STATISTICAL 85 
followed by a large number of minor shocks, and the point is 
whether these minor shocks should be treated as separate 
quakes or regarded as part of the primary shock. Again, 
ought there to be a classification according to intensity? I 
should doubt if agreement of opinion could be reached a@ 
priort. It seems to me to rest with the investigator to decide 
whether he ‘shall classify and group or not, but it then rests 
with him to show that he reaches a conclusion which is a real 
contribution to knowledge. 
There is a growing doubt whether a Fourier analysis of an 
observational quantity is really the best way of expressing 
results with a view to physical explanation of the cause, but 
however that may be, we must agree with Schuster that there 
is a right and a wrong way of making the Fourier analysis, 
and that the right way is to take the data as they stand and 
not to apply any preliminary smoothing process. It appears to 
me that if a smoothing process was permissible it would, 
carried to excess, be an argument for never making observa- 
tions at all. 
It is not sufficient to compute the Fourier co-efficients. We 
have to show that any term so obtained is substantially greater 
than what might be expected as the result of fortuitous occur- 
rence. The criterion given by Schuster is as follows :— 
“Tf a number xz of disconnected events occur within an 
interval of time T, all times being equally probable for each 
event, and if the frequency of occurrence of these events is ex- 
pressed in a series of the form 
(¢-¢ (¢-¢ 
a1 +p, COS 27 os FE NN i Ori COS) 2) 7707s — 
the probability that any of the quantities p has a value lying 
between p and p+ 6p is 
ee np?] 4 
n 
5? 6p e 
and the ‘expectancy’ for p is 
|r|.” 
On this basis Schuster finds that the lunar terms obtained 
by Knott must be discarded, but on the other hand the annual 
