In article
<4154fd79-1b98-429f-8298-d6f06064ac92@[EMAIL PROTECTED]
>,
Robert Dodier <robert.dodier@[EMAIL PROTECTED]
> wrote:
>David Jones wrote:
>> Of course a true "statistician" will use any tool that will provide an
>> answer to his questions, whether it be "frequentist" or "Bayesian".
>The problem with frequentist methods is not so much that they are
>wrong, as that they are misdirected. Having excluded prior information
>and probability of hypotheses from the analysis, frequentists declare
>the questions of real interest out-of-bounds, and proceed to solve
>some other problem that nobody really cares about. Significance
>tests and hypothesis tests are really an institutionalized form of
>looking for the house keys under the streetlight.
This is quite correct. The real problem of inference is
that we do not know the state of nature, so we have to
balance the risks in the various states of nature. This
leads to Bayes as the ideal. However, one needs to use
"engineering" approximations, so using any convenient
loss-prior combination is not a good idea.
>No true statistician would use "any tool" if it meant solving the
>wrong problem.
This is why it is im****tant to understand, rather than
to have a box of tools.
>FWIW
It is worth much; the person who does not know the
difference between statistical significance and
practical significance cannot use statistics in
any intelligent manner.
>Robert Dodier
--
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
|
