Cluster analysis

This chapter discusses some of the methods for using distance between cases to form clusters. One of the fundamental problems with distance measures is deciding which variables to include. The chapter begins by considering the set of three test scores on children who were poor readers that we used in discriminant analysis. Clustering methods all begin with a matrix of distances (or similarities) between each pair of cases. One advantage of using the Manhattan distance is that outlying observations receive less weight, or influence the final result less. The results of a hierarchical clustering are often displayed using a dendrogram or tree diagram. The chapter considers the methods used to decide which cases should be linked into clusters and when. Clusters are amalgamated to form bigger clusters at increasing distances between them, until finally all cases are in one cluster.