48 BULLETIN 1440 ; TJ. S. DEPARTMENT OP AGRICULTURE 
It was decided to use only the period 1903 to 1915 in making the 
multiple correlation study, as apparently the relations were closer for 
the shorter period, possibly because of shifting of relations during the 
period. The following lengths of lag were selected for each variable 
from the results just stated, the six-months lag for the index of business 
cycles (price of industrial stocks) being used so that it would be 
possible to get a six-months " forecast." 
A Corn-hog differential, lagged 18 months. 
B Live weight of hogs, lagged 11 months. 
C Index of prices of industrial stocks, lagged six months. 
D Price of corn, lagged 11 months. 
E Price of hogs. 
Each factor except A was expressed as a percentage of straight- 
line trend and corrected for seasonal, as already explained, before 
being lagged. The multiple correlation of the four factors with E 
for the period 1903 to 1914 gave a multiple coefficient of R e .abcd = 
0.651, for the straight-line regressions. Applying the method of cur- 
vilinear multiple correlation, the regression curves shown in Figure 
30 were obtained. On the basis of these curves, a multiple correla- 
tion index of P e .abcd = 0.847 was obtained. The net correlation 
indexes of the different factors with hog prices were as follows : 
PEA-BCD = 0.581 
PED.ABC = 0.478 
pecabd=0.499 
PEB-ACD = 0.311 
The relative importance of each of the four factors, as shown by 
the determination coefficients, was as follows: 
Per cent 
A Corn-hog differential 34. 2 
B Average live weight 3. 
C Price index of business cycles 16. 6 
D Corn prices 17. 9 
Total determination by all four 71. 7 
When the hog prices estimated by the use of these results from 
values of the other four factors available at least six months in 
advance, were converted back from terms of deviations from the 
trend to actual prices, there was a correlation of 0.932 between the 
actual prices and the estimated prices. This correlation was neces- 
sarily higher as the result of putting back in the trend and seasonal 
portion of the hog prices, for which a perfect estimate could be made. 
Taking six-month moving averages of both the actual hog prices 
and those estimated there was a correlation of 0.967 between the 
smoothed values, showing that the estimated prices correlated with 
the more general swing of actual prices even more closely than with 
the prices from month to month. 
The closeness of this estimate may be gauged from the fact that 
the standard deviation of the error in " predicting" the actual monthly 
price was only 49.5 cents, and between the moving average of the 
predicted prices and the moving averages of the hog prices, only 34.8 
cents. 
The net regression curves shown in Figure 30 deserves some com- 
ment. It should be remembered that these curves represent not 
only the effect upon later prices of the particular variable designated, 
but also the effect of other related variables not taken into account, 
