alind wrote:A question on variance estimates for real parameter estimates....
I have been running some of the pre-defined models (CJS type) to determine survival rates and cap probs. for a female only data set of frogs in N. Calif. I've got 5 years of data lumped to the annual level, so 5 cap-recap occasions.
When I request variance-covariance matrix output for the real parameter estimates, I find the values on the diagonal of that matrix corresponding to the parameter number, correct? If so, my variances all come out smaller than the SE's available in the real parameter estimates window. This is non-intuitive from my past statistical experience - can someone explain.
Thanks much!
If you're running pre-defined models, I'll *guess* you haven't gone through Chapter 1 -> 7 in the MARK book:
http://www.phidot.org/software/mark/docs/book/
You should first, before going much further. If you have, re-read the sidebar starting on p. 27 of Chapter 7, and then, have a look at Appendix 2 (the Delta method) - particularly the material starting on the bottom of p. 15.
In addition, you need to understand what MARK outputs in terms of the VC matrix. In the Notepad output from MARK, the variance-covariance values are below the diagonal,whereas the standardized correlation values are above the diagonal.
However, it is very important to note that the V-C matrix that MARK outputs to the Notepad is rounded to 5 significant digits. For the actual calculations, we need to use the full precision values. To get those, you need to either (i) output the V-C matrix into a dBase file (which you could then open with dBase, or Excel), or (ii) copy the V-C matrix into the Windows clipboard, and then paste it into some other application. Failure to use the full precision V-C matrix will often (almost always, in fact) lead to ‘rounding errors’ in the calculations described in the sections of the book I mention in the earlier part of this note.
