Influence of groups of observations on bayes factors

Abstract

A diagnostic for finding groups of observations influential on Bayes factors is discussed, which extends ideas in Pettit & Young (1990). Ways of reducing the combinatorial explosion involved in detecting more than one influential observation are considered. The effect of masking is also examined. Finally new graphical displays to identify these observations will be explored.

Keywords:

• Bayes factor
• Influential observations
• Masking