© 1981 by Biometrika Trust
MISCELLANEA |
A moment of indecision
Department of Statistics, University of Warwick Coventry
For a suitably thick-tailed prior density and a sufficiently large observation, the sample information dominates the prior information. Similarly, a thick-tailed sampling distribution can be dominated by the prior if the observation is sufficiently large. We show, in the case where the dominating density is normal, that less extreme observations produce very large posterior variances.
Key Words: Bayes inference Large observation Outlier Posterior variance