Integrated habitat models: fixing and counting params

[Eurycea]

I have two questions related to integrated habitat models:

Q1: When I fix parameters, PRESENCE does not correctly count the number of parameters. How do I adjust this for correct AIC output?

Q2: What parameters need to be fixed in PRESENCE for an integrated habitat model of suitable and unsuitable habitat with three states (1) unsuitable and unoccupied 2) suitable and unoccupied 3) suitable and occupied?

I'll denote suitable as S and unsuitable as U.

MacKenzie et al. 2011 in their paper state for this type of model that psiU (occupancy in unsuitable habitat) is zero, and state that the gamma(S,U) and gamma(U,U) are also fixed to zero. Additionally epsilon(S,U) is fixed at one. This all makes sense to me. You can't have a unsuitable , and therefore, unoccupied habitat become colonized- also if a site becomes unsuitable, it has by definition, become extinct.

What doesn't make sense to me is why espilon(U,U) and epsilon(U,S) also are not fixed at zero. If a site is unsuitable, and therefore unoccupied, by definition, it cannot go extinct. Epsilon is only for occupied sites. Additionally, detection probabilities for sites that are not occupied will not get estimated. So why is p(U) not fixed at zero as well? These might not need to be fixed in practice because they automatically drop out of the likelihood, but in MY practice the only thing that seems to make a difference is fixing psiU to zero. I seem to get the same -2LL no matter if I set psiU to zero, fix parms in the previous paragraph, or all of those that I mentioned(however I am working with fake data so I could have screwed that part up). Any advice is appreciated.

[Eurycea]

Working from a known dataset (Exercise17 from the online book) I've been able to hone in on the problem a bit. It seems that PRESENCE gives the wrong answer whenever I fix any parameters, either for "Fix Beta" (which apparently helps with the parameter count problem) or fixing parameter values using the click box and menu, or both.

To achieve the correct results I have to ensure any parameters that the literature advises fixing are free to vary. Additionally I have to set all pS=pU. Doing this, I've been able to match the same -2LL scores and parameter estimates between the spreadsheet model and PRESENCE.

This doesn't solve my problem entirely, but it does seem to confirm there is some odd behavior of the software.

I can also confirm that fixing parameters seems to work fine with single season models. The number of parameters is correctly adjusted and the model results seem reasonable.

[jhines]

I've found the bug which was causing PRESENCE to give different (wrong) estimates from the exercise in the online book and corrected it. You don't need to constrain any beta's, just some of the real parameters. Here are the design matrices and fixed parameters:

Code: Select all

Matrix 1: rows=4, cols=3
            -,a1,a2,
pi            1 0
psiA          0 1
psiB          0 0
========================

Matrix 2: rows=5, cols=3
            -,b1,b2,
eta0AA1       1 0
eta1AA1       1 0
eta0BA1       0 1
eta1BA1       0 1
========================

Matrix 3: rows=5, cols=3
            -,c1,c2,
gamAA1        1 0
gamAB1        0 0
gamBA1        0 1
gamBB1        0 0
========================

Matrix 4: rows=5, cols=2
            -,d1,
epsAA1        1
epsAB1        0
epsBA1        0
epsBB1        0
========================

Matrix 5: rows=9, cols=2
            -,e1,
pA1           1
pA2           1
pA3           1
pA4           1
pB1           0
pB2           0
pB3           0
pB4           0
========================

Matrix 6: rows=0, cols=0
========================
fix(13)=1.000000 /* epsAB1 */
fix(14)=1.000000 /* epsBA1 */
fix(15)=1.000000 /* epsBB1 */
*psiB parm(3) fixed to 0.
*gamAB1 parm(9) fixed to 0.
*gamBB1 parm(11) fixed to 0.
*pB1 parm(20) fixed to 0.
*pB2 parm(21) fixed to 0.
*pB3 parm(22) fixed to 0.
*pB4 parm(23) fixed to 0.

Jim

[Eurycea]

Thank you so much!!! It works great now.

Nate