On 26 May 2000 03:53:36 GMT, [EMAIL PROTECTED] (Wen-Feng
Hsiao) wrote:
> I have 43 subjects to rank 11 stimuli. To obtain the underlying
> variables, principal components, I conduct a PCA (Principal Component
> Analysis) to the obtained ranked-data. However, a friend of mine told me
> that using ranking data to do a PCA is quite dangerous, since the scores
> for the 17 stimuli from each subject is not independent. (I.e. the last
> stimulus always score 17 if we know the first 16 stimuli.) I am not sure
> whether it is right or wrong. Any suggestion?
I don't see where anyone gets "quite dangerous" out of it. If you
have the wrong stat-pack, it might detect the loss of rank and quit,
but that is hardly risky behavior. Toss in a little statistical
"noise", and you are ready to go with any package.
But here are some weaknesses. If you have 11 or 17 stimuli, it would
be nice to have 110 or 170 subjects in order to have a stable
structure, by the usual rule of thumb. 43? Pretty thin.
But you don't have the usual sort of data -- you start your collection
by *throwing away* all the between-subject variance; everyone is
treated as the same average. That could be done by subtracting off
each subject's average; but when you do it with Ranking, you limit
your within-subject information further, by using ranks instead of
meaningful distances.
I don't know of special statistical problems with ranking except that
it probably doubles or triples the sample size needed for stable
correlations for PCA, for the reasons I just said.
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
===========================================================================
This list is open to everyone. Occasionally, less thoughtful
people send inappropriate messages. Please DO NOT COMPLAIN TO
THE POSTMASTER about these messages because the postmaster has no
way of controlling them, and excessive complaints will result in
termination of the list.
For information about this list, including information about the
problem of inappropriate messages and information about how to
unsubscribe, please see the web page at
http://jse.stat.ncsu.edu/
===========================================================================
