Standard errors - robust design

questions concerning analysis/theory using program MARK

Standard errors - robust design

Postby Lauren » Wed Nov 10, 2021 4:16 pm

Hi all.

I hope to run a basic robust design model for my capture-recapture data - 3 primary occasions (7 months apart), of 10 secondary occasions (daily). I am interested in determining survival, emigration and immigration probability between the primary occasions. I have run Huggins robust design models for three movement models: no movement, random movement, Markovian movement.

However, I am concerned by odd standard error estimates. I am aware that, given confounding factors, that odd estimates are to be somewhat expected. The standard errors range from 0 to 100+ - some examples are below.

Does anyone know why such large standard errors are estimated and whether this precludes the use of robust design for my data? My initial thought was a limitation due to only 3 primary occasions, but I have read empirical and theoretical papers that show 3 primary occasions suffice (although is the minimum to be used). I have read the manual and other relevant papers back to back and many times over in search of answers.

Any help would be MUCH appreciated. I have happy to provide the encounter histories if it would help.

Many thanks,
Lauren

Random movement model
Parameter Estimate Standard Error Lower Upper
-------------------------- -------------- -------------- -------------- --------------
1:S 0.9999999 0.3791123E-03 0.4543176E-301 1.0000000
2:S 0.3544058 152.27742 0.3053694E-308 1.0000000
3:Gamma'' 0.8593845 0.0540323 0.7178485 0.9362289
4:Gamma'' 0.5952970 173.88869 0.8182420E-308 1.0000000

No movement model
Parameter Estimate Standard Error Lower Upper
-------------------------- -------------- -------------- -------------- --------------
1:S 0.2865718 0.0727603 0.1666401 0.4465657
2:S 0.2328974 0.0602238 0.1355532 0.3702092
3:Gamma'' 0.0000000 0.0000000 0.0000000 0.0000000 Fixed
4:Gamma' 1.0000000 0.0000000 1.0000000 1.0000000 Fixed

Markovian movement model
Parameter Estimate Standard Error Lower Upper
-------------------------- -------------- -------------- -------------- --------------
1:S 0.6567963 0.0000000 0.6567963 0.6567963
2:S 0.6350457 0.0000000 0.6350457 0.6350457
3:Gamma'' 0.7859071 0.0000000 0.7859071 0.7859071
4:Gamma' 0.6012360 0.0000000 0.6012360 0.6012360
Lauren
 
Posts: 2
Joined: Wed Nov 10, 2021 9:30 am

Re: Standard errors - robust design

Postby cooch » Wed Nov 10, 2021 4:41 pm

Simple enough -- with only 3 primary periods, there are only 2 'open' intervals. Meaning, there are precious few 'fancy' models you can build. For example, the Markovian model is not a model you can (easily) work with with only 2 intervals, since parameters aren't estimable without constraints (which are described at length in Bill's RD chapter).

So, without looking more closely, 'weird' (from the Latin) SE's for fancy models with only 3 primary samples is probably just an artifact of trying to do too much with too little. The basic rule of thumb (in my experience) for RD models is that you can get 'fancy' with 5 or more primary samples.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Standard errors - robust design

Postby Lauren » Wed Nov 10, 2021 5:40 pm

Thank you so much for your swift response - completely understood and I see the necessity to forgo fancy models (unfortunately!).

The simplest model (time-dependent, p=c) results are below. Again, large SEs but smaller than the fancy models. Perhaps it's time to look at other options.

Thank you again, your comments are much appreciated.

Code: Select all
     Parameter                Estimate       Standard Error     Lower           Upper
 --------------------------  --------------  --------------  --------------  --------------
     1:S                      0.9149823       18.746761      0.7847259E-204   1.0000000                         
     2:S                      0.8583194       32.988460      0.7450004E-230   1.0000000                         
     3:Gamma''                0.7137593       46.917532      0.8342824E-195   1.0000000                         
     4:Gamma''                0.6599728       78.411197      0.7279835E-297   1.0000000                         
     5:Gamma'                 0.0676562       54.702105      0.4036600E-309   1.0000000
Lauren
 
Posts: 2
Joined: Wed Nov 10, 2021 9:30 am

Re: Standard errors - robust design

Postby egc » Thu Nov 11, 2021 8:54 am

Build the simplest model possible -- if you still get inflated SE, then either (i) you've made a mistake setting up the model, or (ii) you simply don't have sufficient data. The former is fixable. The latter is not.
egc
Site Admin
 
Posts: 201
Joined: Thu May 15, 2003 3:25 pm

Re: Standard errors - robust design

Postby Bill Kendall » Thu Nov 11, 2021 2:10 pm

I have only one thing to add to what Evan has presented. With only three primary periods (2 intervals), the Markovian model is not an option at all. You need at least 4 primary periods.

For random emigration, with three primary periods in theory you should be able to estimate S1 and gamma1, but then S2 and gamma2 are confounded with one another.
Bill Kendall
 
Posts: 96
Joined: Wed Jun 04, 2003 8:58 am


Return to analysis help

Who is online

Users browsing this forum: No registered users and 9 guests