"Dennis Roberts" <[EMAIL PROTECTED]> wrote in message
news:[EMAIL PROTECTED]
...
> some on various lists have claimed that the coefficient of variation ...
...
> i have NEVER been convinced that this is so
...
> does anyone out there?

FWIW, I've used coefficient of variation as a crude measure of reliability
in the following context: (data are from the field of orthodonty; the
article is going to be published :)

several parameters (various tooth angles or their ratios - so always
positive and <ratio scale> values) were measured for each subject, and each
subject was measured three times (or by three orthodontists, or by three
devices - I can't remember right now), and I calculated CV within each
subject for each parameter, and used median CV (since many values were zero
and just some about, say, 20%) across subjects as a crude measure of
reliability which could then be (purely descriptively, of course) compared
between parameters.

Of course, this is a textbook case for which ICC methodology is apropriate,
but neither the co-authors nor <leading international experts> (who heard
the presentation at a conference and suggested publication in a rather
prestigeous journal in the first place) were happy with that since they said
in that field nobody would understand it -- the CV approach that I then
thought of was just on the edge of understandable and at the same time
"technical" enough to fascinate them :)

Anyway, as stated in the first place, FWIW and just as an example. Which,
just to tease the statistical community, can even be instantaneously and
accurately implemented in Excel :)

Gaj Vidmar
Univ. of Ljubljana, Inst. of Biomed. Informatics


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to