Published online by Cambridge University Press: 08 February 2016
Consider a one-dimensional sequence (temporal or spatial) of samples from a many-species community. The sequence may exhibit no steady change, or gradual, steady change, or one or more abrupt, stepwise changes. The pattern is graphically displayed in a similarity matrix, whose elements are the pairwise similarities between every possible pair of samples.
Statistical study of the patterns of similarity matrices will be shown to be informative. A criterion, Q, is proposed for measuring the “disarray” of a similarity matrix; matrices constructed from “perfect” sequences, exhibiting uninterrupted steady change, have Q = 0. The probability distribution of Q in “random” similarity matrices (constructed from sequences in which the ordering is wholly haphazard) is examined.
Several applications are illustrated, using data on foraminiferal assemblages in marine sediment cores as examples. Use of the criterion permits the following: (1) Objective recognition of regions in which stratigraphy is especially well preserved. (2) Selection for study of cores in which the sequence of change is especially clear, and rejection of cores in which the natural sequence has been obscured by reworking. (3) Judgment as to whether, in sequences exhibiting stepwise changes, gradual change between the steps occurs as well. (4) Discrimination between species-sets that track gradual environmental change and sets that switch abruptly. (5) Objective choice of the best similarity index for measuring pairwise similarities in a given body of data.