B0560
Title: A new look at the Dirichlet distribution with applications in model-based clustering
Authors: Salvatore Daniele Tomarchio - University of Catania (Italy) [presenting]
Antonio Punzo - University of Catania (Italy)
Andriette Bekker - University of Pretoria (South Africa)
Johan Ferreira - University of Pretoria (South Africa)
Abstract: The analysis of compositional data poses many challenges because of their peculiar characteristics. The Dirichlet distribution is the most commonly adopted for modelling this kind of data. A convenient mode-based parametrization that yields the unimodal Dirichlet distribution is used. Such parametrization allows/simplifies the use of the Dirichlet distribution in various branches of statistics. In particular, by using simulated and real data, examples of the usefulness of such a parameterization are provided in robust statistics and model-based clustering.
