EXPERIMENTS IN FIELD PLOT TECHNIC 53 



four, eight, and twelve ranges, and in the whole field of 24 ranges, is 

 shown in Table 27. 



The variability of the whole field of 24 ranges was 12 per cent 

 greater than the average variability of sections of four ranges each. 

 In this case again, five replications in the larger field would have given 

 less accurate results than four replications in the smaller. 



In each of the cases cited above a steady increase in variability 

 is apparent as the size of the experiment field is increased. It is obvious 

 that the substitution of 3-row plots with discarded borders for single 

 rows will result in greater variability, and will require increased rep- 

 lication for the same degree of accuracy. 



From the foregoing statements it will be clear that the number of 

 replications necessary for a given degree of accuracy may vary con- 

 siderably with conditions. The number to be used in any specific ex- 

 periment should be determined from the variability of the field in 

 question and the degree of accuracy required. The variability of the 

 check plots is usually considered a measure of the variability of the 

 field. But when the number of replications to be used or the extent of 

 experimental error is determined from the variability of the check 

 plots, it is assumed that the variability of different varieties of the same 

 crop is approximately the same under the same conditions. This of 

 course is not strictly true. The yield of two varieties may be deter- 

 mined by very different factors, as has been stated, and their relative 

 variability may also be quite different. The variability of 120 plots 



TABLE 28. SOIL HETEROGENEITY of AN EXPERIMENT FIELD AS DETERMINED FROM 



YIELDS OF Two CHECK VARIETIES. 



Oats Variety and Strain Tests. 1921. 



Number Average Probable error of a 



Check variety of plots yield Standard deviation single yield determination 



bu. bu. % bu. % 



each of Kherson and Red Rustproof oats, grown side by side as check 

 plots in the oats variety and strain test of 1921, illustrate the possibil- 

 ity of a serious error in the use of the standard deviation of check 

 plots as a measure of the variability of an experiment field. These 

 determinations are shown in Table 28. 



The field would have been considered decidedly less variable if 

 Kherson had been used as the check variety than if Red Rustproof had 



