Abstract
This article is concerned with the detection of outliers in a binomial sample. A Bayesian approach to the modeling of outliers is presented and examined. It is supposed that most observations are from a binomial distribution with mean n but a small number of observations may be contaminated. That is, they are generated from a binomial sample with mean inflated (or deflated) by a factor δ. Bayes factors for the cases when either proper or improper priors are specified for π and δ are discussed. An alternative approach, to transform the Binomial observations to approximate normality, is also presented.
Key Words: