Program RELEASE GOF tests

[Otshawytscha]

I'm interested in evaluating whether or not a survival difference exists between three release groups (~12,000 fish in each) and am in the process of evaluating the fit of my most global model (i.e., Phi(group*T), p(group*T)) prior to proceeding with formal model selection and inference. I have a very basic question regarding goodness-of-fit testing in this process....

What is the best way to proceed when component Test 2 and Test 3 Program RELEASE GOF tests have insufficient data (as indicated in output) and are nonsignificant (P>0.25 for all; also, overall tests and sums of Test 2 + Test 3 are nonsignificant)?

Thanks!

[cooch]

I'm interested in evaluating whether or not a survival difference exists between three release groups (~12,000 fish in each) and am in the process of evaluating the fit of my most global model (i.e., Phi(group*T), p(group*T)) prior to proceeding with formal model selection and inference. I have a very basic question regarding goodness-of-fit testing in this process....

What is the best way to proceed when component Test 2 and Test 3 Program RELEASE GOF tests have insufficient data (as indicated in output) and are nonsignificant (P>0.25 for all; also, overall tests and sums of Test 2 + Test 3 are nonsignificant)?

Thanks!

Given your question, I *might* guess that you haven't read the relevant chapter (5) in the MARK book. Start there...

[Otshawytscha]

Thanks for the reply.

I have looked through both Chapter 5 of your book (a great resource, BTW) and Burnham et al.'s 'Blue Book'. Based on this, I haven't gained a clear sense of how to proceed in my situation. I've considered collapsing the final 2 encounter occasions into a single one (original: 1 release, 3 recapture events) to bolster detections, but I could not run RELEASE under these circumstances.....The other option I've considered was to collapse groups in the input file, but this is counter to the experimental objectives.

[cooch]

Thanks for the reply.

I have looked through both Chapter 5 of your book (a great resource, BTW) and Burnham et al.'s 'Blue Book'. Based on this, I haven't gained a clear sense of how to proceed in my situation. I've considered collapsing the final 2 encounter occasions into a single one (original: 1 release, 3 recapture events) to bolster detections, but I could not run RELEASE under these circumstances.....The other option I've considered was to collapse groups in the input file, but this is counter to the experimental objectives.

Then you've missed one of the basic points - RELEASE tests a fully time-dependent model. If you are getting contingency tables that are to sparse to be interpreted in any robust way, your only defensible option is to use a less general model - you can retain your groups, but just consider a model without time-dependence.

If you do that, you'll need to use something other than RELEASE, as discussed in detail in Chapter 5.