BOOTSTRAPPING FINITE MIXTURE MODELS

COMPSTAT 2004 section: Clustering, Resampling methods. Abstract: Finite mixture regression models are used for modelling unob- served heterogeneity in the population. However, depending on the speciì- cations these models need not be identiìable, which is especially of concern if the parameters are interpreted. As bootstrap methods are already used as a diagnostic tool for linear regression models, we investigate their use for ìnite mixture models. We show that bootstrapping helps in revealing identiìability problems and that parametric bootstrapping can be used for analyzing the reliability of coe cient estimates.