In article <[EMAIL PROTECTED]>,
Alan McLean <[EMAIL PROTECTED]> wrote:
>[EMAIL PROTECTED] wrote:
>> More importantly, I would say: DON'T DO TESTS. Instead, try to find
>> models that you would be prepared to use to predict the response
>> in as-yet untried circumstances.
>> --
>Hypothesis testing is simply one useful method of identifying 'models
>that you would be prepared to use to predict the response
> in as-yet untried circumstances.'
>Any method has to use past experience ('sample data') to identify models
>and choose between them.
>Hypothesis testing is restricted in its use, but within its limitations
>it is very useful.
I suggest that practitioners of statistics abandon their
current RELIGION and look at the problems. Testing is
needed, but the real problem is when one should accept
a hypothesis known to be false.
In some cases, this can be approximated by a point null,
but one should still use a decision approach to the problem.
The "significance level" should depend rather heavily on
both the problem and the sample size.
I have only seen one way of using fixed level testing
which I consider to be somewhat sensible; a mathematical
psychologist told me this one. He produces a model, and
collects data until it is rejected at the .05 level.
Then he looks at the fit, and decides whether to accept
his model as an approximation. The point of sampling
until rejection is to avoid accepting on a small number
of observations which accidentally fit.
--
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, Dept. of Statistics, Purdue Univ., West Lafayette IN47907-1399
[EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
