270 



Journal of Agricultural Research 



Vol. XII, No. s 



±67.35 per cent; (b) ±44.44 per cent; (c) ±28.42 per cent. Table IX 

 summarizes the results. It seems probable, therefore, that a difference 

 between two 16 adjacent tree plots of less than 50 per cent of the mean 

 production should be considered with caution before attributing it to 

 differential treatment. The scattering of the units of the plots increased 

 the accuracy very decidedly, four units giving more accurate comparison 

 than two. Even with this scattering, differences of less than 30 or 40 

 per cent are well within the realm of chance. The apparent cause for the 

 4-unit plot having a probable error less than the theoretical is accounted 

 for by the fact that the variation in productivity of the soil was known 

 when the distribution of the units was made. The results might not have 

 approached so closely to the theoretical if the distribution had been 

 decided upon before harvest. 



Table IX. — Comparison of the reliability of a plot of 16 adjacent trees with that of two 

 units of 8 trees each and of four units of 4 trees each. Jonathan apples 



The large probable errors which are apparently always present with 

 plot trials of fruit trees emphasize the importance and value of obtaining 

 individual tree records of experimental orchards before differential 

 treatment is started. The probable errors will likely differ somewhat 

 from year to year, and possibly be further influenced by the advanced 

 age of the trees. Nevertheless, if but one or two years ' records of mature 

 trees are available before differential treatment is commenced, at least 

 some idea can be obtained of the casual variation of the plots — that is, 

 limitations can be placed beyond which observed differences in plots 

 may be due to chance rather than to the factors under experimentation. 

 In the absence of such previous records, the employment of frequent 

 controls or standard treatments may be indicative of the probable error 

 of the entire area. 



REIyATlON OF THE SHAPE OF THE PLOT TO THE VARIABIUTY OF THE 



COMPARATIVE YIELDS 



The shape of small plots may be of great importance when cultural 

 operations are considered. 



Lyon {igi2) found no satisfactory evidence that long and narrow 

 plots are less likely to error than square plots when no control plots are 



