STATISTICAL SIGNIFICANCE OF SPECIES CLUSTERS IN ASSOCIATION ANALYSIS

Cluster analysis techniques are often used in preliminary studies of community orga- nization to identify groups of associated species on the basis of distributional co-occurrence. Adequate statistical tests have not been available for determining whether the associations recognized are sufficiently nonrandom to be considered significant, or might reasonably be expected on the basis of random distribution alone. For presence/absence occurrence data, an approximate test of significance may be performed by generating random occurrence matrices within the constraints of a randomiza- tion model which treats as marginal constants both the observed number of occurrences of each species and the observed number of species present at each locality. Null (random) distributions of node values of dendrograms, derived from the randomly generated occurrence matrices, are used to assign approximate critical test values. These values may be applied directly to the dendrogram derived from the observed distribution data to determine which of the observed species clusters may be considered statistically significant. Groups of significantly associated species are reasonable entities within which to examine ecological relationships further.