1918.] The Fijth Indian Science Congress. Ixxxi 
work because we have not organized and paid for it. If we 
have fairly reliable data for thirty or forty years regarding our 
question, it is obvious that there should be material enough to 
settle it, and the only point at issue is how this is to be done. 
Now quite simple considerations lead us to the idea of the 
correlation coefficient between two quantities as the extent to 
which the variations of each are determined by those of the 
other. If we have a number of bags of rupees made of various 
light materials, the variations of the weights of the bags will 
be almost entirely governed by the numbers of rupees inside, 
and the correlation “between the weight of a bag and the 
be exactly — 1. On the other hand, if the variations of one 
quantity have very little dependence on those of another, such 
as the cost of a book and its literary merit, the correlation 
The formula, being based mainly on the data from 1876 to 
908, agrees fairly well with the actual rainfall during that 
to 1917 afford us a perfectly satisfactory test of its value. The 
: defect with the actual, 
seven times out of nine. 
urning to the cold weather the forecast turns mainly on 
whether the season is severe or mild in December : for there 
are fairly long odds that it keeps through January and Feb- 
in December. The line 
cae ‘ calculated’ indicates for successive years the condi- 
ions in December in the extreme north of India, Persia, and 
r ® subsequent rainfall and snowfall of the winter as actually 
Dh aeeeie in North-West India. You will see that in most 
1 See Fig. I. 
