90 
Journal of Agricultural Research voi. xxvii, no. 2 
if the latter is not significant, no adjustment of individual yields need be 
made. The following values show the average percentage reduction in 
the probable errors resulting from adjustment when the correlation be¬ 
tween tests and checks is as stated : 
1 
r= 
1 
Reduction in 
variability * 3 
100X1— V1—f 2 
r** 
Reduction in 
variability * 3 
100X1— Vi —r 2 
1 
Per cent. 
Per cent. 
0.4 
8.4 
0.8 
40.0 
i *5 
13-4 
.866 
5 °.o 
! .6 
20.0 
•9 
5 6 -4 
I • 707 
29.3 
1.0 
100.0 
It is evident that correlations of less than 0.6 will reduce the variability 
so little that adjustment of individual yields may hardly be worth while. 
In the case of relatively low correlations, however, the moving average 
may be used as a criterion in another way. Individual yields may be 
modified by adjusting so as to change the mean as well as the standard 
deviation. Thus, if one strain occurred only on relatively poor soil in 
each of four replications, the four yields might be identical, but all rela¬ 
tively too low. This fact should be brought out by the average of the 
indices of productiveness for the four soils in which this strain was grown. 
Therefore, if individual adjustments are not made, the averages of the 
indices may be obtained and taken into account in drawing conclusions 
from the data. As an example of this use, the average percentage 
values of the 3-row checks for each of the strain numbers are given in 
Table III, column 7. These seem to be a fair measure of the reliability 
of the comparison, and if the difference between two strains were materi¬ 
ally greater than the difference between the corresponding indices more 
significance could be attached to it. 
CONCLUSION 
The fundamental concepts on which the proposed method is based are : 
(1) That the mean yield of any one seed class is a fair measure of its pro¬ 
ductiveness, (2) that the deviations of the individual replicates from this 
mean are due in part to soil variation, (3) that deviations of the yields of 
successive groups of contiguous rows from the computed production of 
an equal number of average rows of similar kind are due in part to soil 
variation, and (4) that the correlation of the percentage yields of such 
groups with the percentage yields of the individual rows upon which 
the groups are centered measures their tendency to concomitant response 
to the soil variation. Granted these, and they seem entirely sound, the 
adjustment of yields on the basis of their regression on a moving average 
follows logically. 
