Finding Groups in Ordinal Data: An Examination of Some Clustering Procedures

The article evaluates, based on ordinal data simulated with cluster.Gen function of clusterSim package working in R environment, some cluster analysis procedures containing GDM distance for ordinal data (see Jajuga et al. 2003; Walesiak 1993, 2006), nine clustering methods and eight internal cluster quality indices for determining the number of clusters. Seventy two clustering procedures are evaluated based on simulated data originating from a variety of models. Models contain the known structure of clusters and differ in the number of true dimensions, the number of categories for each variable, the density and shape of clusters, the number of true clusters, the number of noisy variables. Each clustering result was compared with the known cluster structure from models applying (Hubert and Arabie 1985) corrected Rand index.