Cluster analyses usually provide more than one cluster solution. The final number of clusters in
SAS® was determined using the squared multiple correlation (R2), the cubic clustering criterion
(CCC), the Pseudo-F, and the pseudo-T2 (PST2) statistic.
Results
Using our example data set, we were unable to obtain meaningful cluster distributions that best
describe both types of data – ordinal or binary – using the six hierarchical agglomerative and kmeans
in SPSS. Similar resistance patterns (phenotypes) that were found in one cluster also existed
in one or more other clusters, which indicated lack of homogenous clusters. In SAS®, cluster
algorithms (hierarchical agglomerative and k-means) with squared Euclidian distance were unable
to cluster MIC values into meaningful clusters with or without standardization or transformation.