Orthogonal sets of a priori contrasts do not require adjustment of p-values
because they ARE simply components of the ANOVA. That is why they are so
powerful for hypothesis testing. They are also greatly under utilized.
William J. Resetarits
Program Director
Population and Community Ecology Cluster
Division of Environmental Biology
National Science Foundation
4201 Wilson Blvd., Suite 635
wrese...@nsf.gov
Voice (703) 292-7184
Fax (703) 292-9064
-----Original Message-----
From: Ecological Society of America: grants, jobs, news on behalf of Asaf Sadeh
Sent: Wed 7/7/2010 11:20 AM
To: ECOLOG-L@LISTSERV.UMD.EDU
Subject: [ECOLOG-L] planned contrasts replacing 2-way ANOVA
I am planning a 2x2 factorial experiment whe
Dear list members,
I am planning a 2x2 factorial experiment where I record several response
variables to test several hypotheses.
Some of these hypotheses can be tested using the usual 2-way ANOVA. That is,
the
significance or non significance of the two main effects and their
interaction is expected to provide sufficient information to draw conclusions
regarding these hypotheses.
However, another hypothesis focuses on one of the four treatment
combinations, and would be best approached by comparing this focal treatment
with the other three. However, if I use the 2-way ANOVA, some possible results
may be inconclusive, and require further "post hoc" analyses with
a reduced threshold for significance.
Since my a-priori hypothesis calls for 3 specific contrasts, it makes
statistical sense to me that I can skip the 2-way ANOVA and only perform these
3
contrasts without reducing the significance threshold. The logic behind this is
that these 3 planned contrasts replace the 3 component tests of the 2-way ANOVA
procedure (2 for the main effects and one for their interaction) that are
each always done without any change to the significance threshold.
Though I have found support for this approach on online statistics lectures
notes, I have never seen it taken in any published study, and I wonder if it is
indeed kosher (and if it can be expected to pass reviewers).
I would highly appreciate input on this, and especially references to back this
approach in case it is valid.
Thank you very much,
Asaf
