=1, b2=covariate
p2: b1=1, b2=covariate
p3: b1=0, b2=0

Thank you for your time!
Daisy :)
Daisy
 
Posts: 9
Joined: Sat Feb 07, 2015 7:38 pm
Location: Michigan, USA

Re: Fixing p for survey period with no detections

Postby jhines » Sun Nov 22, 2015 9:47 am

Your design matrix is correct. That will model detection to be the same for surveys 1 and 2 and be a function of your covariate. The 3rd p will be fixed to zero. PRESENCE assumes if a row of the design matrix contains only zeros, that the parameter is fixed to zero, so there is no need to explicitly fix the parameter to zero using the 'fix parameters' button in this case. If you want to fix a parameter to some other value, then you would use the 'fix parameters' button to give it a value.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Fixing p for survey period with no detections

Postby darryl » Sun Nov 22, 2015 3:45 pm

Hi Daisy,
By doing as you suggest the implication is that you think there's a covariate relationship for detection that holds in the first 2 surveys, but not the third. Was there something quite different about the 3rd survey that would make this assumption reasonable? Note that because you've go the same covariate effect for all survey periods (ie your covariate name is in the 1 column for the whole design matrix) then provided there's some detections in the first 2 surveys it MAY be fine to also use the covariate for the third survey.; the covariate values may suggest that detection probability was really low in the 3rd survey so not surprising there's no detections. Jim's comment is more of an issue if you were trying to estimate a separate covariate effect for that last survey. Note that a similar issue applies when you have all 1's as well (ie no variation to model).

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

Re: Fixing p for survey period with no detections

Postby Daisy » Sun Nov 22, 2015 7:18 pm

Thank you both for your answers.

I'm studying a watersnake that anecdotally is more detectable earlier in the season. There is little vegetation around the wetlands in spring, and as the season progresses plants leaf out and it is harder to detect snakes in dense vegetation. I didn't measure that as a covariate though. I believe that detection is also a function of temperature (which I'm using as a covariate) - the hotter it is, the less likely snakes are to be out in the open basking, and temp increases as the season goes on.

So a combination of factors (which are correlated with progression of the season) likely influence detectability.

I think it makes more sense based on this to not fix p[3], because temp as a covariate probably does impact there being no detections in the third survey.

:?: If p[3] were fixed to zero, is that essentially the same thing as just omitting the 3rd survey's values for observations and covariates from this analysis altogether?

Thanks again for your helpful responses!
Daisy
Daisy
 
Posts: 9
Joined: Sat Feb 07, 2015 7:38 pm
Location: Michigan, USA


Return to analysis help

Who is online

Users browsing this forum: No registered users and 8 guests