Bayesian multivariate density estimation for observables and random effects
Jim E. Griffin · 2011
Multivariate density estimation is approached using Bayesian nonparametric mixture of normals models. Two models are developed which are both centred over a multivariate normal distribution but make different prior assumptions about how the unknown distribution departs from a normal distribution. The priors are applied to density estimation of both observables and random effects (or other unobservable random quantities). Markov chain Monte Carlo methods are described for estimation of all models. The models are applied to density estimation for observables and the application of a nonparametric linear mixed model to repeated cholesterol measurements from the Framingham study.