Date: Fri, 22 May 1998 08:31:27 -0600
Reply-To: Nathalie Cote <cotenc@CTRVAX.VANDERBILT.EDU>
Sender: "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From: Nathalie Cote <cotenc@CTRVAX.VANDERBILT.EDU>
Subject: ANOVAs on percentages
Content-type: text/plain; charset="us-ascii"
I'm using SPSS as a tool to analyze my dissertation data and I'm puzzled
about whether using orthogonal contrasts in the MANOVA command is the way
to go.
Let me describe the situation. I have two groups of participants. From
think-aloud protocols, I have coded for each person how many comments or
"events" they made in each of 7 categories. So there's a between subjects
factor of GROUP (1,2) and a within subjects/repeated measure factor of
EVENT TYPE (7 levels).
My original plan was to do an ANOVA, predicting that there would be an
interaction between GROUP and EVENT TYPE and that some EVENT TYPES would be
more frequent than others. (I didn't get the interaction but I did get the
main effect of EVENT TYPE.) I want to do contrasts among the EVENT TYPE
means to verify that the differences are where I expected them to be.
Any given subject can have an unlimited total number of events/comments
spread across the 7 event types. Here's a problem: if there's a few people
in one group who talked a lot, that could really affect the means. I'd
wanted to convert the frequency of comments in each category to a
proportion out of total events for each person, because that controls for
how talkative the person was. But I'm told that analyzing proportions is
problematic because they are not completely free to vary across the
categories. So I'm hoping that by embedding the contrasts in the ANOVA, any
possible differences in overall frequency of comments between the groups or
in the interaction between the groups and EVENT TYPE will be controlled for
in the comparisons among the event type means. Am I completely off base?
Could I perhaps do some transformation to the proportions and use that
rather than using frequency of comments?
I've spent a good bit of time with stats books in my lap and I'm still not
confident I understand what to do here. Can anyone help?
****************************
Nathalie Cote'
Research Associate
Learning Technology Center
Box 45 Peabody
Vanderbilt University
Nashville, TN 37203
USA
phone +1-615-343-2614 or +1-615-322-8070
fax +1-615-343-7556
Nathalie.Cote@vanderbilt.edu
