ON SEISMOLOGICAL INVESTIGATIONS. 



53 



lines and tried with fair success; but it would lead to confusion to 

 multiply provisional sets of tables, and it is preferable to wait until 

 they have been thoroughly tested and corrected. But the impression 

 given by the work hitherto done is that these oscillations in the curve 

 for SP ax-e real and will explain many apparent anomaHes and diffi- 

 culties; and it is hoped that in the next Report satisfactory evidence 

 of these facts may be presented. 



X. — General Preliminary Discussion of the 1914 Results. 



It will be seen that the above discussion was conducted by the 

 study of a few particular earthquakes; not from all those given in the 

 bulletins for 1914. 



Some hesitation was felt about the form in which any discussion of 

 the 1914 residuals should take, i.e. how much provisional correction of 

 tables and epicentres should be attempted first. The tables were 

 apparently capable of improvement, and this would involve a readjust- 

 ment of some epicentres. Ultimately it was decided to try collecting 

 the results simply as they are printed, but limiting the selection to the 

 better stations: 34 observatories were included, and 15 were omitted, 

 the selection not being difficult when the mean errors had been formed. 

 The residuals for P and S were grouped for every 5° of A , except that 

 the first group extended from the epicentre to 10°. The result was 

 more definite and satisfactory than had been expected. 



It was feared that it would be difficult to draw the line between large 

 errors and definite mistakes, but when the residuals were tabulated in 

 this form there were found to be very few cases of doubt, and their effect 

 on the means was almost neghgible. The means were taken in a variety 

 of ways (one of which was to select the median or the middle residual) 

 with inclusion or exclusion of doubtful cases ; but the various alterna- 

 tives were so closely accordant that the simple arithmetic mean was 

 ultimately adopted throughout. The mean eiTors thus found were as 

 in Table XVII. 



Table XVII. 



It will be seen that both P and S show clearly the change from a 

 sensibly positive error at 110-15° to a negative error at 210-25° and 

 afterwards. This drop occurs earlier than is suggested tentatively in 

 Table V., but gives substantially the same phenomenon as was to be 



