The Multivariate Analogue of JAN

Abstract

The purpose of the univariate version of JAN (Judgement Analysis) is to take a situation in which judges are rating subjects on a single attribute and, through regression techniques, "capture" the policy of each judge and iteratively cluster the judges on the basis of policy similarity. The present paper extends this procedure to situations in which the judges rate the subjects on a multiplicity of attributes. Computational simplification is discussed, and a numerical example is given.