In several areas of research on ecological assemblages, it is useful to be
able to analyse patterns of spatial variation at various scales. Multivaria
te analyses of dissimilarity or similarity in assemblages of species are li
mited by problems of non-independence caused by repeated use of the sample-
units. Where rank-order procedures are used, no comparative quantitative me
asurements of dissimilarity at different scales are produced. An alternativ
e method is described that uses the sample's average assemblage (or centroi
d). These estimates are themselves averaged to give centroids for larger sp
atial scales. Dissimilarities from the centroids at each scale are then cal
culated using independent replicates for each scale from those in each samp
le. The dissimilarity measures can then be examined by analysis of variance
to detect spatial scales of differences for each sample at every level of
a hierarchy of scales. The method is illustrated using data from mangrove f
orests and rocky shores, involving up to 97 taxonomic groups (species, othe
r taxa). Differences among assemblages at the scales of sites (tens of mete
rs apart) or locations at shorts (hundreds of meters apart) were identified
. Consequences of different numbers of replicates are discussed, with some
potential problems (and their solutions) in application.