From: nonlinear@rogers.com   
      
   William Siler wrote:   
   > Paul Victor Birke  wrote in message news   
   <3FE76A21.3040101@rogers.com>...   
   >   
   >>Just got a questionnaire from Home Depot this afternoon   
   >>   
   >>the commom memebership lingo of the following five (for the store in   
   >>this case)   
   >>   
   >>excellent   
   >>above average   
   >>average   
   >>below average   
   >>poor   
   >>   
   >>while we might set average = 0.5 and above and below at say 0.75 and   
   >>0.25 and finally excellent and poor at 0.95 and 0.05 how do we handle   
   >>the density distributions for each of the membership labels 1-5 above?   
   >>   
   >>that is we may have a total of 10,000 received satisfaction questionnaires   
   >>   
   >>What do we do with the distribution at each wrt to defining our   
   >>memebership function or modifying same.   
   >>   
   >>make a new membership nomalized on the max of the 5 above and redefine   
   >>the membership labelling somehow?   
   >>   
   >   
   > A problem here is that the concentration on fuzzy control that has   
   > been so fruitful also has a flip side; fuzzy people sort of   
   > automatically assume that a fuzzy set represents a number, and has a   
   > membership function to convert a number into grades of membership of a   
   > fuzzy set. Not all fuzzy sets are linguistic variables; not all fuzzy   
   > set represent numbers.   
   >   
   > This is further complicated here by the fact that in this case, we are   
   > supplied with the grades of membership by the answered questionnaire.   
   > I presume that the instructions told you to check one of the answers   
   > excellent, above average, ...   
      
      
   Indeed   
      
   If this is the case, in our set of   
   > replies the checked reply has grade of membership 1, and the others   
   > zero. No membership function is needed.   
      
   But look at wha happens in  a day or week when you are looking at the   
   whole result.  Herein, we have the density of each grade giving us   
   information, yes?!   
   >   
   > Actually, what we have is a categorical variable being input, and no   
   > dependent variable at all.   
      
   There is a dependent varialbe that of graded satisfaction.   
      
   We are into a whole new ball game. Fuzzy   
   > theory can be used here, but in a very non-standard way, to help us   
   > visualize what the set of replies means. The basic question is, before   
   > deciding on a method, is to define what we want the analysis to   
   > produce. Do we want to describe the customers? The store help? The   
   > different stores? Then we can go about finding a proper method.   
   No doubt this is a bit unclear, how the information is in fact being   
   used.  That is really the point of my question, how to use the   
   information especailly the PDF information for each grade label.   
      
   >   
   > See my GoM analysis description in my previous posting (I give the   
   > URL) to get an idea of what GoM can do.   
      
   Dear Bill   
      
   I will have a look.   
      
      
   There's also latent class   
   > analysis, factor analysis, logistic analysis, ANOVA and ANCOVA, all   
   > with different virtues and drawbacks for a particular purpose. First,   
   > we need to define the purpose.   
      
   Indeed!   
      
      
   all the best   
      
   Paul   
   >   
   > William Siler   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)