Abstract: Order-statistics models may be used to estimate rank covariances, and the choice between the various models is determined by general considerations such as maximum entropy and/or the symmetry of the ranking process. In practical cases these covariances are fairly close to the sample covariances, and may be used in preference to sample covariances when the sample sizes are small (of order 10 to 20).
Key words and phrases: Covariance matrix, logistic models, mean ranks, order-statistics models, rank preferences.