June 1, 1925 
Relation of Weather to Yield of Cotton 
1085 
The cotton plant can not survive in 
a freezing temperature, and it is con¬ 
ceivable that a temperature could be 
too high. Somewhere between the 
extremes there must be an optimum 
point. There is probably also an opti¬ 
mum condition of rainfall. Since there 
is considerable variation in rainfall and 
temperature from one season to an¬ 
other, there must necessarily be con¬ 
siderable deviation from this optimum. 
As weather shows the greatest year- 
to-year variation of the four factors 
mentioned and is of influence through¬ 
out the entire State, there is presum¬ 
ably a relation between changes in 
weather conditions and changes in 
average yield. The attempt was 
accordingly made to discover this 
relation between weather conditions 
and yield, and to ascertain to what 
degree the latter may be explained by 
the former. Residual variations in 
yield (changes left over after the effects 
of weather have been eliminated) were 
then considered attributable to one or 
more of three things: (1) Errors in 
the original estimates of yield and 
weather data used; (2) an insufficient 
statistical expression of weather condi¬ 
tions; and (3) influence of the factors 
mentioned other than weather. As a 
matter of fact, these residuals were 
very slight, thus furnishing statistical 
support for the hypothesis advanced. 
TABULATION OF DATA 
For use in the quantitative analysis, 
data on yield per acre and rainfall and 
temperature were selected. Official 
figures of yield per acre of the United 
States Department of Agriculture were 
selected after comparison with the 
sources from which the final figures are 
taken. This is an expression of yield 
in pounds of lint cotton produced per 
acre. 
Selection of weather data to be used 
was governed to a considerable extent 
by what is available without reworking 
the original Weather Bureau records. 
Undoubtedly, percentage of sunshine, 
humidity, winds, • rainfall, and tem¬ 
perature taken for short periods of 
time, particularly in critical growing 
periods, are of importance. As only 
monthly data, however, are readily 
available over the period of time 
selected, 1900 to 1922, inclusive, the 
shortest weather period that could be 
used was automatically defined. Fur¬ 
thermore, temperature and rainfall are 
the only satisfactory State average 
weather figures covering the period. 
Accordingly the following series (Table 
I) for the State selected (Louisiana) 
were chosen for comparison and cor¬ 
relation with yield. 
A number of other rainfall and tem¬ 
perature series were experimented 
with, but they showed slight relation 
to yield and are not included here. 
Table I records the original data 
described in the preceding. 
Table I .—Average rainfall and tem¬ 
perature for certain months and yield 
of lint cotton per acre, Louisiana, 
for the years 1900 to 1922 inclusive. 
The method of handling the described 
data is an extension of multiple correla¬ 
tion methods 4 designed by M. Ezekiel, 
of the Bureau of Agricultural Econom¬ 
ics, and is briefly as follows: 
By the usual methods of multiple 
correlation, 5 the net regression lines 
showing net effect of each weather 
factor upon the yield are plotted. The 
values of the dependent as obtained 
from the regression equation are deter¬ 
mined, as are the residuals—plus or 
minus differences between the actual 
yield and yield estimated from regres¬ 
sion equation. These residuals are in 
turn plotted against each weather fac¬ 
tor as deviations from each of the re¬ 
gression lines, the lines then being 
curved to pass through the plotted 
points in so far as consistent with the 
hypothesis of a “smooth curve” func. 
4 Ezekiel, M. J. B. a method or handling curvilinear cokkelation for ant number of vari¬ 
ables. Jour. Amer. Statis. Assoc. 19:431-453, illus., 1924. 
* Tolley, H. R., and Ezekiel, M. J. B. a method of handling multiple correlation problems. 
Jour. Amer. Statis. Assoc. 18:993-1003. 1923. 
