1042 
Journal of Agricultural Research 
Vol. V, No. 22 
yield takes into account the soil differences in both directions across the 
field. To a certain extent it is dependent upon the assumption that the 
soil changes in a uniform manner from one side of the field to the other. 
Harris (6) has pointed out that this is not always the- case, but that the 
diagrams of experimental fields indicate that differences in soil are more 
likely to occur as a spotting of the field. However, a closer study of the 
observed yields in many experimental fields indicates that theie is a 
tendency for areas of good soil (high yield) to grade off through areas of 
medium soil to regions of poor soil. Ordinarily, the changes from one 
extreme to the other are not abrupt (see fig. 3, 4). The diagrams 
published by Montgomery (10) indicate this to some extent, although 
such diagrams do not show the graded changes as well as a study of the 
actual yields of contiguous plots. 
a 
b 
c 
d 
e 
f 
g 
h 
i 
j 
k 
1 
Fig. i.—D iagram illustrating the method of obtaining the "cal¬ 
culated ’ ’ yield. (For explanation, see text.) 
Further, if the distribution of the high and low “calculated” yields in 
figures 2 and 3 are compared with the high and low observed yields, it 
will be seen that the former show approximately the same “spotting” 
as the latter. This method does tend to lessen the variability and to 
smooth the results. While it is not ideal and does not obviate all the 
difficulties, it seems possible that this method may prove useful in 
estimating soil differences. 
For cases like Montgomery's wheat experiment (10) or Mercer and 
Hall's field trials (9), where there are a number of plots all planted 
with the same variety, the contingency calculated yields may be used 
directly. For such experiments these calculated yields represent a 
smoothing of the original observations. In the case of field trials or 
variety tests, where different plots have different treatments or are 
planted with different varieties, such a smoothing tends to mask the 
actual differences between the plots. In such cases a further procedure 
is necessary. 
