UW Aquatic & Fishery Sciences Quantitative Seminar

Marti Anderson

University of Auckland Department of Statistics

Multivariate methods for multi-species data

Abstract

Abstract: In recent years, there have been a number of developments in the statistical analysis and display of multivariate data in the form of counts of abundances of species. Such counts are well known to "mis-behave", by having many zeros and overdispersion, with right-skewed distributions. In addition, there are often many more species (response variables) than there are sampling units (cores, trawls, quadrats, etc.). Traditional statistical methods (such as MANOVA or PCA), which are also intrinsically based on Euclidean distances among sample units, are either inappropriate or impossible to use in such cases. In this seminar, I will give a general conceptual overview of dissimilarity-based methods which do not make any specific assumptions regarding species distributions, and which allow rigorous tests of multivariate hypotheses in community analysis through the use of permutation testing procedures. I will also discuss recent methods for modelling and predicting the position of new sample units in canonical spaces, which opens the doors for monitoring and impact assessment. Examples will include the analysis of New Zealand fish biodiversity surveys and the assessment of "ecosystem health" in benthic soft-sediment estuarine systems of the Auckland region.