A0566
Title: Practical aspects of shape mixture constructions emanating from a Dirichlet setup
Authors: JT Ferreira - University of Pretoria (South Africa) [presenting]
Tanita Botha - University of Pretoria (South Africa)
Andriette Bekker - University of Pretoria (South Africa)
Abstract: The Dirichlet distribution is arguably the most well-known multivariate distribution for implementation on the unitary simplex. Different generalisations exist, which include a shape mixture of Poisson weights known as the noncentral Dirichlet. This noncentral representation depends on the noncentrality parameters through the confluent hypergeometric function of several variables and admits both singly- and doubly noncentral representations; computational aspects are explored when the estimation of this singly and doubly noncentral Dirichlet is of interest. We investigate to what degree the additional parameter(s) and their effect on the doubly noncentral Dirichlet, compared to the singly alternative, affects the practical implementation of the model. Real data examples are used for this investigation by using maximum likelihood estimation for the parameters and further strengthened by simulation studies.
