identifiable parameters POPAN

[Neus]

Hi,
I'm using POPAN through MARK to estimate population size. The better model fitting my data is the following: Phi(t) p(.) pent(t) N
I'm getting bad values in some of the confidence intervals (see document below) and they aren't one of these non-identifiable parameters on POPAN, like final survival and catchability, initial entrance and catchability.
Furthermore, the estimates and standard error seem to be correct, but the interval confidence isn't good. I've taken a look to my data input and I have enough
recaptures in this occasions, so the program should be able to calculate the interval, right? Do you have any idea about these results?
Thanks a lot in advance for your help,
Best,
Neus

Real Function Parameters of {Phit p. pentt N}
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
------------------------- -------------- -------------- -------------- --------------
1:Phi 0.8021802 0.0969556 0.5504390 0.9307011
2:Phi 0.5825828 0.0884451 0.4062545 0.7400525
3:Phi 0.6953082 0.0935921 0.4897954 0.8443468
4:Phi 0.6403141 0.0780853 0.4780657 0.7757813
5:Phi 0.7940056 0.0602867 0.6517651 0.8881196
6:Phi 0.7271820 0.0888791 0.5255480 0.8651176
7:Phi 0.4572373 0.0765912 0.3151029 0.6066925
8:Phi 0.9891431 0.0948781 0.2748759E-005 1.0000000
9:Phi 0.6559149 0.0834476 0.4801216 0.7973542
10:Phi 0.9549631 0.1555741 0.0173668 0.9999607
11:Phi 0.4262651 0.1517626 0.1804391 0.7148716
12:p 0.3676606 0.0522541 0.2723366 0.4745889
13:pent 0.0340735 0.0362285 0.0040619 0.2337794
14:pent 0.1165290 0.0266389 0.0735851 0.1796741
15:pent 0.2258782 0.0355393 0.1638169 0.3029333
16:pent 0.0040281 0.0278998 0.4867249E-008 0.9997025
17:pent 0.1946594 0.0319722 0.1394638 0.2649734
18:pent 0.0678945 0.0313643 0.0268342 0.1613653
19:pent 0.0033718 0.0120553 0.2989614E-005 0.7928975
20:pent 0.0858931 0.0197687 0.0542521 0.1333851
21:pent 0.2241176E-006 0.1241583E-003 0.1596743E-223 1.0000000
22:pent 0.0192333 0.0136638 0.0047190 0.0750242
23:pent 0.0220939 0.0110282 0.0082394 0.0578849
24:N 624.53815 59.865398 528.04326 766.22387

Net Birth+Immigration Estimates of {Phit p. pentt N}
95% Confidence Interval
Grp. Occ. B-hat Standard Error Lower Upper
---- ---- -------------- -------------- -------------- --------------
1 1 21.280213 22.232756 -22.295990 64.856415
1 2 72.776820 18.500577 36.515688 109.03795
1 3 141.06956 27.648923 86.877665 195.26145
1 4 2.5157240 17.364314 -31.518333 36.549780
1 5 121.57222 24.899195 72.769802 170.37465
1 6 42.402680 19.507275 4.1684208 80.636939
1 7 2.1057957 7.5060559 -12.606074 16.817666
1 8 53.643505 13.566636 27.052897 80.234112
1 9 0.1399700E-003 0.0775423 -0.1518429 0.1521228
1 10 12.011920 8.4623683 -4.5743221 28.598162
1 11 13.798513 7.1106045 -0.1382725 27.735298

[sbonner]

Neus,

The problem is that the parameter estimates are at the boundary of their range (i.e., close to 0 or 1 for phi's or pent's). The asymptotic intervals computed by default in MARK are not appropriate when this is the case. One solution is to use the option to compute profile likelihood intervals instead. If you search for "boundary + profile likelihood" on this forum you will come up with numerous previous posts discussing this issue. You should also read the help on profile likelihood confidence intervals in MARK and section 1-22 in the Gentle Introduction to MARK.

Hope that helps,

Simon