A least squares algorithm for a mixture model for compositional data

✍ Scribed by Ab Mooijaart; Peter G.M. van der Heijden; L.Andries van der Ark

• Publisher: Elsevier Science
• Year: 1999
• Tongue: English
• Weight: 136 KB
• Volume: 30
• Category: Article
• ISSN: 0167-9473
• DOI: 10.1016/s0167-9473(98)00098-x

✦ Synopsis

The estimation of a model for compositional data is studied where the data are approximated by a mixture of latent compositions. This model is variously known as "endmember analysis" or "latent budget analysis". Two estimation procedures are available. The ÿrst uses a procedure which is incorrect in the sense that, although it claims to be a least squares procedure, it does not always minimize a least squares criterion. The second uses a maximum likelihood procedure starting from assumptions that are often violated for compositional data. In this paper we propose a constrained (weighted) least squares procedure for the estimation of the model.