Bayesian inference for mixed effects models with heterogeneity
2016 (English)Report (Other academic)
We are interested in Bayesian modelling of panel data using a mixed effects model with heterogeneity in the individual random effects. We compare two different approaches for modelling the heterogeneity using a mixture of Gaussians. In the first model, we assume an infinite mixture model with a Dirichlet process prior, which is a non-parametric Bayesian model. In the second model, we assume an over-parametrised finite mixture model with a sparseness prior. Recent work indicates that the second model can be seen as an approximation of the former. In this paper, we investigate this claim and compare the estimates of the posteriors and the mixing obtained by Gibbs sampling in these two models. The results from using both synthetic and real-world data supports the claim that the estimates of the posterior from both models agree even when the data record is finite.
Place, publisher, year, edition, pages
Linköping: Linköping University Electronic Press, 2016. , 22 p.
LiTH-ISY-R, ISSN 1400-3902 ; 3091
Bayesian inference, mixed effects model, panel/longitudinal data, Dirichlet process mixture, finite mixture, sparseness prior
Probability Theory and Statistics Control Engineering
IdentifiersURN: urn:nbn:se:liu:diva-126680ISRN: LiTH-ISY-R-3091OAI: oai:DiVA.org:liu-126680DiVA: diva2:916319
FunderSwedish Research Council, 621-2013-5524