Fish_Boy wrote:Has anyone ever calculate a negative C-Hat estimation for CJS? I assume this has something to do with overparameterized model for phi(t)p(t)? Any advice would be greatly appreciated.
10000 449;
10001 5;
10010 4;
10100 2;
10100 -1;
10110 -1;
11000 5;
11000 -1;
11001 2;
11010 1;
11011 2;
11100 4;
11101 1;
Fletcher chat {phi(t)p(t)} = 1.7452813
RELEASE: 3.3157/6 = 0.553
median c-hat (10 design points, 100 replicates per design point, upper bounrd = 3) = 3.45
If you let median c-hat go from 1-5 (default), you can get a negative value.
The significant variation in estimates among the 3 methods (above), reflects the fact that you have basically no data, and you're working with only a single cohort. If you look at the reduced m-array (below), you have ~9% total recaptures. There isn't enough there for MARK to simulate the bootstrapped samples, which is why the median-c-hat is 'flaky'. And, the Fletcher c-hat has issues with losses on capture, so that is suspect. This is also reflected in the real parameter estimates, which are basically trash, by and large.
Best you can do is set p = constant, and hold your nose. But hold it tight, because even estimates from {phi(t)p(.)} are basically garbage.
- Code: Select all
Group 1 Group 1
Occ. R(i) j= 2 3 4 5 Total
--- ------ ----- ----- ----- ----- -----
1 478 16 4 4 5 29
2 15 5 3 2 10
3 8 1 1 2
4 7 2 2
5 10 0