GOF for Pradel (seniority) model

[lhliow]

I have been trying to search older posts for an answer to this question:

I am running Pradel seniority models (to get at phi, gamma and lambda) but can't find a standard GOF test that will run on MARK. An older post from 2010 on the Pradel model had Evan Cooch and/Gary White saying "do standard CJS gof on either RELEASE or another program on MARK and use the c-hat" (not sure which, but the replies we a bit mingled).

Has the view been revised since? Or are there newer/more appropriate ways to go GOF for JS type models? I am a paleontologist so I don't encounter the statistical ecology literature on a daily basis. Or should I go ahead and pretend I am doing CJS then use the chat to adjust my Pradel models? the c-hat I got from pretending it is CJS is about 1.5 from RELEASE. And in any case, I am not terribly interested in model comparison in my study, I simply need to get time-varying estimates and group estimates.

Thank you.

Lee

[Eric Janney]

Pradel's seniority model is the reverse analogue of the basic CJS model. So, GOF assassment for that dataset fit to the global CJS model will be analogous to the GOF for that dataset fit to the seniority model. I think you are safe using 1.5 as your c-hat

[cooch]

I have been trying to search older posts for an answer to this question:

I am running Pradel seniority models (to get at phi, gamma and lambda) but can't find a standard GOF test that will run on MARK. An older post from 2010 on the Pradel model had Evan Cooch and/Gary White saying "do standard CJS gof on either RELEASE or another program on MARK and use the c-hat" (not sure which, but the replies we a bit mingled).

Canonical statement of GOF - see p. 8 of the following:

http://www.phidot.org/software/mark/doc ... chap13.pdf

Has the view been revised since? Or are there newer/more appropriate ways to go GOF for JS type models? I am a paleontologist so I don't encounter the statistical ecology literature on a daily basis. Or should I go ahead and pretend I am doing CJS then use the chat to adjust my Pradel models? the c-hat I got from pretending it is CJS is about 1.5 from RELEASE. And in any case, I am not terribly interested in model comparison in my study, I simply need to get time-varying estimates and group estimates.

Thank you.

Lee

Eric is correct - with the additional comment that the seniority-only model is the only model (among those available for the 'Pradel data type') which is a literal flip of the CJS model. And, for which there is general acceptance that a median c-hat GOF test is probably robust. For other forms of the Pradel model, not so much.

[lhliow]

Thanks Eric and Evan,

I read the Pradel chapter and re-read it again in your link, Evan. The problem is that I want from my data, phi, gamma AND lambda. Not just gamma (seniority). What should I do? Should I adjust the c-hat according median c-hat even though it is "wrong" or should I just leave it and say that there is no way to do a correct gof?

Thanks,

Lee

[Eric Janney]

I'm not sure why the other forms of the Pradel have the same fit as the gamma only form? Evan - Why would the assessment of whether the data fit the model change depending on whether you estimate phi and gamma in the liklihood and then derive lambda or f versus estimating lambda or f directly from the likelihood. I can't think of a situation in which the data would fit in the first scenario but not the second?

[cooch]

I'm not sure why the other forms of the Pradel have the same fit as the gamma only form? Evan - Why would the assessment of whether the data fit the model change depending on whether you estimate phi and gamma in the liklihood and then derive lambda or f versus estimating lambda or f directly from the likelihood. I can't think of a situation in which the data would fit in the first scenario but not the second?

Simple - different likelihoods. Different functional forms of the model being fit to the data. The parameterization for Pradel seniority and CJS are essentially equivalent in all important respects, so GOF that works for the latter can be applied to the former.

[gwhite]

All:
The only lack of fit that you can detect in the open models (Jolly-Seber, Pradel, Link-Barker) is from the recapture portion of the likelihood (i.e., the Cormack part of the CJS) -- something Ken Burnham has stressed. So, GOF tests from RELEASE are generally what we've recommended in the past. The likelihoods for all 3 of the Pradel parameterizations and for the Link-Barker are identical when all parameters are estimated without losses on capture (and lambda >= phi for the Pradel lambda parameterization).

This equivalence of the models is part of the reason that I have the CJS data type available in these open models as an alternative data type, and why you can run RELEASE from any of these data types.

Gary