In article
<b286e083-9e7c-471a-9ed2-568e99e40212@[EMAIL PROTECTED]
>,
Luna Moon <lunamoonmoon@[EMAIL PROTECTED]
> wrote:
>On Apr 2, 11:09 am, Ray Koopman <koop...@[EMAIL PROTECTED]
> wrote:
>> On Apr 2, 11:57 am, Luna Moon <lunamoonm...@[EMAIL PROTECTED]
> wrote:
>> > On Apr 1, 10:57 pm, Ray Koopman <koop...@[EMAIL PROTECTED]
> wrote:
>> >> On Apr 1, 10:13 pm, Luna Moon <lunamoonm...@[EMAIL PROTECTED]
> wrote:
>> >>> Hi all,
>> >>> Suppose I have a model and I've used MLE to estimate the parameters
>> >>> for the model. What are the good methods that I can use the test
the
>> >>> goodness of the MLE estimation results?
>> >>> Thanks!
>> >> If your model is sufficiently close to correct then the inverse
>> >> of the matrix of second derivatives of the negative log likelihood,
>> >> evaluated at the likelihood-maximizing estimates, is usually a
>> >> consistent estimate of the covariance matrix of the estimates.
>> >> More generally, you can always get an empirical covariance matrix
>> >> by bootstrapping or jackknifing.
>> > How does that relate to evaluation of the performance?
>> The same way that the standard error of any estimator relates to its
>> performance. What do you mean by "evaluation of the performance"?
>How about model mis-specification?
>Thanks!
There was a letter to the editor of a science journal
some time ago, saying that he was not interested in
the errro of a measurement from what it was measuring,
but from the true value.
One can test for model misspecification by testing against
other models, including generalizations of this one, or
by seeing how well the data points follow the predicted
distribution.
How does one test anything? But a Bayesian approach also
gives an idea of what significance level is appropriate for
the problem and sample size, and classical statistics does
not do anything of the sort.
--
This address is for information only. I do not claim that these views
are those of the Statistics Department or of Purdue University.
Herman Rubin, Department of Statistics, Purdue University
hrubin@[EMAIL PROTECTED]
Phone: (765)494-6054 FAX: (765)494-0558
|
