On 17 Jul 2002 07:42:10 -0700, [EMAIL PROTECTED] (Alfred Barron)
wrote:

> Hi all,
> 
> I have a client doing a bunch of samll sample 
> t-tests, about n = 3, 4 sample size/group. 
> For the moment, I'll skip the power.
> 
> Does anyone know of any studies or simulations 
> showing how poorly the nominal alpha level
> coverage can be ?

"how poorly"  under what abnormal conditions?

t remains rather well-behaved with equal Ns
(especially in contrast with bad behavior for unequal Ns,
with unequal variances).

Comparisons between long-tailed distributions  produce
empirically computed "t-test" distributions where the tails 
are shorter than for t;  and conversely, short-tailed 
distributions produce longer tails.  

Does that help?


-- 
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