Mark-Resight Logit Normal w/ no individual ID's issues

[AlaskaMatt]

Hello,

After reading everything I can get my hands on regarding the Mark-Resight Logit Normal model in MARK for estimating abundance with no individually identifiable marks, I am finally at an impass with trying to figure it out "on my own". I think I am very close to getting the analysis to work, but I must have a small error somewhere. Fortunately my design is super simple, so I hope the answer is obvious to someone out there.

My basic design is 4 Primary occasions with 2 secondary occasons each (one real and one "dummy"). I have two groups and I will post my .inp file and results file below for those interested.

After creating the "artificial encounter history" and including "dummy secondary sampling occasions" to satisfy the requirements for a Robust Design, I am only getting a beleivable N estimate for one of my two groups. The error seems to be occuring during the 4th N estimate for group 2. The N it gives in the results is actually equal to the number of marked individuals in the population and is an odd jump from N's 1-3.

I have set all sigmas to "0", all p's equal for the real resightings, and all "dummy" p's to "0".
I have tested my artificial encounter history to make sure they meet the requirements stated in CH 18.
The artificial encounter histories and "dummy" sightings were created following advice from these two forum posts: [Artificial individual encounter history file] and [Robust design for marked populations?] along with guidlines laid out in "The Gentle Intro..."

If my question is unclear or lacking information, please let me know and I will do my best to clear things up. I would definitely prefer a terse reply to being ignored for my inability to communicate my problem well.

Thank you kindly,
-Matt

Here are my .inp files (original and adjusted) and my screwy results file:

Original .inp :
/* No Individual Marks 2 Groups */
/* 4 occasions, 4 primary, 1 secondary each */
/*Begin Input File CHUB SUCKER */
1111 541 327;
0111 212 62;
0011 318 59;
0001 701 585;
0000 934 2822;
.000 816 877;
..00 829 642;
...0 802 529;
Unmarked Seen Group=1;
816 829 802 1397;
Unmarked Seen Group=2;
877 642 529 1621;
/* End Input File */


Adjusted .inp :
/* No Individual Marks 2 Groups */
/* 4 occasions, 4 primary, 1 secondary each */
/*Begin Input File CHUB SUCKER */
10101010 541 327;
00101010 212 62;
00001010 318 59;
00000010 701 585;
00000000 934 2822;
..000000 816 877;
....0000 829 642;
......00 802 529;
Unmarked Seen Group=1;
816 829 802 1397;
Unmarked Seen Group=2;
877 642 529 1621;
/* End Input File */

Screwy Results (12:N Session 4) :
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
-------------------------- -------------- -------------- -------------- --------------
1:p Session 1 0.2630006 0.0035103 0.2561785 0.2699385
2:p Session 1 0.0000000 0.0000000 0.0000000 0.0000000 Fixed
3:p Session 1 0.1106104 0.0022255 0.1063232 0.1150483
4:sigma Session 1 0.0000000 0.0000000 0.0000000 0.0000000 Fixed
5:N Session 1 5807.2531 101.97778 5613.7322 6013.6537
6:N Session 1 11779.708 298.50979 11215.897 12386.703
7:N Session 2 6672.6826 102.92243 6477.3274 6880.9513
8:N Session 2 10532.134 245.40068 10070.750 11033.392
9:N Session 3 7399.0214 100.95541 7207.4857 7603.4002
10:N Session 3 10152.532 218.24872 9743.5557 10599.786
11:N Session 4 10463.373 141.06748 10194.002 10747.138
12:N Session 4 5903.0000 0.4836052E-003 5903.0000 5903.0014 [/size]

[bmcclintock]

Hi Matt,

Looks like a convergence issue. Run the model and check the box "Use Alt. Opt. Method". That should do the trick.

Cheers,
Brett

[AlaskaMatt]

Thanks Brett!

It appears to have worked! I at least have a more beleivable N-estimate for that group...
I am a bit hesitant to go around presenting results simply because they are "beleivable". In an effort to understand my model more completely so I can confidently defend my work, would you be willing to point me in the direction of any liturature that will explain to me what checking the "Use Alt. Opt. Method" actually does, and/or reasons I was having convergence issues?
As with many of the posters to this forum, I am brand spanking new to estimating population parameters (abundance specifically) and MARK, and my personal goal is to not just "trust the machine", but to actually know what the heck I'm doing. Eventually. Anything you have to say will be helpful.

Thanks again!

-Matt

Here's that last N- estimate after clicking the box. It's the only estimate that changed. :

Real Function Parameters of {p(.) sigma(.) N(t)}
95% Confidence Interval
Parameter Estimate Standard Error Lower Upper
-------------------------- -------------- -------------- -------------- --------------
13:N Session 4 20553.654 452.38441 19693.477 21467.484

[cooch]

Thanks Brett!
...would you be willing to point me in the direction of any liturature that will explain to me what checking the "Use Alt. Opt. Method" actually does, and/or reasons I was having convergence issues?

Start with the sidebar on p. 39 of chapter 10. 'Alternate optimization' is 'simulated annealing', which is introduced in said sidebar -- albeit somewhat briefly -- in the context of optimizing 'problem' data types (commonly, multi-state problems with the possibility of multiple modes in the likelihood. Simulated annealing simply does a better job at working through those problems....

As with many of the posters to this forum, I am brand spanking new to estimating population parameters (abundance specifically) and MARK, and my personal goal is to not just "trust the machine", but to actually know what the heck I'm doing. Eventually. Anything you have to say will be helpful.

Thanks again!

Good for you -- exactly the right attitude. Some of the answers to various questions -- as relate to MARK -- are in fact answered in 'the book'.The problem is, if you call it a problem, is that no one chapter can be entirely self-contained. The assumption is that before you get to Brett's chapter, you'll have read and fully assimilated most of the preceding 17 chapters (or, at the very least, chapter 1 -> 10).