Thanks to all of you for your most helpful comments. I think I see the
problem more clearly now.
Indeed, some of the proportions we want to analyze have a large clump at
zero. For these variables, there is no useful transformation that makes
them more symmetric. It might be more informative to treat them as
dichotomic ("none" vs. "some"), as Rich suggested.
We have other proportions, however, that clump at lower values, with no
or little zeros. For these variables, one of the transformations you
suggested might be useful. By the way, the book chapter by Emerson in
Hoaglin, Mosteller & Tukey ("Fundamentals of Exploratory Data Analysis")
seems helpful to me as a good introduction to the topic (and this is
what I needed).
> Why would symmetry be necessary?
The reason why I asked the question in the first place: We would like to
include the proportional variables together with other continuous (and
more symmetrically distributed) variables in MANOVAs. This makes sense
conceptually, because the single variables can be regarded as aspects of
more complex constructs.
Many thanks again,
Tobias
--
---------------------------------------------
Tobias Richter, Dipl.-Psych.
University of Cologne
Psychological Department, General Psychology
Herbert-Lewin-Strasse 2
50931 Koeln, Germany
Phone +49-(0)221-4703848
Mob. +49-(0)179-2203314
E-Mail [EMAIL PROTECTED]
---------------------------------------------
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
