A finite mixture approach to joint clustering of individuals and multivariate discrete outcomes. Issue 11 (24th July 2017)

• Record Type: Journal Article
• Authors: Martella, Francesca; Alfò, Marco
• Is Part Of: Journal of statistical computation and simulation. Volume 87:Issue 11(2017)
• Page Start: 2186
• Page End: 2206
• Publication Date: 2017-07-24
• Subjects: Finite mixtures -- discrete data -- joint clustering -- maximum likelihood estimation; 91C20 -- 62H30; Mathematical statistics -- Data processing -- Periodicals; Digital computer simulation -- Periodicals; 519.5028505
• DOI: 10.1080/00949655.2017.1322593
• Languages: English
• ISSNs: 0094-9655

Abstract

ABSTRACT: In this work, we modify finite mixtures of factor analysers to provide a method for simultaneous clustering of subjects and multivariate discrete outcomes. The joint clustering is performed through a suitable reparameterization of the outcome (column)-specific parameters. We develop an expectation–maximization-type algorithm for maximum likelihood parameter estimation where the maximization step is divided into orthogonal sub-blocks that refer to row and column-specific parameters, respectively. Model performance is evaluated via a simulation study with varying sample size, number of outcomes and row/column-specific clustering (partitions). We compare the performance of our model with the performance of standard model-based biclustering approaches. The proposed method is also demonstrated on a benchmark data set where a multivariate binary response is considered.