Page 1 of 1

Contraints on first and last p in POPAN models

PostPosted: Fri Nov 18, 2022 5:24 pm
by stshroye
I know that in fully time-dependent POPAN models, the first and last p need to be constrained in order for the superpopulation estimates to be valid. Schwarz and Arnason (1996) suggest fixing p(1) = p(k) = 1. However, is there any reason why they should not be fixed to something more informative? For example, what if there is evidence from CJS model {phi(.) p(t)} that the last p is much < 1?

Re: Contraints on first and last p in POPAN models

PostPosted: Wed Jan 24, 2024 7:42 pm
by jlaake
But your example depends on phi being constant and you don't know that. When you follow A&S suggestion, the last survival is simply the product of phi and p for that occasion. Likewise for occasion 1, it is p(1) and pent(1) that are confounded. A better solution is to use covariates or other information that lets you use other constraints like p(1)=p(2) and p(k-1)=p(k).