Visualizing non-hierarchical and hierarchical cluster analyses with clustergrams

In hierarchical cluster analysis dendrogram graphs are used to visualize how clusters are formed. Because each observation is displayed dendrograms are impractical when the data set is large. For non-hierarchical cluster algorithms (e.g. Kmeans) a graph like the dendrogram does not exist. This paper discusses a graph named “clustergram” to examine how cluster members are assigned to clusters as the number of clusters increases. The clustergram can also give insight into algorithms. For example, it can easily be seen that the “single linkage” algorithm tends to form clusters that consist of just one observation. It is also useful in distinguishing between random and deterministic implementations of the Kmeans algorithm. A data set related to asbestos claims and the Thailand Landmine Data are used throughout to illustrate the clustergram.