ABSTRACT
A strength of the Bayesian graphical modelling techniques of BUGS is the way they can represent the typical complexities of real data. This chapter explains various generic issues encountered in data analysis and how they can be addressed in BUGS. For example, data commonly include missing values and measurement errors. A realistic model may need to account for censoring, truncation, grouping, rounding, or constraints on parameters, or use a sampling or prior distribution not already included in BUGS. We also discuss prediction, controlling “feedback” in graphical models, classical bootstrap estimation, and expressing uncertainty surrounding “ranks” or positions in a league table. Each of the techniques we describe may be deployed as part of any model in BUGS, with typically only a few extra lines of code.
