FIELD PLOT TECHNIC 



57 



error and, therefore, under the assumed conditions of the experi- 

 ment, of little significance, as is indicated by Table X taken from 

 Pearl and Miner (1914). 



Use of Probable Error in Eliminating Strains. The probable 

 error obtained by means of the checks may also aid in selecting 

 an elimination value below which varieties or strains may be dis- 

 carded without danger of throwing away a valuable one. This 

 figure is necessarily more or less arbitrary and will depend upon 

 the desired degree of accuracy. The magnitude of the figure 

 which is multiplied by the probable error will also depend some- 

 what upon the desired amount of elimination. The method 

 used at the Minnesota Station is to subtract the product of three 

 times the probable error for the method of test multiplied by 

 \/2 from the highest or one of the higher yielding strains. The 

 difference gives a figure below which it is considered safe to dis- 

 card without danger of eliminating a high yielding strain. If 

 the yield of a strain falls below the elimination figure for two or 

 three years, it is discarded from further trials. 



The Pairing Method of Securing a Probable Error. Under 

 certain conditions it is impracticable to devote so large a share 

 of the experimental field to check plots. Wood and Stratton 

 (1911) have suggested a means of securing a reliable probable 

 error without the aid of checks. Briefly, their method consists 

 of systematically pairing similarly treated plots and finding their 

 mean yields. The deviation of this mean from the yield of the 

 original plots is expressed in percentage of the mean. The fol- 

 lowing illustrates the procedure: 



Now if all A plots are similarly treated, A' would be paired 

 with A" and A" with A'", etc. 



In this method the probable error is expressed in percentage of 

 the mean. If the number of pairs is sufficiently great the devia- 

 tions + and will yield a normal frequency curve. As in the 

 method of determining the probable error by means of checks, 

 it is desirable to have a large enough number of variants to secure 

 at least an approach to the normal frequency distribution. 



