On Tue, 5 Jul 2016 06:15:42 -0700 (PDT),
mohitmehta.pro@gmail.com
wrote:
Hi,
I need to perform clustering on ordinal variables. I have responses of certain questionnaire , which has 12 questions. The response of 12000 customers is recorded into 12 columns. The values in columns can be anything from 1 to 5.
What method we can use to cluster them. I need output like which questions are key to study one cluster.
Any help is much appreciated.
If these questions make up a "rating scale" where the designers
have attempted to provide equal intervals, then the responses
that range from 1-5 can be (probably) regarded as "interval";
"ordinal" should probably be reserved to describe responses that
are more questionable.
The simple clustering of items for a rating scale is achieved by
performing a Factor Analysis, with varimax rotation to get a
set of factors that you will describe by the strongest loadings.
If your questions are not designed as a rating scale, there could
be a number of other possibilities. For instance, if two questions
are paramount, your best "clusters" might be formed by grouping
continuous cells in their 5x5 table of crosstabulations.
For more comments, you might repeat your question and
provide specifics about the questions and the anchor points.
--
Rich Ulrich
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)