Aghicultural Experiment. 461 



and the coirespoiiding average of the two outside rows was 15G lb., 

 an increase of 59 per cent, on the averag-e of the i vo inside rows, 

 the lack of competition on the outside rows having i much g-reater 

 relative effect in the dry season, in which the expei.ment was con- 

 ducted than if plenty of water had been available. 



C'()-()]'i;i;attve Exi'Kiumrnt. 



Tn co-operativ^e exj)erinients the ditHcultx of harvesting carefully- 

 areas of from one-tenth acie to one acre or more frequently arises. 

 I'Lxperiment has shown Ihat l)y harvesting caiefuUy a number of 

 nc(uirately measured areas distiibuted systematically about the plot, 

 a result may be obtained whicli gives as accurate a fig'ure as harvest- 

 ing tlie wliole area. 



Fjivk Srofnv ExrKRTMKMS. 



In feeding experiments with fattening or growing animals, the 

 individual variation is so great that the results obtained from groups 

 of less than ten animals are of very little value, and certainly need 

 to be accepted with caution. (Ireat care in the selection of animals 

 in respect of breed, sex, type, and age is also necessary. 



With feeding" experiments for milk production even greater 

 difficulty is met with. The milk yield is affected by the weather, by 

 slight indisposition, by oestrum, by pregnancy, by the time of 

 calving, by change of attendant, and varies with the stage of lacta- 

 tion. Even with the alternation system, in which the groups under 

 the different treatment are interchanged after definite intervals, it is 

 very difficult to obtain reliable results, owing to the disturbing effect 

 of the change, and the progressive variation in milk production as 

 the lactation advances. 



Error Due to Bias in the Observer. 



Knowing the variation liable to occur in the results of single- 

 plot trials, a survey of agricultural experiments so far conducted in 

 South Africa leads one inevitably to the conclusion that the observers 

 in some cases were biased in favour of the result they desired to 

 obtain. However fair-minded the observer may be, there is always 

 a danger of involuntary bias. An instructive example of the possi 

 bility of error of this type is afforded by plotting the frequency 

 curves of the results recorded in the United States Department of 

 Agriculture Year Book for 1918 of 109 observations of the depth of 

 s])ring and fall ploughing in Indiana. These curves show maxima 

 at 5, G, 7, and 8 inches, and minima at the half inches. It is 

 extremely improbable that such a curve accurately represents the 

 actual depth of ploughing; it does, however, represent the recorded 

 depth of ploughing. In this case the error is not in the results, but 

 in the observers, and is due to a tendency in the majority of them 

 to observe and record the depths in units of inches rather than in 

 halves. 



It is obvious from the above remarks that a great deal of care 

 and a fair amount of experience and knowledge are required for the 

 proper conduct of agricultural experiment, whether with field crops 

 or wdth animals, and although farmers can learn much from experi- 

 menting on their own, yet they should not regard the results 

 obtained from single-plot trials as being' final and decisive, unless 

 the difference due to the factor under test is very large indeed. 



