I agree with Ken here.
Sometimes, although not often, a cluster analysis ends up re-discovering
known
structure in the data and/or showing clusters that are obvious
graphically.
When that's true it is moot whether the cluster analysis is needed at
all.
Depending partly on how many observations or variables you have,
another possible graphical method
is a parallel coordinates plot (often called a profile plot).
One implementation is -parplot- from SSC.
Nick
[email protected]
Ken Higbee replied to Vidya Mahambare
> I am using the command "cluster kmedians" in order to cluster a sample
of
> objects on the basis of a certain variables.I want to check how the
> clusters change when a different group of variables is used. However,
when
> I run the command using a different variables or by adding or deleting
> certain variables, the cluster do not change. This does not seem
correct.
> Please suggest what could be the problem.
There is no problem, and in fact many researchers who are doing
cluster analysis would love to have the situation you describe.
It sounds like your data fall naturally into k distinct groups
and those groups are distinctly separable based on one or more of
your variables.
I don't know how many variables are involved in your cluster
analysis, but you might want to use -graph matrix- on some of the
variables and the distinctness of the groups might become
apparent to you.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/