COMPSTAT 2016: Start Registration
View Submission - COMPSTAT
Title: Model-based ordination method for overdispersed count data Authors:  Jenni Niku - University of Jyvaskyla (Finland) [presenting]
Francis Hui - Australian National University (Australia)
Sara Taskinen - University of Jyvaskyla (Finland)
David Warton - University of New South Wales (Australia)
Abstract: Unconstrained ordination methods are commonly used in ecology to visualize the relationships between different sites in terms of their species composition. Classical unconstrained ordination methods, such as non-metric multidimensional scaling, are algorithm-based techniques, which are developed and implemented without directly taking into account the statistical properties of the multivariate data. The ignorance of important data properties (such as mean-variance relationship) can then yield to misleading results. We consider a model-based approach to unconstrained ordination. The method uses a generalized linear latent variable model to produce an ordination plot. As usual, a model-based approach gives us tools for diagnostics, model selection and statistical inference. Examples and simulations are presented to illustrate the method in case of overdispersed count data. The results are compared with the results based on classical unconstrained ordination methods.