Integrated habitat models and random occupancy eps=(1-gam)

questions concerning analysis/theory using program PRESENCE

Integrated habitat models and random occupancy eps=(1-gam)

Postby Eurycea » Fri Jul 31, 2015 4:16 pm

Is there a way to run random occupancy models within an integrated habitat framework? There is a drop down box for this in the regular multi-season models but not in integrated habitat models.

Nate
Eurycea
 
Posts: 99
Joined: Thu Feb 25, 2010 11:21 am

Re: Integrated habitat models and random occupancy eps=(1-ga

Postby Eurycea » Wed Aug 05, 2015 6:08 pm

Fiddling around it seems that I can code -1 in the DM to accomplish this. However, to be sure, I ran a single season model using the -1 method vs. specifying the seasonal occupancy parameterization with provided NSO example data. The seasonal occupancy blows up with crazy high AIC. The -1 results seem ok though. Maybe I need a clean file to run the alternative parameterization?
Eurycea
 
Posts: 99
Joined: Thu Feb 25, 2010 11:21 am

Re: Integrated habitat models and random occupancy eps=(1-ga

Postby jhines » Thu Aug 06, 2015 10:13 am

I'm not sure I understand your question completely, but I'll attempt to answer. Since we're using the logit link, you can set one parameter (eps) to be equal to 1-gam by using -1 in the design matrix in the same column as the '1' for gam. This is because the logit scale is centered around 0.5, and the real scale is centered around zero.

One problem with the 'random occupancy' model is that since you're setting eps=1-gam (colonization compensates perfectly for extinction), it implies that occupancy is constant from season to season. If the data don't support this, the model will have a hard time converging on reasonable estimates.

How did you run the single-season model with multi-season data? If you treated all surveys as occurring in 1 season, where does the '-1' come in? I'm referring the the '-1 results' which you said were OK.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Integrated habitat models and random occupancy eps=(1-ga

Postby Eurycea » Thu Aug 06, 2015 11:02 am

For the initial question, I did figure out the -1 trick, thank you. So I believe I can now run the random model with the integrated habitat option as gamAA=gamBA=(1-epsAA).

I understand that the random model may not be very good and I may not include it in the final model set. But I wanted to be sure I could run it if I needed to. Trying to get all my ducks in a row before running through the candidate model set, which I have not finalized yet.

But I'm a bit confused about what you mean by occupancy being constant from season to season with an eps=1-gam model. I thought the random model would allow occupancy to vary seasonally, but that the change in occupancy by site is not dependent on the previous state. Isn't constant occupancy with gam=eps=0?

I misspoke about the single season model. I meant to write multi-season single species (the default multi-season model).

I coded a DM using the Init Occ, colonization, extinction, detection parameterization, using the NSO example data, like this:

Code: Select all
psi1 1 0 0 0 0
gam1 0 1 0 0 0
gam2 0 0 1 0 0
gam3 0 0 0 1 0
gam4 0 0 0 0 1
eps1 0 -1 0 0 0
eps2 0 0 -1 0 0
eps3 0 0 0 -1 0
eps4 0 0 0 0 -1


With seasonal detection, I got a -2LL of 1429.32.

I ran then a seasonal occupancy, eps=(1-gam) model, coding the DM like this:

Code: Select all
psi1 1 0 0 0 0
gam1 0 1 0 0 0
gam2 0 0 1 0 0
gam3 0 0 0 1 0
gam4 0 0 0 0 1


With seasonal detection, I got a -2LL of 17308.24. Same # parms (10).

Edited to add: an equilibrium model gam(.) eps(.) with seasonal detection gives -2LL ~ 1337.

I expected the same results. What am I missing here?

Thank you,

Nate
Eurycea
 
Posts: 99
Joined: Thu Feb 25, 2010 11:21 am

Re: Integrated habitat models and random occupancy eps=(1-ga

Postby jhines » Thu Aug 06, 2015 12:41 pm

Sorry, my thought about eps=1-gam as implying constant occupancy is only true if initial occupancy is 0.5. You're right, that model just implies changes in occupancy don't depend on previous occupancy.

How did you create the DM for the 1st model (default multi-season, -2LL=1429.32)? Each parameter (psi, gam, eps, p) is on a different tab, so it's impossible to set eps = 1-gam. If you set the DM for eps to -1 for each of the 4 eps, PRESENCE will assign 4 different beta's since they are in different sections of the overall design matrix. The effect would be the same as if you just set the eps DM to identity (1's on diagonal), with the resulting beta's having the opposite sign. Also, that model would have 14 parameters (1 psi, 4 eps, 4 gam, 5 p), not 10.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Integrated habitat models and random occupancy eps=(1-ga

Postby Eurycea » Thu Aug 06, 2015 12:48 pm

Sorry, my thought about eps=1-gam as implying constant occupancy is only true if initial occupancy is 0.5. You're right, that model just implies changes in occupancy don't depend on previous occupancy.


Gotcha.

How did you create the DM for the 1st model (default multi-season, -2LL=1429.32)? Each parameter (psi, gam, eps, p) is on a different tab, so it's impossible to set eps = 1-gam.


I did it as above, using the "combine matrices" right click menu option in the DM, so all parms show up in the same matrix.
Eurycea
 
Posts: 99
Joined: Thu Feb 25, 2010 11:21 am

Re: Integrated habitat models and random occupancy eps=(1-ga

Postby jhines » Fri Aug 14, 2015 11:11 am

I've corrected a bug in the way PRESENCE handles a combined design matrix. (I forgot that feature was still in there. :) This should make the two models give the same results.

Jim
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Integrated habitat models and random occupancy eps=(1-ga

Postby Eurycea » Thu Sep 03, 2015 2:30 pm

Jim,

Can you confirm this bug is fixed for the integrated habitat model? If I run a gamAA=1-epsAA model the -2LL blows up. My DM for these parms looks like this:

Code: Select all
gamAA1   0   0   0   0   0   0   1   0
gamAA2   0   0   0   0   0   0   1   0
gamAB1   0   0   0   0   0   0   0   0
gamAB2   0   0   0   0   0   0   0   0
gamBA1   0   0   0   0   0   0   0   1
gamBA2   0   0   0   0   0   0   0   1
gamBB1   0   0   0   0   0   0   0   0
gamBB2   0   0   0   0   0   0   0   0
epsAA1   0   0   0   0   0   0   -1   0
epsAA2   0   0   0   0   0   0   -1   0
epsAB1   0   0   0   0   0   0   0   0
epsAB2   0   0   0   0   0   0   0   0
epsBA1   0   0   0   0   0   0   0   0
epsBA2   0   0   0   0   0   0   0   0
epsBB1   0   0   0   0   0   0   0   0
epsBB2   0   0   0   0   0   0   0   0



I've also tried it with gamAA=gamBA=1-epsAA. That won't run at all. PRESENCE says to check the .pao and .dm files. It also has an odd behavior where the latter model will run if I have empty column of 0's next to epsAA, but gives the same results as the prior model.
Eurycea
 
Posts: 99
Joined: Thu Feb 25, 2010 11:21 am


Return to analysis help

Who is online

Users browsing this forum: No registered users and 11 guests