DUGi

Log-ratio methods in mixture models for compositional data sets

http://dugi.udg.edu/item/http:@@@@hdl.handle.net@@2072@@299173

When traditional methods are applied to compositional data misleading and incoherent results could be obtained. Finite mixtures of multivariate distributions are becoming increasingly important nowadays. In this paper, traditional strategies to fit a mixture model into compositional data sets are revisited and the major difficulties are detailed. A new proposal using a mixture of distributions defined on orthonormal log-ratio coordinates is introduced. A real data set analysis is presented to illustrate and compare the different methodologies

This research was supported by the Ministerio de Economía y Competividad through the projects “METRICS” and “CoDa-RETOS” (MTM2012-33236; MTM2015-65016- C2-1-R: MINECO/FEDER,UE) and the Agència de Gestió d’Ajuts Universitaris i de Recerca (AGAUR: 2014SGR551)

Institut d’Estadística de Catalunya (Idescat)