3/]/] Journal of Agricultural Research volxx.no. 5 



From the series of correlations as a whole it appears that of the 152 

 coefficients showing the relationship between crop yields in different 

 years, 133 are positive while only 19 are negative in sign. If the differ- 

 ences in capacity for crop production demonstrated in different years 

 were due to purely transient causes, one would expect to find an approxi- 

 mately equal number of positive and negative correlations with the gen- 

 eral average value sensibly zero. Instead we find the proportion of 133 

 to 1 9. This is a deviation from the ratio 76 to 76, which one might ex- 

 pect on the assumption that there is no correlation between the yields 

 of plots in a series of years, of 



57±°- 6 745V I 5 2 X5Xo.5 = 57±4.i6. 



The deviation from equality is 13.7 times as large as its probable error 

 and is unquestionably significant. 



If we consider that coefficients which are 2.5 times or more as large as 

 their probable errors represent statistically significant interrelationships, 

 we find that of the 82 relationships which may be regarded as falling in 

 this class 78 are positive whereas only 4 are negative in sign. 



Averaging the values of the coefficients considered in Table IV, we 

 note that the average for the 133 positive values is + 0.3346, whereas that 

 for the 19 negative values is — 0.1475. Taking the constants altogether, 

 the average value is + 0.2743. 



There is, therefore, an overwhelming body of evidence to show that 

 plots, even of the small size and the apparent uniformity of those of 

 the Huntley Station, which yield higher in one year will yield higher 

 persistently throughout a series of years. 



It is now desirable to determine whether the same relationships hold 

 when these plots are divided into smaller subplots. It is possible to 

 subdivide a number of the plots into 2 subplots, each one-half the original 

 size. Correlations may be determined for the 92 yields of these half 

 plots in the same manner as for the total yields on the 46 original plots. 

 The results appear in Table V. 



The constants are positive throughout. In general, they are statis- 

 tically significant in comparison with their probable errors. As a matter 

 of fact, only 2 of the 22 constants are less than twice as large as their 

 probable errors. Thus, they indicate a real biological relationship 

 between the productions of the half plots in different years. Those 

 which give a higher yield in one year give a higher yield in another year. 



For a smaller number of the crops it is possible to divide the original 

 plots into quarter plots, thus securing 1 84 subplots to be used as a basis of 

 calculation. The coefficients of correlation between the yields in the 

 several years are shown in Table VI. 



