Mixture models for multi-species and environmental data

This poster was presented for the International Biometric Conference, held in Florence from 6 to 11 July 2014.

  • Ecological inference and management decisions often depend on data from many species.
  • A proper and useful statistical analysis quantifies the important patterns of variation, whilst reducing the complexity in multi-species data.
  • Currently, analysis is frequently done by: 1) performing species-by-species analyses (e.g. univariate regression and extensions) and then
  • combining results, or 2) by combining data (clustering) and then performing a group-by-group analysis.
  • Neither of the standard approaches are entirely satisfactory as important aspects of the variance in the data can be lost when moving from step
  • to step. Also, the propagation of uncertainty is difficult and is subsequently (often) ignored.
  • We introduce two models, based on mixture models, that address these issues. One model type, species archetype models (SAMs) exploits
  • similarities in individual species’ responses to the environment. The second type, regions of common profile (RCP) models, exploits similarities in
  • the assemblage patterns at each site.
Document type: 
Posters and banners