Make sure that the relationship between objects
and clustering variables is reasonable (rough
guideline: number of observations should be at
least 2m, where m is the number of clustering
variables). Ensure that the sample size is large
enough to guarantee substantial segments.