Date: Mon, 24 May 1999 10:17:32 -0400
Reply-To: Sarai Batchelder <sbatchelder@EROLS.COM>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: Sarai Batchelder <sbatchelder@EROLS.COM>
Subject: post-hoc confusion and questions
Content-Type: text/plain; charset="iso-8859-1"
I have approached with a question from a student regarding an analysis she
recently completed. In GLM she performed a complicated repeated measures
anova with the following factor levels.
between subjects: 2x4x2
within subjects: 4x2
On the within subject factor, both emerged as significant main effects, as
well as a signifcant interaction. The 4 level factor is different
'situations', or stories, presented to the subjects. She then went on to
perform followup manovas to determine the nature of the effect, by doing
separate manovas for each 'situation'. In these analyses, other factors
became significant which had not been significant in the overall analysis.
I recommended that she correct the F value in the followup analyses using
the MS error term from the omnibus manova.
Some of the questions we have are:
1. Is my advice to correct the F value of the followup manovas good advice?
Is it standard?
2. How does one write up or describe this procedure of correcting F? Is
there a term for this?
3. What should/can be done with the new significant effects if she does not
correct the F values? (I think this is an issue due to some input from a
dissertation committee member.) Is it legit to report it? Should she
mention it but speak about this finding as speculative at best?
4. Is there a way to include post hoc analyses or simple main effects in
the overall analysis and avoid all these issues? In 6.1 this is done using
the MWITHIN command, but she has a later version with the GLM module and was
told to use EMM, but can't seem to get it.
Sarai Batchelder, PhD
Beth Israel Medical Center
Mood Disorders Research
New York City