MANOVA or Multiple ANOVA

Main Page | Report this Page

Science Forum Index » Statistics - Education Forum » MANOVA or Multiple ANOVA

Page 1 of 1

Author

Message

Guest

Posted: Sun Apr 20, 2008 1:25 pm

Can anyone tell me the advantage of running a MANOVA rather than
multiple ANOVA for each dependent variable? I ran 3 ANOVAs on 3
different dependent variables and was criticized for this approach,
but I do not know why. Is this similar to running multiple t-tests
instead of an ANOVA? Any help, and/or references would be appreciated!

Art Kendall

Posted: Mon Apr 21, 2008 7:30 am

Guest

Quote:

Guest

Posted: Mon Apr 21, 2008 7:37 am

On Apr 21, 5:30 am, Art Kendall <Arthur.Kend...@verizon.net> wrote:

Quote:

yes it is analogous. The approach is often called using an omnibus
test. The reasoning goes like this. As with the Anova situation you
can have a significant overall test when none of the specific test are
significant. This is due to the error term in the omnibus tests
accounting for more of the total.

Some people contrast the approach with an omnibus test followed post hoc
by all pairs of differences with an approach with a priori contrasts
specified.

Art Kendall
Social Research Consultants

beginner1....@hotmail.com wrote:
Can anyone tell me the advantage of running a MANOVA rather than
multiple ANOVA for each dependent variable? I ran 3 ANOVAs on 3
different dependent variables and was criticized for this approach,
but I do not know why. Is this similar to running multiple t-tests
instead of an ANOVA? Any help, and/or references would be appreciated!- Hide quoted text -

- Show quoted text -

Thank you for responding! So, if since I ran multiple ANOVA tests,
would adjusting the alpha level (like a Bonferroni) be appropriate, or
should one always use the omnibus test?

Guest

Posted: Mon Apr 21, 2008 1:26 pm

If the null hypothesis is not true, then the error is not preserved for
the individual ANOVAs. I'm not a fan of univariate ANOVAs as follow-ups
to MANOVA--the two approaches are almost entirely different and have
different aims. My preference for follow-up of MANOVA (if I have to use
MANOVA) is to describe the discriminant weights.

If I'm just interested in controlling the error rate without regard to
some underlying discriminant structure, I'll just use ANOVAs with some
correction for multiplicity.

Mike Babyak

Richard Ulrich

Posted: Mon Apr 21, 2008 7:31 pm

Guest

On Sun, 20 Apr 2008 16:25:11 -0700 (PDT), beginner1.mat@hotmail.com
wrote:

Quote:

The main *advantage* of running MANOVA is that
only with MANOVA can you detect effects that depend
on (an unexpected) pattern among the outcome variables.

The *criticism* probably arises because of the multiple-test
circumstance, and controlling by Bonferroni levels leaves
you without the complexity of MANOVA to deal with.

Constructing a composite out of the three outcomes, or
naming one of them as primary, increases statistical power.
This makes use of what you know and want to know, and
is often the preferable approach.

--
Rich Ulrich

http://www.pitt.edu/~wpilib/index.html

Page 1 of 1 All times are GMT - 5 Hours The time now is Tue Oct 07, 2008 11:37 pm Science Forum Index