Model-based approaches to multivariate abundance analysis using the mvabund package (Ulrike Naumann)

16:00, October 25, Kennedy Seminar Room 2

Abstract: Multivariate abundance data are abundances collected simultaneously for many ecological taxa (species, orders, functional groups…). This type of data is commonly collected in ecology and the environmental sciences. Multivariate data are common in the environmental sciences, occurring whenever we measure several response variables from each replicate sample. Questions like how does the species composition of a community vary across sites, are multivariate questions. Traditional methods used for analysis try to calculate a measure of similarity between each pair of samples, thus converting a multivariate dataset into a univariate one. However, this approach leads to low statistical power, and does not account for important properties of the multivariate data, such as the mean-variance relationship, or occurrence of rare species (many zeros).

The mvabund package uses model-based approaches by developing a novel set of hypothesis testing tools using the generalised linear models (GLM) framework. We use resampling-based hypothesis testing to make community-level and taxon-specific inferences about which factors or environmental variables are associated with the multivariate abundances. These inference tools take into account correlation between species, which is not possible using standard glm tools. Some more recent extensions of the mvabund package will be demonstrated as well.