Statistical methods for analysing changes in community structure fall under the three general headings of univariate, graphical/distributional and multivariate. These methods are applied to a variety of benthic community data (macrobenthos, meiobenthos, corals, demersal fish), from a variety of localities (intertidal/subtidal, temperate/tropical) and over both spatial and temporal scales. Four general conclusions emerge from this comparative study:
(1) The similarity between sites or times based on their univariate or graphical/distributional properties is usually different from their clustering in multivariate analyses.
(2) Species dependent (multivariate) methods are much more sensitive than species independent (univariate and graphical/distributional) methods in discriminating between sites or times.
(3) In examples where more than one component of the fauna has been studied, univariate and graphical/distributional methods may give different results for different components, whereas multivariate methods tend to give the same results.
(4) By matching multivariate ordinations from subsets of environmental data to an ordination of faunistic data, the key environmental variables responsible for community change may be identified.