36o EXPERIMENTS WITH FERTILISERS [chap. 



or other it must exist in all work involving measure- 

 ments, and the only scientific method of dealing with it 

 is to estimate its magnitude and to draw no conclu- 

 sions from results which are not well outside that 

 magnitude. For example, the experimental error of a 

 field plot on average soil and under ordinary farming 

 conditions in this country may be taken as about lo 

 per cent. ; this means that if the yield of the standard 

 Plot A be taken as lOO, and another Plot B yields 109, 

 while a third C yields 91, the conclusion cannot be 

 drawn that B is better than A, and A better than C, 

 because the same variations in the results might have 

 been seen had the three plots been treated exactly 

 alike. Furthermore, unless the real difference brought 

 about by two different methods of treatment is greater 

 than 10 per cent, it is hopeless to expect to reveal 

 the difference at all by a single pair of experimental 

 plots. 



These points may be illustrated by actual examples 

 drawn from the Rothamsted experiments, where the 

 soil conditions are fairly uniform, though by no means 

 exceptionally so, and the control and management is 

 about as good as they can be under ordinary farming 

 conditions. 



On the grass fields are two unmanured plots almost 

 at the two extremities of the field, and taking a fifty 

 years' average, one of these plots (No. 12) is 10 per 

 cent, better than the other (No. 3), owing to some 

 fundamental superiority of soil or situation. Table 

 cm. (pp. 361-2) shows the actual results given by these 

 plots year by year, reduced to a common standard by 

 taking the yield of Plot 3 in each year as 100. 



It will be seen that though on the average of the 

 whole period Plot 12 is represented by no when Plot 

 3 is 100, yet there are twelve occasions when Plot 12 



