Multivariate analysis (MVA) becomes necessary when there are many response variables measured on the same experimental unit. In ecology, most biodiversity surveys end in a list of species per sample with some abundance measure for each species, e.g. counts, biomass, presence/absence or percentage cover. A myriad of multivariate analysis tools exists for use with such data depending on the question we want to answer. Most researchers are familiar with the association-based MVA methods, such as multi-dimensional scaling or correspondence analysis. Recently more model-based methods have been developed, one of which is the use of generalised linear models to analyse multivariate abundance data. Natasha Karenyi introduces the ‘mvabund’ package as a way to perform these model-based multivariate analyses.
