MEASUREMENTS OF RELATIONSHIPS 



21 



and x/y ratios together) because the correlation is much closer for 

 mediocre values of x and y than for extreme values (see especially 

 the regression of x on y}. U is .292 and r from cos T7 is accorcU. 

 ingly .61. 



This case illustrates the fact that the relation of y to x may not 



be the same as that of x to if even when the form of distribution and 

 variability is the same for both cases. It also illustrates a rather 

 close approach to the so-called 'normal' correlation. 



FIG. 5. 



Table X. gives graphically the correlation in the case of age at 

 death of husband with age at death of wife in 935 pairs from 

 records of the Society of Friends. This is taken from the table 

 on p. 498 of Vol. I. of Biometrika, the table being due to Mary 

 Beeton in cooperation with Karl Pearson. This case shows a rela- 

 tionship between two series neither of which is anything like normal 

 in form of distribution, which are not of the same form of distribu- 

 tion and which therefore are in strictness incomparable in varia- 

 bility. 



