07-31-2015 08:35 AM
This pertains to the output data from PROC FASTCLUS. This may sound very ignorant, but I'm wondering if you get a measure of separation between clusters (a kind of "Between STD") when you subtract Within STD from Total STD? I would then be interested in the ratio between this "Between STD" vs Within STD. If not, should R^2 ratio be preferred as a measure of separation of clusters vs within cluster variation?
08-05-2015 04:12 PM
The VariableStat table contains a column labeled RSQ/(1 - RSQ). The documentation says that that ratio is "the ratio of between-cluster variance to within-cluster variance." Is that what you are looking for?
I don't think what you suggest would be true (but I don't fully understand what you are proposing). In my experience, it is usually the VARIANCE (not the STD) that you try to decompose into interpretable independent components.