Statistical design and analysis for a 'biological effects' study

Abstract

Statistical aspects of "biological effects" field surveys are discussed, with particular reference to the GEEP Workshop. Recommendations are made on design criteria, for example, selection of sites and samples, and replication strategies (including formulae for sample size determination). The role of transformations is discussed, both for univariate sub-lethal response data and the multivariate data arising from benthic community studies. Statistical analysis is categorised into testing methods, for establishing biological differences between field sites, and descriptive techniques, for representation of those differences. The former includes a non-parametric randomisation test for use with site-species arrays and the latter a survey of various multivariate ordination and clustering methods. A final section outlines a procedure for comparison of different pollution indices, combining their power to detect specific contaminant inputs with their associated "costs".