OF ATTRIBTTKS IN STATISTIC 



291 



1'influence des causes qui moclifient les dldmens sociaux,'* but it is either explicitly 

 or implicitly used in most statistical discussions of causation. To take an example 

 from QUKTKI.KT'S pamphlet, I give below a table of the chances of condemnation of 

 various categories of prisoners in the French Assize Courts during the years 1825-30. 

 Here A must stand for condemnation, B, C, D for the various attributes of the 

 accused (superior education, being a wonum, being a man, &c.). The chances 

 tabulated are then (AB)/(B), (AC)/(C), &c., except No. 8, which is (A)/(U). 

 QUKTKI.KT went further than a tabulation" of the simple chances, and used as a 

 measure of the " degree of influence " of the cause the function 



(AB)/(B) - 



(14), 



these measures being given in the second column. 



33. Now from the work in 22, p. 272, we have at once 



AD (A) _ (AB) -(A)(B)/(U) _ (AB) - (AB),, _ (AB)(o0) - (aB)(A0) . , 

 (T.) "(U)- (H) (B) (U)(B) 



so that if (AB)/(B) > (A)/(U), A and B are certainly positively associated, and if 

 (AB)/(B) < (A)/(U) negatively associated. It does not follow, however, that if 

 (AB)/(B) > (AC)/(C), A and B are more closely associated than A and C. If we 



write 



Pl = (AB)/(B) 

 Ps = (AC)/(C) 



and if K , K., are the values of the functions K for AB and AC, then 



* It is entitled " Lettre a M. Willernio de 1'Institut de France." Bnixelles, 1832. (Royal Statistical 

 Society's Library. Tracts, S. 4, vol. S.) 



VOL. CXCIV. A. 2 O 



