Ivi 



Tables for Statisticians and Biometricians [VIII IX 



But the deviations from the observed values of d in the four cases are very 

 considerable. Notwithstanding, if we proceed to determine n st we have : 



from (/3) : n st = d% d d + d 2 



= 56-437 - 27-113 - 28-348 + 12-752 = 13-728, 

 from (a) : n st = 56'674 - 27'313 - 2 8'467 + 12'847 = 13'741, 

 or (/3) only improves on (a) by '013, a quantity of no practical importance. 



Now the standard error of 13'74 in 1000 = x / 18 ' 74 * 9 ^ = 3'68 nearly, 

 corresponding to a probable error of 2'48. 



Clearly 13'74 + 2'48 easily covers the probability of 12 arising in a random 

 sample. Or, the observed cell content of 12 is quite consistent with the hypothesis 

 of normality holding for the correlation of father's and son's statures. 



Illustration (ii). We will take another example from the same correlation table, 

 which indicates a greater variety in the methods of treatment; namely, we 

 will consider the cell for fathers of stature 67"'875 68"'875 and for sons 

 67"'875 68"'875. It contains 27 cases, and the full table condensed for our 

 purposes is as follows : 



We have at once : 



w u = 77, w w ' = 43, w,, = 169, n v ' = 108. 

 Thus the four tables take the forms : 



