The generalization of Edwards's argument for the measure of association of the rows and columns of 2 x 2 table, to that of an r x s table whose rows and columns are assumed unordered, shows, not surprisingly, that association ought to be measured by some function of the (r-l) (s-1) cross-ratios. Such a function is suggested by the introduction of a metric on certain equivalence classes. The properties of such metrics are examined, and in particular comparisons are made with Good's suggestion of the use of the algebraic rank of the contingency table, and with Lindley's significance test for association in the r x stable.
GENERALIZATION OF EDWARD'S ARGUMENTFOR THE 2 x 2 TABLE EDWARDS (1963) has shown that any measure of association of rows and columns for a 2 x 2 contingency table should be a function of the cross-ratio (Pll P22)/(P12 P21)' where the table is represented by (Pij; I~i,j~2). This argument is generalized below to the r x s contingency table, and some suggestions for measures of association are given.Let Pr,s be the class of r x s contingency tables, that is, Pr,s={(Pij):Pij>O,I~i~r,l~j~s and~~Pij= I} It is presumed that no orderings of the rows and columns of tables of P, s are available.Let~be an equivalence relation such that for p,qEPr,s,p~q means that the "association" of rows and columns of P is the same as the "association" of rows and columns of q. A precise definition of "association" is deliberately omitted at this point, but following Edwards, two propositions about it are made.Proposition I If P, q E Pr,s are such thatPij qij -= -, l~i~r, l~j~s-l, Pi. qi.then p~q (employing the usual dot notation for summation over a suffix).
64ALTHAM -Column and Row Measurements for a Contingency Table [No.1, Proposition II If P, q E Pr,s are such that Pij _ qij ---, 1~i~r-l, 1~j~s, P.j q,j A then p--q. Let O,.!! be a relation such that for p,qEPr,s' po"!!q iff PijPrs qijqrs--= --, 1~i~r-l, 1~j~s-l.
PisPrj qisqrjThis is easily seen to be an equivalence relation. The generalization of Edwards's result is the following.
OR ATheorem. If p,q EP r.s and p--q, then p-q. Proof. Define .
PijPrs qijqrs ()(.ij=--=--, l~i~r, l~j~s.PisPrj qisqrj Define 7Tij = 7To ()(.ijPisqrj, where 7To is the normalizing constant, i.e. such that 7TEP r,s' Then 7Tij/7Ti. = qij/qi., 1~i~r, 1~j~s-l, so that 7T1.,q and 7Tij/7T.j = Pij/P.j, 1~i~r-l, 1~j~s, so that 7T1.,p.Hence, by the transitivity of 1." p1.,q as required.Similar versions of this result for r x s x t tables are easily obtained, given suitable modification of Propositions I and II. There are essentially four different types of dependence of rows, columns and layers to be considered, and accordingly four different types of cross-ratio arise as being relevant for the description of these four types of association. An example follows. LetPr,s,!={(Pijk):Pijk>O, l~i~r, l~j~s, l~k~t, and :S:S:SPijk= I}.Let !!., be an equivalence relation such that for P, q E P; s t>P!!., q means that the second order interaction of the rows, columns and layers of P' is the same as that of q. The...