On 5 Mar 2003 07:00:12 -0800, [EMAIL PROTECTED] (dennis roberts) wrote:

> 
> 
> build a 95% confidence interval ... see what you see ... and be done with it 
> 

If I were reading the report, I would not be satisfied to
know that the tiny sample of  *10*    did not  
"differ at the 5% level."   

I would hope for no-trends, and I would expect that
the experimenter would tell me whatever it was that
came out worst.  Anything at the 10% level? which
is the 5%, one-tailed, if someone pushes the question.

(That does not answer the question of whether the
experimenter should continue with the first 10 if
they looked "somewhat bad"  or should draw another
10  in order to perform the expensive followup.)

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to