Re: ANOVA result when number of groups changes

On 7 Apr 2006 21:16:42 -0700, "nels0360" <nels0360@xxxxxxxxx> wrote:

Well, I got it to pass the normality and equal variance tests. I just
did a log transformation since it was conerning bacteria growth. Now I
have a good data set to work from.

I am still curious about the analysis of only two of the groups instead
of all ten at the same time. I would assume it had to do with the
great variance when I included the groups with very high means and the
high variance that accompanied them. Seems as though the log
transformation took care of that.

Most of the multiple-test procedures do use the pooled
variance term, so that could let a simple t-test between
two groups test as nominally significant, when the contrast
was not marked that way by a multiple-test procedure.

The more general explanation is the distinction between
"nominal" and "family-wise" testing. "Nominal" is what you
get when you look up a t-test in a table; "family-wise" is
what you get when you compensate for doing multiple tests,
and want to preserve the 5% level for making even one mistake.

Consider the simple case of Bonferroni testing. If you want
to provide a 5% family-wise test for five independent tests,
you need to use the nominal 1% cutoff for each of the tests.
The ANOVA post-hoc <whichever> uses assumptions of equal
and pooled variances in order to be less conservative
than "Bonferroni", but the principle is preserved: The separate
tests are more conservative than a 5% "nominal" test

There are some good comments (posted long ago) in my stats-FAQ at
http://www.pitt.edu/~wpilib/statfaq/97postho.html
or you can search groups
< group:sci.stat.* bonferroni post-hoc > .

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.

Relevant Pages

  • Re: ANOVA result when number of groups changes
    ... I got it to pass the normality and equal variance tests. ... did a log transformation since it was conerning bacteria growth. ... high variance that accompanied them. ...
    (sci.stat.consult)
  • Re: ANOVA result when number of groups changes
    ... I got it to pass the normality and equal variance tests. ... did a log transformation since it was conerning bacteria growth. ... high variance that accompanied them. ...
    (sci.stat.math)