Goodness of fit testing for transient models??

Forum for discussion of general questions related to study design and/or analysis of existing data - software neutral.

Goodness of fit testing for transient models??

Postby bootzies » Tue Nov 04, 2014 1:37 am

Hello list.

I have a query that may be quite elementary, but I am having trouble working it out.

I have a long time series (30 years) of mark recapture data on >5000 individuals. I used RELEASE on the CJS model and identifeid a transient effect (i.e. there was a different survival probability for the year after marking than in subsequent years, most likley due to non-resident inidivuals moving through the survey area). As a result, I introduced a time since mark effect into the model of survival, to allow this transience to be accounted for in the model structure.

Now I am stuck....... I want to check the goodness of fit for my new and improved CJS model with the transient effect (as I want to ensure that by including the transient effect I have dealt with the lack of fit). Is there a way to do this? When I run the CJS with transient effect model through RELEASE in MARK, I still get highly significant results for test 3. Do the RELEASE tests only test against the standard CJS assumptions? Or do they take account of the parameterisation of the model object that they are being run on?

If it is not appropriate/possible to use RELEASE to check GoF for my updated transient (i.e.non standard) CJS model - then how can I do it? Do I just need to use the median-c/bootstrap tests to estimate c-hat and then correct for any overdispersion without actually being able to identfy it's source?

Thanks in advance for your help!

Alice
bootzies
 
Posts: 21
Joined: Wed Jan 29, 2014 2:31 am

Re: Goodness of fit testing for transient models??

Postby Morten Frederiksen » Tue Nov 04, 2014 4:01 am

Hello Alice,

The goodness-of-fit test for your model with a transient effect on survival is obtained simply by ignoring the test component 3.SR in RELEASE (or U-CARE, which I would recommend using), and summing the chi-square values and degrees of freedom for the other components.

See the classical paper by Pradel et al (1997), Biometrics 53, 60-72.

Morten
Morten Frederiksen
 
Posts: 42
Joined: Thu May 29, 2003 4:53 am
Location: Roskilde, Denmark

Re: Goodness of fit testing for transient models??

Postby bootzies » Wed Nov 05, 2014 11:53 pm

Thanks Morten for your advice, it has been very useful.

I am also trying to assess whether including a transient effect in a multistate model improves goodness of fit compared the the multistate model without the transient effect. I am a little unclear how to calculate GoF from the component U-CARE tests for the multistate model. I have read the Pradel et al 2003 paper (Biometrics, vol 59), but am still a bit unsure.

I think that the U-CARE Test 3G.SR is the test for transience (is that right?), so I assume that excluding this component and calculating an overall c-hat using the chi-sq and df of all other components (Test3G.SM, TestM.ITEC and TestM.LTEC) should work in similar way to the method that you described above for the RELEASE tests on the CJS model. Is this correct?

Am I right in thinking that the test WBWA is a sub-test of Tests 3 (so I don't need to include the results of this in the calculations for GoF)?

Thanks again!

Alice
bootzies
 
Posts: 21
Joined: Wed Jan 29, 2014 2:31 am

Re: Goodness of fit testing for transient models??

Postby Morten Frederiksen » Thu Nov 06, 2014 8:22 am

Hi Alice,

You're right that test 3G.SR is the multistate equivalent of test 3.SR, and by excluding this component you get a GOF test for a model with transience. However, you need to include test WBWA, which is a specific test for memory effects (not part of test 3). Check Choquet et al (2009), Ecography 32, 1071-1074 - the official reference for U-CARE.

Morten
Morten Frederiksen
 
Posts: 42
Joined: Thu May 29, 2003 4:53 am
Location: Roskilde, Denmark

Re: Goodness of fit testing for transient models??

Postby bootzies » Fri Nov 07, 2014 12:22 am

Thanks again Morten - you have been extremely helpful!
Alice
bootzies
 
Posts: 21
Joined: Wed Jan 29, 2014 2:31 am


Return to analysis & design questions

Who is online

Users browsing this forum: Google [Bot] and 13 guests