03-22-2012 12:19 PM
Suppose I have many variables (both numerical and character) and I want to reduce them.
I know Proc VARCLUS can work for correlated numerical variables, howvere, is there any method that can work for categorical/character variables?
03-22-2012 12:36 PM
I apologize for not expressing my questions clearly:
My goal of reduction is to reduce the number of categorical/character variables; for example, say I have 200 character variables, I want to reduce this number to 30 (or 50) which is more amenable to analysis (such as building a logistic regression model).
03-22-2012 01:46 PM
Your choices vary according to whether your variables are ordinal or nominal, but look at the DISTANCE procedure and the various proximity measures it provides. Some are applicable for discrete variables. You can then get a "distance" matrix (dis/similarity matrix) that can be used to cluster variables.
I haven't done this myself (I use correpondence analysis instead) butI think it will work.