Interpretation of average ranks

Abstract

Given only the average ranks ξ 1 ,…,ξ k allocated by m judges to k objects, we can use order statistics models to find a value for the average of Kendall's τ between the judges' rankings and the true ranking. The choice of model is most naturally determined by an extremum principle conditional on the given average ranks: in particular, if the entropy of pairwise comparisons is maximized we are led to the shifted extreme-value model.