A visual heuristic for detecting outliers

It is well known that the conclusions of many statistical procedures can be greatly affected by outliers. In this paper a visual heuristic is discussed, which may be used for detecting single and multiple observations that stand in strident contrast with the rest of data. The proposed technique is a simple line segment plot that shows, for each candidate outlier, the fraction of observations that have that observation as opposite point, that is the point furthest apart in terms of a given metric. This procedure can detect outliers in univariate and multivariate data and it alleviates the problems related with masking or swamping. It also gives good advice on the number of outliers.