bootstrap GOF for Pradel model?

[phettinga]

Hi,

Is it possible to use the bootstrap GOF test in MARK with the Pradel model to find c-hat? I am able to do it in MARK but a course website for a graduate level Demographics Methods course says this should not be possible. Thanks

[dhewitt]

You asked whether it is possible, and then told us you did it. So it is possible. I suspect what you're really asking is whether the method used in MARK to do it is valid, since some undocumented (to us anyway) reference has indicated it cannot be done "correctly". A few things are missing here:

1. What Pradel model are you using?
There is more than one model structure available in MARK.

2. What specifically does your reference say is wrong with the idea?

3. How does the bootstrap estimate compare to median c-hat, the preferred approach?
We've found them to be similar for some applications of Pradel models, but I always go with the median c-hat approach.

- Dave

[phettinga]

Hi,

I am trying to estimate lambda within the likelihood i.e. not as a realized parameter. I checked the gentle introduction and the Williams, Nichols and Conroy text and nothing is said about not being able to estimate c-hat using the bootstrap GOF test. But because one source said you couldn't (with no additional explanation given) I decided to make sure and post my question.

When I try to use median c-hat in MARK I get an error message. I am not even sure how you are able to do it. I read the gentle introduction manual and can't find any mention of median c-hat being the preferred approach either. Rather it is called promising and notes that all methods should be tested.

[dhewitt]

Fair enough. On the bootstrap, what is this "source" that you refer to?

As for the preference for median c-hat, if you don't trust me then just search the archives of this list and you'll find plenty of recommendations to that end. To the best of my knowledge, there is no definitive answer to the best approach, which is why I try them all.

As for running median c-hat, I have not used it on the models with lambda in the likelihood. But, for the seniority-only model (reverse CJS survival model), it works fine. What version of MARK do you have? What does the error message say?

[phettinga]

My source is a graduate level course called Demographic Methods. The error message I get when I try to use the Median c-hat bootstrap method is "The median estimate of c-hat only works for live recaptures and dead recoveries at this time" I just tried using the seniority only model and got the same message. Even with the dipper data I get this message. What version of MARK are you using?

[cooch]

Time to finish this thread off...

My source is a graduate level course called Demographic Methods.

Which doesn't tell anyone much unless we know who is teaching the course (or, the author of the materials).

The error message I get when I try to use the Median c-hat bootstrap method is "The median estimate of c-hat only works for live recaptures and dead recoveries at this time" I just tried using the seniority only model and got the same message. Even with the dipper data I get this message. What version of MARK are you using?

Chapter 12, and I quote:

If you try to run a bootstrap, or median
MARK will quickly tell you that a neither of these GOF
tests are available for the Pradel recruitment only model (and
indeed, they aren't available for any of the Pradel
models). At first glance, it might seem that you could simply flip
the encounter histories back around the `normal' forward direction
(as if you were going to do a CJS analysis), and simply use the CJS
bootstrap GOF test. However, this is inappropriate in general, since
the likelihood is based on the full encounter history. However, it
`may' be reasonable if your only interest is in estimating the
parameters. For the moment, we can cautiously suggest that
if all you're interested in is , and are using the Pradel
recruitment only model, then since this model is identical to taking
your encounter histories, flipping them, and running them through
the CJS model, then the from the CJS model may be
appropriate. But - don't quote us (this is still a work in
progress). However, for the other Pradel models, this is undoubtedly
incorrect.

In other words, if your interest is in the seniority model only, then there is (perhaps) a reasonable GOF test. If you're interested in any of the other models, there is no currently accepted GOF test, and (as such) nothing is available through MARK. Although you can run the RELEASE GOF test, it has essentially no meaning for Pradel models (and probably shouldn't be available when the Pradel data type is selected).

[dhewitt]

Big fat OOPS. That's what happens when you check your RMark code too quickly I guess. Since I was using the seniority-only model, I ran the median c-hat for the CJS model. Which means I must have reverently obeyed the book, eh?

Sorry for the confusion.