Rich Ulrich <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>... > On Wed, 13 Aug 2003 10:13:47 +0200, Torsten Franz > <[EMAIL PROTECTED]> wrote: > > > > Do you expect the same result for different input? ! > > Definitely not, although the underlying data set is the same. My > > question was: why are the results different? Because the matrices are > > different? Is the answer so simple? And which conclusions I have to draw > > for the interpretation of the MDS? > > > > I tried to do MDS on correlations, 15 years ago when I > was trying to figure out if MDS was worth anything, and > I never figured out how I was supposed to reverse the > correlations in order to have a 'distance' metric that would > work. Did you have some concrete advice from somewhere?
Try adding the coauthor name "Young" to your Google search. If you interpret a correlation coefficient as the cosine of an angle, a, you can represent the data in terms of points on a unit 3-dim sphere. When the angle between two position vectors emanating from the sphere center is a, the distance between the points is 2*sin(a/2). Hope this helps. Greg > Since the results I could get on correlations did not look > nearly as meaningful to me as what Factor Analysis > gave me, I decided to forget about MDS. > > I googled for > < "Multidimensional scaling" FAQ > and two of the first 4 hits > were to my own FAQ -- that is not encouraging. > < "Multidimensional scaling" tutorial > gave a better looking > set of references. > > Has somebody cared enough about MDS to update the > computer programs? It's long been my impression that > 'marketing' was using MDS. From google, it also seems > like MDS sometimes is included in the tools of data mining. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
