On Mon, 8 Dec 2003 14:28:25 +0800, "Erica So"
<[EMAIL PROTECTED]> wrote:

> After conducting the two way anova with REPEATED measure, I have
> difficulties on understanding the Post hoc test. Could anyone suggest some
> reference that I can refer to? Both website and reference book are ok.

There are designed contrasts for oneway ANOVA  which 
go by names like SNK and Tukey's.  They have certain things
in common:  The contrasts are a  subset of  possible tests, and 
they may have rules for applying  them,  and they are based 
in theory  on the shared within-group variance.   Some 
people always insist that there should be  an overall test 
that is 'significant' and so they call them  "post hoc" tests.

Two-way tests and in particular, repeated measures, don't
have the shared variance that can safely, generally,  be used.
So, as it happens, there aren't any "post-hoc"  tests that I know
of, going by any of those names,  that are widely recognized.
I once read a really awkward description in one textbook, 
many, many pages  for trying to apply a 'post-hoc'  test, and
that left me convinced that *that*  test was not viable.  I think
I've heard that some stat-pack  has something, but I don't 
know what and I surely would not trust it without reading
its documentation and sources carefully.


In biostatistical applications, followup tests for repeated 
measures are usually conducted as paired t-tests, using 
the variance term that relies on only the data being tested.
They are not usually labeled as 'post-hoc'  tests, though
the logic is not too different if you merely apply the need for
one overall test  to justify the much-larger set of followups.

Unequal correlations as well as unequal variances stand in
the way of using the pooled tests of the oneway ANOVA,
that go by those particular names as 'post-hocs'.


You may want to read about "Bonferroni"  correction;  
it can be used in  various circumstances.

So, what is it that brings to the surface  the question of 
post-hocs for a repeated measure?  suggested by a 
stat-pack, a reviewer, or  convenient analogy?

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
"Taxes are the price we pay for civilization." 
.
.
=================================================================
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