18 BULLETIN 1224, U. S. DEPARTMENT OF AGRICULTURE. 
county is near 100. This is in group 10, where the correlation between 
rent and productivity is the same as the correlation between value 
and productivity. In the rest of the areas the number per county 
is much less. In most of the counties the number of one-year tenant 
farms ranges from 25 to 75, and in not a few the number is less than 
25. The total number of farms per county ranges from about 2,000 
to 3,000. Thus the number of one-year tenant farms constitutes a 
very small percentage of the total number of farms per county. It 
is not probable, then, that so small a sample will generally be repre- 
sentative of the entire county. 
The difference between the average values on the one-year tenant 
farms and the county average values do, then, represent, in part, 
real differences in the grade of the rented farms and the county 
average grade. But they do not reflect this difference in grade 
accurately, for if they did the correlation between adjusted rents 
and values would be considerably higher for each group than the 
correlation between unadjusted rent and value. In other words, the 
error is not in the average rents, for they are almost as highly correlated 
with county average value as the rents are when adjusted to the 
county basis, and in group 11 the unadjusted rents are more highly 
correlated with value. If the average values on these one-year 
tenant farms were more accurate, then the adjusted rents would 
show a much higher correlation with value than the unadjusted 
rents, because the unadjusted rents are not representative of the 
entire count}-. Then, too, the adjusted rents would be as highly, or 
more highly, correlated with the productivity factors than are county 
average values. 
Evidence from these correlations between the adjusted and the 
unadjusted rents and county average values is in line with what one 
would expect from the census valuations of land. The estimated 
value obtained by the census enumerator for each farm is of necessity 
roughly and hurriedly made. As a result the value placed upon any 
individual farm is wholly unreliable, and the average for small groups 
of farms is likely to be very erratic. But for large numbers of farms 
the census average value is highly accurate, as shown by the corre- 
lation between county average values and productivity factors. 
It may be concluded, therefore, that the cash-rent data are reliable 
and that they give highly reliable averages even when based upon a 
very small number of farms, but that the census values can not be 
relied upon unless based upon a large number of farms, and then 
they, too, are highly dependable. It is clear that the county ratios 
of rent to value, in general, can not be relied upon, largely because 
of the inaccuracy of the census values when based upon the small 
number of one-year tenant farms found in most of the counties, but 
also in part because there are not enough rented farms to give average 
rents that are entirely reliable. But since the areas included in the 
county groups are homogeneous areas, the group ratios can be relied 
upon to give an account of the relation of land income to land value 
in 1920 that is highly accurate. The greatest error for the county 
groups is apparently 0.4 per cent (see p. 12), and this is due to the 
upward bias of the census valuations in this area. Although data 
were not available to measure the bias in the census values for all 
areas, it was probably less in other regions than in those where 
