GOF testing and sufficient data (or not)

questions concerning analysis/theory using program MARK

GOF testing and sufficient data (or not)

Postby jcgarvin » Mon Mar 24, 2008 8:25 pm

I'm just beginning to analyze an 8 year set of mark/recapture/re-sight data with a live recaptures model. I've run median c-hat as well as program RELEASE to test the GOF of my most parameterized model or global model which is S(age*sex*t)p(age*sex*t). The summary table from RELEASE indicates that I do not have sufficient data for most of the tests. What does this mean? I've tried running RELEASE on less parameterized models (e.g. S(t)p(t)), and still have insufficient data on most of the tests. I only have roughly 130 individual encounters, and some have duplicate encounter histories. My median c-hat estimate was around 1.68, so I realize that is somewhat high.

Thanks so much for your help,
Julie Garvin
jcgarvin
 

Postby abreton » Mon Mar 24, 2008 9:15 pm

"RELEASE indicates that I do not have sufficient data for most of the tests. What does this mean?" It means just what it says - you'll need more data for these tests for the models you're fitting. I assume you tried dropping both age and sex and assessing fit of the Phi (t) p (t) model? If this is a reasonable model given your data then you might consider this option. Alternatively try running the GOF tests for single-strata data in UCARE (more options, slightly different algorithms). You can download UCARE and its user manual from http://www.cefe.cnrs.fr/BIOM/en/softwares.htm (see also resources from the Analysis Forum software links).
abreton
 
Posts: 111
Joined: Tue Apr 25, 2006 8:18 pm
Location: Insight Database Design and Consulting

Postby annaren » Wed Apr 02, 2008 10:28 am

I have had similar problems with insufficient data so what I have done is made a simpler model based on the results form UCARE, i.e. which test had sparse data. If it was test 2 CT then I took out time for the recapture etc. I then ran the median c hat test on my basic model and adjusted it accordingly as long as it the estimated c hat was below 3. If anyone has any comments on this method it would be greatly appreciated as I realise that the median c hat test is "still being developed" and other GOF tests should be done along side. However as RELEASE and UCARE only test the full time dependent model I think this may be my only option.
Thanks
annaren
 
Posts: 25
Joined: Tue Mar 04, 2008 12:00 pm
Location: University of Aberdeen


Return to analysis help

Who is online

Users browsing this forum: No registered users and 2 guests

cron