IMPROVEMENT OP WHEAT BY SELECTION 87 



result shows that not the slightest trust can be placed in a 

 competitive trial of two wheats, if the trial is made only 

 once over. In the season 1913-14 the writer grew nine 

 plots of Hunter's wheat side by side in a very even 

 paddock. The plots were about half-an-acre each in area. 

 The seed and the treatment in every respect were the 

 same. The following were the yields : 



Plot 1 50 bushels per acre. Plot 6 56 bushels per acre. 



2 51 7 54 



3 60-5 8 60 



* 60 , 9 56 



5 58 



Now, differences approaching 10 bushels per acre are 

 of the greatest significance in wheat growing: indeed, it 

 will be shown later that an increased yield of 4 bushels 

 per acre is sufficient profoundly to affect the industry. 

 If, then, one not alive to the importance of the experi- 

 mental error had grown one kind of wheat on plot 2, 

 and another kind on plot 3, he would at once have 

 declared the latter infinitely the better wheat. Or if he 

 had gone to more trouble and grown a new wheat on plot 

 8, and his old standard wheat on each side of it, on plots 

 7 and 9, he (and 99 per cent, of observers) would un- 

 hesitatingly have declared the new wheat an improve- 

 ment of the highest value, whereas we know from the 

 above trial that it would not necessarily be a scrap the 

 better. Practically every agricultural experiment made 

 in this country has been a trial of plot .against plot, and 

 the results are worse than worthless. 



To be able to arrive at any reliable result, plots must 

 be replicated sufficiently frequently to allow all inequali- 

 ties of soil to be smoothed out in taking the average, and 

 the greater the accuracy of result aimed at the more 

 frequent must be the replication. Mathematical calcu- 

 lation shows that in experiments of this kind 10 to 15 

 trials are necessary before the probable error of the 

 average is reduced so low as to be negligible. 



