Identifying Outliers in Multi-Output Models

A firm may be different from other firms either in terms of the mix or scale of its input-output vector. This paper develops separate mix and scale measures of dissimilarity, and shows that these can be additively aggregated into a measure of absolute dissimilarity. The mix measure is particularly relevant in the context of frontier analysis since it will identify the firms that exert the most influence on the resulting efficiency scores. Copyright Kluwer Academic Publishers 2004