370 The Experimental Error in Field Trials, [aug., 



One or two other practical points are also raised by the 

 recognition of an average error of 10 per cent, in the results 

 of a single plot. In designing field experiments, it is useless 

 to include small differences in treatment which are not likely 

 to induce more than 10 per cent, differences in the yield, 

 unless the experiment is going to be repeated very widely 

 or carried on for several years. For example, in dealing 

 with hay it would be of little use to set out conclusions from 

 comparative plots, one with i cwt. and the other ij- cwt. per 

 acre of nitrate of soda, or in an experiment on roots, one plot 

 with 4 cwt. of superphosphate per acre as against another 

 with 6 cwt. ; in each case the differences due to the manuring 

 are likely to be smaller than the experimental error. Finally, 

 in view of a io per cent, experimental error, it is no good pre- 

 tending to an impossible accuracy in recording the results. 



For example, we have read the following figures : 



Weight of roots per acre. 

 Tons. cwt. lb. 



Swedes, Champion ... ... ... n 9 28 



Crimson King ... ... 14 16 48 



Now, putting aside the barbarous units of tons, cwt., lb., so 

 cumbrous to write and difficult to read, what possible value 

 can be attached to the figures representing cwt. and lb. As 

 only the produce from tV acre was actually weighed and 

 then multiplied up to obtain the figures quoted, the inclusion 

 of a few clods of earth more or less with the roots, or an error 

 of a few inches in the measurement of the plot so as to 

 include one root extra would make a difference in the cwts. ; 

 moreover, as duplicate plots would be likely to differ in the 

 tons, very little attention need be given to the cwts. Had 

 the results been written 11*5 and 14*8 tons respectively, the 

 figures would still be a degree more accurate than the reality. 

 Of course, the experimenter must record what he actually 

 obtains to the nearest pound or ounce if his scales tell it to 

 him, but to set out such figures in the published report is to 

 make a parade of accuracy which cannot be sustained. 



Returning, however, to our original point of view, we may 

 conclude that as absolute correctness in our results is impos- 

 sible, it is only by recognising and measuring the extent of 

 the inevitable error that we can reach a due measure of 

 accuracy in the conclusions we draw from them. 



