Further nonparametric tests for comparing dissimilarity matrices based on the relative neighborhood graph

Abstract Using the edge sets of their relative neighborhood graphs, contingency tables may be constructed to compare dissimilarity matrices. Two kinds of tests of independence are considered, one based on the relationship between the minimum path lengths in the relative neighborhood graphs, the other on the product-moment correlation coefficient between the dissimilarities corresponding to the unit path lengths of the intersection. To compare a dissimilarity matrix with a transformation of it, such as by a clustering, various elements of the contingency table are compared with the binomial distribution having known parameters, while for independently obtained dissimilarity matrices, standard tests of marginal independence are appropriate.