With the help of certain inequalities concerning the elements of the dispersion matrix of a set of statistics, and of the information matrix, the following results have been proved. Some of these inequalities are extensions of results given by Fisher (1) in the case of a single parameter.

(i) Efficient statistics are explicit functions of the minimal set of sufficient statistics.

(ii) Functions of the minimal set of sufficient statistics, satisfying the property of uniqueness defined in the text, are best unbiased estimates. Under certain conditions estimates possessing exactly the minimum possible variance can be obtained by the method of maximum likelihood.

(iii) In large samples maximum likelihood estimates supply efficient statistics in the case of several parameters.

(iv) The importance of replacing the sample by an exhaustive set of sufficient statistics (referred to in this paper as the minimal set) as a first step in any methodological problem has been stressed by R. A. Fisher in various articles and lectures. The above discussion supplies a formal demonstration of this view so far as the problem of estimation is concerned.

A general form of the covering principle and relative differentiation of additive functions. II*.

By A. S. Besicovitch

P. 5, line 7 from the bottom, after ‘F-singular’ add ‘in the class of B-sets’.

P. 5, the bottom line, after ‘function Θ(X)’ add ‘in the class of B-sets’.

P. 7, line 9 from the bottom, forG⊃Xnread.

P. 8, line 4, for.

P. 8, line 5, for.

* Vol. 42 (1946), pp. 1–10.