RELATION OF THE VARIABILITY OF YIELDS OF FRUIT 

 TREES TO THE ACCURACY OF FIELD TRIALS^ 



By L. D. Batchelor, Professor of Plant Breeding, and H. S. ReED, Professor of Plant 

 Physiology, University of California, Citrus Experiment Station.- 



INTRODUCTION 



The value of the outcome of any trial depends upon the probability 

 that a similar result will be obtained if the trial is repeated. In recent 

 years the agricultural experiment stations of all countries have greatly 

 increased the number and size of their field trials. A casual examination 

 of such trials usually shows a wide range in the reliability of the results. 

 The yields of control plots in different parts of the same tract will often 

 differ as much among themselves as the yields of fertilized and un- 

 fertilized plots differ from each other. The purpose of this paper is to 

 present the results of a study of the variation in recorded yields of fruit 

 trees taken singly and in groups of various sizes, and especially to deter- 

 mine the effect upon variability of various combinations and repetitions 

 of unit plots. 



It is recognized that the results of a single experiment are often 

 untrustworthy ; yet experimenters have published single results and have 

 based practical advice upon them. However well planned field trials 

 may be, the interpretations of the results can hardly be considered of 

 infinite reliability — that is, results which will invariably be obtained 

 when the trials are repeated. The best that can be done is to construct 

 the most probable results from the more or less varying observed results 

 of individual trials. If it is impossible to obtain perfect accuracy, it is 

 not impossible to fix the limits of error and thus to determine whether the 

 differences obtained are due to the treatments appUed or to unavoid- 

 able errors — that is, whether the differences are significant. A number 

 of trials are necessary before a reasonably reliable result can be obtained. 

 The average result of a series does not always represent the truth. 

 When averages are used, they should always be accompanied by their 

 probable errors, which are a measure of their reliability. 



Before attempting to interpret the results of a plot experiment with 

 fertiUzers, it is necessary to know whether the differences observed are 

 any greater than those which might have occurred had none of the plots 

 been fertilized. The scientific method is to recognize the inevitable error 

 and, while reducing it by every possible precaution, at the same time to 



> Paper 44, University of California, Citrus Experiment Station, Riverside. Cal. 



' The writers wish to make acknowledgment of their indebtedness for aid and criticism to Prof. H. H. 

 Love, of Cornell University; Prof. E. B. Babcock, Dr. H. B. Frost, and other members of the University 

 of California. 



Journal of Agricultural Research, Vol. XII, No. 5 



Washington, D. C. Feb. 4, 1918 



Iw Key No. Cal.— 14 



(24s) 



