GOF (RDsurviv) and c-hat for Mark

[fanny]

I have test the GOF of my data (Robust design) with RDsurviv.
For the most general model, the GOF is not good.
It is possible to use this model in Mark (and create sub-models) and just add in Mark the value of the c-hat calculate from this most general model from RDsurviv?
Or I can't use this data in Mark?
Thank you very much for the help!

[ganghis]

Fanny,

To my knowledge, there hasn't been much simulation testing done on the effectiveness of goodness-of-fit procedures in RDSurviv. In complicated models like the robust design, I suspect there are some issues with pooling small cell probabilities that have not been looked into. I would suggest doing your analysis in MARK, and using the parametric bootstrap GOF (with the median c-hat option), which Gary has found works okay (but is not infallible).

Cheers, Paul

[fanny]

Thank you for teh answer!

But if I try to do a bootstrap in Mark with data from Robust design, the program tell me that is not possible to do a bootsrap with this data!!!

[cooch]

Thank you for teh answer!

But if I try to do a bootstrap in Mark with data from Robust design, the program tell me that is not possible to do a bootsrap with this data!!!

Paul was being literally 'precise' in his suggestion - the bootstrap is a general approach, although MARK does treat the 'bootstrap' GOF as separate from the 'median' GOF, even though both make use of a bootstrapping approach.

What Paul suggested is that you use the median GOF approach - which is one of the options available. It is discussed at length in the GOF chapter, which you should read first.

[modonnell]

I'm curious to know if there are any new advances in testing goodness-of-fit for the robust design or median c-hat still the best available/accepted method?

Matt

[Todd]

I do not see any option under Tests for median GOF testing in MARK using robust design.

Are we missing something?

[flounder]

I have also used a robust design model in MARK to estimate seasonal abundance of anadromous fish within a river system. It doesn’t seem that MARK provides a means to assess goodness of fit for robust design models.  I tried to use RDSURVIV but the results are suspect because the GOF for the best models all return as 1.0.  So, I am trying to identify the most generally accepted method for performing and reporting a goodness of fit test for this model. There doesn’t seem to be consensus on the best way to run a GOF on this type of model in most of the places I have researched.  Any advice, i.e. is there another program ?

[Eurycea]

I have also used a robust design model in MARK to estimate seasonal abundance of anadromous fish within a river system. It doesn’t seem that MARK provides a means to assess goodness of fit for robust design models.  I tried to use RDSURVIV but the results are suspect because the GOF for the best models all return as 1.0.  So, I am trying to identify the most generally accepted method for performing and reporting a goodness of fit test for this model. There doesn’t seem to be consensus on the best way to run a GOF on this type of model in most of the places I have researched.  Any advice, i.e. is there another program ?

Is there any consensus on this, GOF testing if you are running a closed Robust Design? I have the same problem as above.

[cooch]

The trick is to consider the classical RD as a mixture of closed abundance and open population modeling. Since the GOF doesn't apply in the usual sense for the closed part (see relevant sections of chapter 14), then the problem can be reduced down to an open population problem. So, one approach that seems to be fairly robust (in many cases) is to simply collapse your secondary samples to a simple binary variable (seen, not seen), and do a GOF on the resulting encounter history - i.e., collapse the secondary closed samples to a single sample (seen once or not), then treat it like a CJS model, and do the GOF on that.

Bill Kendall has thought more about this than anyone else (no surprise there) - but until he weighs in to correct the approach I've described (which has seen some application) - thats what I suggest you try.

[Eurycea]

The trick is to consider the classical RD as a mixture of closed abundance and open population modeling. Since the GOF doesn't apply in the usual sense for the closed part (see relevant sections of chapter 14), then the problem can be reduced down to an open population problem. So, one approach that seems to be fairly robust (in many cases) is to simply collapse your secondary samples to a simple binary variable (seen, not seen), and do a GOF on the resulting encounter history - i.e., collapse the secondary closed samples to a single sample (seen once or not), then treat it like a CJS model, and do the GOF on that.

Bill Kendall has thought more about this than anyone else (no surprise there) - but until he weighs in to correct the approach I've described (which has seen some application) - thats what I suggest you try.

Thanks for the suggestion. I have tried this approach and while I get an estimate of median c-hat for the fully time-dependent CJS model (no groups or anything fancy),

Code: Select all

     Estimated c-hat = 1.3451563 with sampling SE = 0.0090726

     95% Conf. Interval c-hat = 1.2839771 to 1.4063354

     One-sided 95% Upper Bound on c-hat = 1.3943265

           Beta Variance-Covariance Matrix
           3171.8494690      -2378.8760775
          -2378.8760775       1784.1742311

the results of the known fate model are odd:

Code: Select all

                                                              95% Confidence Interval
 Parameter                  Estimate       Standard Error      Lower           Upper
 -------------------------  --------------  --------------  --------------  --------------
    1:                      1.0000000       0.1486852E-007  1.0000000       1.0000000   

Is this partial convergence failure or nothing to worry about? I would not think I'd have such an issue with a fairly simple model, compared to the robust design stuff.

Thanks,

Nate