[ADMB Users] all point estimates are reasonable although the objective function fails to be calculated
shyunuw at gmail.com
Thu May 19 09:00:11 PDT 2011
Hello, Dr. Sibert.
Thank you very much for your response. Yes, I double-checked whether
the cor, std files (xxx.cor, xxx.std) were generated from the TPL of
concern. Yes, they were. I monitored running of the resultant
executable file. Initially the objective function (f) value was
*calculated*, but ultmately f was nan (i.e., not calculated).
Interestingly those cor, and std files were successfully generated,
and all point estimates looked very reasonable.
On May 18, 11:55 pm, John Sibert <sib... at hawaii.edu> wrote:
> You should be deeply suspicious.
> Even though the point estimates look "reasonable" (agree with your
> preconceptions), you will never be able to do any rigorous statistical
> testing such as likelihood ratio tests against alternative hypotheses.
> Or even a simple AIC comparison.
> Are you sure the std and cor files came from the same fit as the nan?
> Delete them and run the model again.
> Where did the nan come from? Check for divide by zero.
> On 05/18/2011 05:04 PM, Saang-Yoon wrote:
> > Dear all.
> > I have interesting results where although the ultimate objective
> > function (i.e., negative loglikelihood) is nan (i.e., failure), all
> > point estimates of parameters look very reasonable, and interestingly
> > correlation and standard deviation outputs (i.e., xxx.cor, xxx.std)
> > are also generated. Under the circumstance (i.e., f = nan), are
> > reliable those point estimates and their standard deviation
> > estimates? I am very interested in learning your ideas.
> > Saang-Yoon
> > _______________________________________________
> > Users mailing list
> > Us... at admb-project.org
> John Sibert
> Emeritus Researcher, SOEST
> University of Hawaii at Manoa
> Visit the ADMB projecthttp://admb-project.org/
> Users mailing list
> Us... at admb-project.orghttp://lists.admb-project.org/mailman/listinfo/users
More information about the Users