Right input to `Known Marks` in Mark-Resight Model

[Kenup17]

Hello there,

I've recently run into some trouble while using the "PoissonMR" model in Rmark. I was able to run the analysis well until I added my first occasions wich included "-0" entries (i.e., the first 'absences' of individuals). My encounter history looks like this:

Code: Select all

030102+0010305050407040208010403060201+0 1 0;
..........+00201+001-0-0-0-0-0-0-0-0-0-0 1 0;
..........020203030402+0+00707010201+002 1 0;
....................071104070602060204+0 1 0;
..............................+0+0+002+0 1 0;
..............................+005+0+0+0 1 0;
..............................04+0060605 1 0;
..............................05030101+0 1 0;
0510060407050805+006050901+0+0-0-0-0-0-0 0 1;
0203050805060403+006050405050808040505+0 0 1;
..............................010302+002 0 1;

And my 'Known Marks' counts are:

Code: Select all

Known Marks Group = 1 ;
1 1 1 1 1 3 3 3 3 3 0 0 0 0 0 0 0 0 0 0;
Known Marks Group = 2 ;
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 0 0 0 0 0;

The model output is, to quote 'The Princess Bride', inconceivable. All beta parameters are estimated with 0 standard errors, and real parameters like alpha are absurdly inflated (over 3000).

However, since this trouble began after the presence of '-0's in my encounter history, I had a hunch that the trouble lied in the 'Known Marks' counts. Instead of setting it to '0', which I was supposed to do when I do not know the exact number of marks, I tried using the minimum number of marks present in each occasion. The counts therefore looked like this:

Code: Select all

Known Marks Group = 1;
1 1 1 1 1 3 3 3 3 3 3 3 3 3 3 7 7 7 7 7;
Known Marks Group = 2;
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2;

The models I fitted this way seemed pretty reliable. Am I missing something? After re-reading chapter 18 of Cooch & White, I still get the impression that the right input to 'Known Marks' is the first one, however my outputs keep telling me otherwise.

Thank you!

Best regards,

[bmcclintock]

Hi Kenup17,

It's difficult to say what the problem is without more information, but I can assure you that specifying the minimum number of marked individuals under 'Known Marks' is not a valid solution when the number of marks is unknown.

I'm guessing your problem could be arising from a number of issues related to model specification, optimization, initial values, etc., but what jumps out to me is the paucity of marked individuals (particularly early in the study).

Cheers,
Brett

[Kenup17]

Hello Brett,

I am aware of the low number of individuals, however, even using only resighting occasions with just 3 individuals resulted in reasonable estimates (with some large C.Is of course, but that was expected).
I already imagined that the Known Marks input was wrong. However, tweaking my dataset in R showed that the model only has trouble running when there are "-0" entries (i.e., the number of marks in unknown).
I exported the input data to the original MARK interface to be sure it was not a problem with the package 'RMark'

Using the following ch and known marks count (with very few individuals):

Code: Select all

030102+0010305050407 1;
..........+00201+001 1;
..........0202030304 1;
0510060407050805+006 1;
0203050805060403+006 1;

Known Marks Group = 1 ;
3 3 3 3 3 5 5 5 5 5;

The model ran just fine, with the following first 3 alpha estimates, for example?

Code: Select all

Parameter                  Estimate       Standard Error      Lower           Upper
 --------------------  --------------  --------------  --------------  --------------
     1:alpha g1 a0 t1    3.3333318       1.0540939       1.8200671       6.1047756                           
     2:alpha g1 a1 t2    3.7134669       1.7833696       1.5205615       9.0689103                           
     3:alpha g1 a2 t3    4.3333339       1.2018504       2.5415415       7.3883440


However, if I replace some '+0's for '-0's (which is actually the right input for my data):

Code: Select all

030102-0010305050407 1;
..........+00201-001 1;
..........0202030304 1;
0510060407050805-006 1;
0203050805060403-006 1;

Known Marks Group = 1 ;
3 3 3 0 3 5 5 5 0 5;


And run this model in MARK interface, the output is:

Code: Select all

  Parameter                  Estimate       Standard Error      Lower           Upper
 --------------------  --------------  --------------  --------------  --------------
     1:alpha g1 a0 t1        290246.04       0.0000000       290246.04       290246.04                           
     2:alpha g1 a1 t2         1729399.9       0.0000000       1729399.9       1729399.9                           
     3:alpha g1 a2 t3        269371.64       0.0000000       269371.64       269371.64                           


And the following warning is shown: Error number 0 from VA09AD optimization routine.

I'm guessing that MARK is having trouble estimating the number of marked individuals? But so far, it is the only aspect of the model that is hindered by the low number of individuals.

I tried using the estimates from the first model as initial values for the second, and it seemed to work (i.e., no more garbage estimates).

However, when using the full data-set:

Code: Select all

030102-0010305050407040208010402060101-0 1;
..............................+005-0-0-0 1;
..............................05030101-0 1;
..........+00201-001-0-0-0-0-0-0-0-0-0-0 1;
..............................04-0060605 1;
..........020203030402-0-00705010201-002 1;
..............................+0-0-002-0 1;
....................040502030301030203-0 1;
0510060407050805-006050901-0-0-0-0-0-0-0 1;
..............................010301-002 1;
0203050805060403-006050405050808040505-0 1;

Known Marks Group = 1 ;
3 3 3 0 3 5 5 5 0 5 0 0 0 0 0 0 0 0 0 0;


The aforementioned approach did not help in any way. I'm guessing due to the large number of occasions in which the number of marked animals must be estimated.
In any case, it seems that you're right and this is a convergence problem.
Do you (or some other user) have any further insight on this?

Thanks in advance for the patience!

Cheers,