Clustering in linear mixed models with Dirichlet process mixtures using EM algorithm
Felix Heinzl, Gerhard Tutz · Open access LMU (Ludwid Maxmilian's Universitat Munchen) · 2011
In linear mixed models the assumption of normally distributed random effects is often inappropriate and unnecessary restrictive. The proposed Dirichlet process mixture assumes a hierarchical Gaussian mixture. In addition to the weakening of distributions assumptions the specification allows to estimate clusters of observations with a similar random effects structure identified. An Expectation-Maximization algorithm is given that solves the estimation problem and that exhibits advantages over in this framework usually used Markov chain Monte Carlo approaches. The method is evaluated in a simulation study and applied to dynamics of unemployment in Germany as well as lung function growth data.