56 REPORTS ON THE STATE OF SCIENCE. — 1916. 



We now return to the mean values, which exhibit the following 

 distinct features: — 



(a) A large positive error at about =13°. The values for P and S 

 correspond in almost exactly the ratio TSO, and thus confirm one 

 another. The observations rejected are : 



For P +144^ and +81^. There is no question as regards the 

 former. If the latter be retained the mean is increased to + 15^. As 

 this group is very important, the errors may be given in full. They 

 are: — 



The whole set is as 



It seems clear that the means cannot be far from the values assigned 

 on any reasonable supposition. And it is also clear that the excessive 

 scattering is due to the abrupt rise and fall of tha error, which is small 

 in adjoining groups. It must rise to sensibly more than the mean 

 values. The use of the erroneous tables to fix the epicentres will also 

 have tended to diminish these errors by compromise ; so that a maximum 

 error for P of + 17^ and for S or + 30^ would not be an unreasonable 

 interpretation of the figures. 



(b) The rapid fall to a negative error at about A =23° continuing 

 to A =40°. A rise again at 33° is shown by P but not by S, and for 

 the present we will disregard it. 



(c) A positive error from about 46° to 70°. This is more marked 

 in P than in S ; but it seems possible that S is already affected by the 

 negative error [d), which reduces the positive excess. 



(d) A negative error which develops rapidly in S after 80°, and may 

 have commenced earlier as remarked in (c). It was this error which 

 chiefly attracted attention in the two former Repoi'ts, in which tentative 

 corrections for it were given with some success as regards S. But the 

 corrections suggested for P wei'e apparently too large. 



This correction appears to have an important significance. The 

 ratio of times for S to times for P is nearly constant, but with the 

 adopted tables tends to rise in value for large values of A. "When, 

 however, the corrections now found are applied, which diminish the 

 values of S (when A >80°) much more than those of P, the rise in 

 value of the ratio disappears, and it seems possible that it Is definitely 

 consta:it and of value 1'800. At any rate, the departures from this 



