Default Bayesian model determination for generalised liner mixed models

Download

PDF Download (6Mb)

Description/Abstract

In this thesis, an automatic, default, fully Bayesian model determination strategy for GLMMs
is considered. This strategy must address the two key issues of default prior specification
and computation.

Default prior distributions for the model parameters, that are based on a unit information
concept, are proposed.

A two-phase computational strategy, that uses a reversible jump algorithm and implementation
of bridge sampling, is also proposed.

This strategy is applied to four examples throughout this thesis.