A Default Prior Distribution for Contingency Tables with Correlated Factor Levels

Antony M. Overstall and Ruth King

University of St. Andrews


A default prior distribution is proposed for Bayesian analysis of contingency table data. The prior is specified to allow for correlation between levels of the factors. Different correlation structures are considered, including a conditional utoregressive structure and a continuous distance matrix. To demonstrate the proposed prior distribution a dataset is considered involving estimating the number of injecting drug users in the eleven National Health Service boards regions of Scotlandusing an incomplete contingency table where the correlation structure relates to geographical regions.


Contingency table; Correlation; Default prior.