Title: Generalized linear latent variable models in the analysis of ecological data
Authors: Sara Taskinen - University of Jyvaskyla (Finland) [presenting]
Jenni Niku - University of Jyvaskyla (Finland)
Abstract: Very high-dimensional multivariate abundance data, which consist of records (counts, presence-absences, biomass) of a large number of interacting species at a set of units or sites, are very common in ecological studies. When analysing such multivariate abundance data the interest is often in visualisation of correlation patterns across taxa, hypothesis testing of environmental effects and making predictions for abundances. In several recent studies, a model-based joint analysis is shown to be a promising method for studying non-normal multivariate abundance data. One particular approach is the use of generalized linear latent variable models. These are constructed by fitting generalized linear models to each species, while including latent variables to account for residual correlation between species, for example, due to unmeasured covariates. Notice that in other fields, e.g in psychometrics and social sciences such models have been popular for a long time, however models used in such fields are not often suitable for ecological data. Some latest developments in the field of GLLVMs applied to ecological studies are discussed. The theory is illustrated using examples.