“…They note that the second of these cases, e 1 (A,B), has several desirable theoretical properties, one of which (that of statistical consistency as defined by Kaufman and Rousseeuw [22]) is not exhibited by e 2 (A,B) (i.e., Ward's method). They suggest that clusters based on e 1 (A,B) will be superior to those based on e 2 (A,B) for separating clusters with the same, or close, cluster means but different distributions of points around those means, and use dermatological, gene expression and simulated multivariate normal datasets to demonstrate that e 1 (A,B) can indeed out-perform e 2 (A,B) in some circumstances [18]. The Székely-Rizzo clustering method is available as a contributed package, called Energy, in the R statistical system (available from http://www.r-project.org/), and this was used for all the experiments reported below.…”