POPAN link function

questions concerning analysis/theory using program MARK

POPAN link function

Postby X33277 » Sun Jan 08, 2023 1:59 pm

Hello,
I have a question regarding the POPAN formulation of the JS model regarding the link function. Since N is not bound between 0 and 1 the log (0<) or identity (-8 to 8+) is recommended. In chapter 12 a POPAN example is given using the log link and another with the identity link. My questions is how to decide between the links. I get very different estimates of N using the log link (N=1379 SE 56) compared to the identity link (N=944 SE 4.6). The estimate of survival is 1.7% less using the log link and is associated with higher estimates of p. Dividing # of captures/ p provides similar abundance estimates within each model with the POPAN estimates but not between them with the log link always having higher estimates of abundance. The trend in abundance is similar between the links except in the last 6-8 years of the study where the log link indicates an increasing trend in abundance and the identity link shows a declining trend in abundance. The derived abundance estimates from the log link are double that of the identity link in the last few years of the study.
The model is phi(.) using the logit link, p(t) using the logit link, PENTS(t) using the MLOGIT(1) link and N using the log or identity link. The identity matrix was used and simulated annealing was implemented.

Side note is there is harvest that I thought may bias the estimates of phi low so I built a MS CJS model using a live and dead state. Harvest was low with psi = 0.012. The estimate of phi was 0.902 which was almost the same for the identity link model while the estimate was lower 0.886 for the log link. The difference between the estimates of N for the log and identity link was the same even if I fixed the estimate of phi to 0.902.
Just wondering if there are any suggestions on when to use log vs identity link for POPAN and if having large differences in estimates of N is common between the log vs identity link. I also tried the Burnham JS model but had issues with convergence.
Any thoughts or suggestions are greatly appreciated.
Kind regards,
X33277
 
Posts: 10
Joined: Tue Mar 12, 2019 9:02 pm

Return to analysis help

Who is online

Users browsing this forum: No registered users and 7 guests