Trap dependence

[mekeys]

Another question - this time regarding estimateing c hat. My prefered current model is phi(t) p(M2-t/t) with full time dependence for both survival and resight, resighting now with an added age class. However, the fit is still not very good. The U-Care GOF gives 82 df and Chi2 1091.1023 - which is an estimated c hat of 13.31!

[cooch]

Another question - this time regarding estimateing c hat. My prefered current model is phi(t) p(M2-t/t) with full time dependence for both survival and resight, resighting now with an added age class. However, the fit is still not very good. The U-Care GOF gives 82 df and Chi2 1091.1023 - which is an estimated c hat of 13.31!

First, use median c-hat to estimate lack of fit statistic. The U-CARE approach (like the RELEASE approach for CJS models) is sensitive to binning issues. Second, if even the media c-hat approach gives poor fit, it means there are other things to consider: lousy general model (perhaps you need structure in phi as well), or lousy data. The former is solvable. The latter isn't.

[mekeys]

Thanks again for your guidance. I will try the median c hat approach

[mekeys]

Thank you for your help so far. One more question... hopefully the final question. I have followed your guidance and created 2 age clases with full time dependence in the resighting PIM and this has resulted in much more sensible survival estimates. I now want to run various environmental constraints, but only on survival (phi), since this is the area of study interest. The Model has 92 parameters - too many for a full design matrix which means I need to create a design matrix - I am assuming the best option is to use the identity design matrix. However, I am unsure how to express the resighting PIM with 2 age classes within the identity matrix. The mannual explains how to constrain these 2 age classes, but not how to just enter them without constraints, while constraining a single phi age class on the other side of the matrix. Thanks again in advance.

[cooch]

Thank you for your help so far. One more question... hopefully the final question. I have followed your guidance and created 2 age clases with full time dependence in the resighting PIM and this has resulted in much more sensible survival estimates. I now want to run various environmental constraints, but only on survival (phi), since this is the area of study interest. The Model has 92 parameters - too many for a full design matrix

No. You can't automatically generate the full DM, not because you have too many parameters (there is effectively no practical limit), but, rather, because the automatic DM tool you probably used only works if the underlying parameter (PIM) structure is simple time-dependence. The moment you added TSM structure, that is no longer the case. You'll need to build the DM by hand - Chapter 7.

which means I need to create a design matrix - I am assuming the best option is to use the identity design matrix. However, I am unsure how to express the resighting PIM with 2 age classes within the identity matrix. The mannual explains how to constrain these 2 age classes, but not how to just enter them without constraints, while constraining a single phi age class on the other side of the matrix. Thanks again in advance.

You need to read, study, and work through Chapter 6 and Chapter 7. At the end of that exercise, you'll see that building a DM for phi(t)p(a2 - t/t) is trivial.

[jm8360]

I have a similar problem to mekeys - trap dependence in sea bird data. I think that my trap-happiness is permanent, and due to site fidelity by breeding birds.
I have therefore transformed my data in U-care using 'split for trap dep analysis'. Then I have loaded this new file into mark and split the recapture parameters into two age categories, but still allowing them to vary with time. My PIMs for p look something like this:

6 1 2 3 4 5
7 2 3 4 5
8 3 4 5
9 4 5
10 5
11

My global model still does not fit the data very well - please can someone let me know if my PIM for P is correct?
Is this the equivalent of p(t*m)?

Thank you very much for your help in advance,
Jess

[Miguel]

Is this the equivalent of p(t*m)?
[/quote]

I imagine that by now you have figure out that you are indeed correct that your PIM is equivalent to p(t*m).

Miguel

[Miguel]

If it is transient (typically), the easiest way to handle it is to consider it as a form of capture heterogeneity. In that case, the most straightforward approach - which is the one you're hinting at - is to simply add 'age' structure (TSM - time since marking - is the preferred vernacular) to the encounter parameter.

Even

I have also encountered transience in a multi-state analysis but I applied TSM to survival rather than the captures otherwise the first survival rate would be biased by transience. TSM survival models greatly improved model fit while TSM recaptures models did little to improve fit. Shouldn't Mekeys also use TSM survival models? Is it best to just try both?

Thanks,
Miguel