ABSTRACT
Hierarchical divisive methods, also known as disjoint cluster analysis, starts with all the objects in one cluster. This cluster is then divided into two clusters. Divisive clustering usually produces clusters in which all the members within a cluster are very similar. These are called monothetic classifications. The Howard-Harris method selects the one variable, in the set of variables that has the largest variance. The method uses a K-means iterative solution to determine the membership of the two clusters. The divisive clustering procedure is valuable because it can be applied to as many as 2000 subjects and as many as 20 variables. The method is part of the PC-MDS package of programs provided by BYU. The FORTRAN code for this method is given in Scaling Methods. Chambers and Kleiner suggested different ways of displaying the results of clustering. The dendogram can be altered by shortening the end lines or by making a straight line connection to the nodes.
