Constructing Software from Graphical Models

Graphical models have been widely used to describe complex statistical models and to communicate these models to people who do not have a statistical background. In this paper we discus the WinBUGS software package which is based on these two ideas. A declarative language, the BUGS language, has been developed to describe complex statistical models. This language is processed by a compiler to produce a low level representation of the statistical model suitable for computational. Inference is then made by using Markov Chain Monte Carllo (MCMC) simulation. Throughout this paper a full probability or Bayesian approach to statistics is assumed.