356 RAYMOND PEARL. 



mately one fifth (19.6 per cent.) of the mean value of the thing 

 counted it indicates a source of error not lightly to be dismissed. 



III. 



It is desirable next to examine somewhat more closely into the 

 nature and distribution of the discrepancies among the observers. 

 A point of particular interest is to determine to what extent 

 the counts indicate a definite and persistent bias on the part of 

 an observer. There may be great variation in the counts of 

 several observers of the same set of things and yet each observer's 

 judgments may be distributed quite at random about the mean. 

 In order to get more light on this and some other matters 

 Table IX. has been prepared. This table gives in successive 

 columns for the four kernel classes, first, the mean deviation 

 from the mean, all deviations being taken together without refer- 

 ence to sign (i. e., the mean total deviation), and second, the 

 mean net deviation from the mean, got by taking the algebraic 

 sum of the deviations. All four ears are used in getting these 

 mean deviations. An example will make clear the method of 

 obtaining the values given in this table. An examination of 

 Tables II. to V. inclusive shows the following set of deviations 

 from the means in the counts of yellow sweet kernels by observer 

 Xo. V. 



-i- Deviations from Mean. Deviations from Mean. 



7.53 (ear 8) 2.47 (ear 10) 



5.60 (ear 9) 10.00 (ear 1 1) 



13.13 == sum of -f deviations 12.47 :: sum of - deviations. 



= 6.40 = mean total deviation from mean. 

 4 



13.13 - 12.47 



+ 0.165 -- mean net deviation from mean. 



4 



The last column of the table gives the total deviation from 

 the mean of each observer, all cars being taken together and the 

 deviations summed without regard to sign. 



It is strikingly evident from the mean net deviations in this 

 table that each observer was "a law unto himself." \< nl\ 

 everv one of the fifteen evidently had a different system ot si >rting. 



