On Thu, 14 Aug 2003 12:49:24 +0200, Torsten Franz <[EMAIL PROTECTED]> wrote:
me > > > > I was asking, partly from technical curiosity, What algorithm/ or > > whatever/ did you use to convert correlations to distances? > > The other part of my reason for asking was my suspicion > > that you hadn't noticed that conversion is needed. Or, Did > > the computer program offer options for default treatment? > > You still haven't settled that suspicion. TF > > As far as I understand my software (STATISTICA) generats the matrices > directly from the raw data. I do not understand why I've to convert > correlations to distances. But I do not look behind the software, so > maybe I've used procedures which are not assumable. >From a website on Statistica, it looks like the program is willing to analyze correlations. Maybe it assumes there is only one way worth pursuing, for transforming r's to distances, but there were no details on the page I read. me > > > > Hybrid distances? nominal with Interval? Seems awkward, > > at best, to do it or to describe it, whatever MDS is offering. TF > > At the end I used only ordinal variables: I've ranked the nominal > variables after their frequencies of values. Now *that* seems difficult to justify, unless the dimension is 'how conventional is the response' -- I have seen one scale where that is intended. I suppose you could do that when referring to attitudes, but I have only seen it once. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." Justice Holmes. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
