Karl Pearson and David Heron 
203 
relationship ; that we dispute entirely the legitimacy of such selection which is of 
a singularly arbitrary character, both in the extent to which it is applied and the 
region to which it is circumscribed. If we apply it to continuous frequency 
surfaces, so that a certain Q remains constant, all other Q's are changed, and r 
taken as a measure of the relationship of the variates as a whole is often 
Diagram IV. Q for frequency surface a right six-face. To illustrate how variation of Q depends 
on form of distribution and how it has no relation whatever to true correlation. Actual corre- 
lation - 3846. 
O'O 
oo 
immensely changed. A correlation table has a multiplicity of Q's and one product- 
moment r ; a process of selectiou, which changes all Q's but one, and widely 
modifies r, has attributed to it by Mr Yule some special merit by which that Q, 
in preference to any other, is considered for the time being to measure the 
" association " of the variates ! 
26—2 
