ON SEISMOLOGICAL INVESTIGATIONS. 51 
The mean of the errors is + 0°73 m., and though there is a slight 
tendency to increase from the second group to the third, the material 
is fairly homogeneous. Now, comparing this with the mean for P, 
viz. + 031 m., it is clear that we are dealing with a much less 
definitely marked phenomenon, as is indeed well known. Part of 
each of these mean errors is due to errors of reading, &c.; and this 
part should be approximately the same in both. If we were to calcu- 
late and remove it, the ratio between the two, already greater than 
2 to 1, would be sensibly increased. 
In determining A from P and §, the superior accuracy of P is there- 
fore rendered more or less useless by the uncertainty of S. Galitzin’s 
azimuthal method of determining the epicentre has thus obvious 
advantages; if the epicentre is well determined from the azimuths 
at several stations, and if the time of the catastrophe is determined 
from the Ps at these stations, we should appear to have the material 
in the best shape for improving the tables of P and S, especially the 
latter. 
But this is a digression from the present investigation, which is 
primarily concerned with the performance of the Milne instruments. 
Putting aside for the present any question of correcting the tables 
for S, and therefore the position of the epicentre (as determined from 
Pulkovo), and consequent correction of the calculated times, it is clear 
that we can compare the performance of the Milne pendulums with 
other instruments on a common basis (though not the ultimate basis) 
by collecting their records for the same earthquakes in the same way. 
This is done in the following Table III., which corresponds to 
Table I. 
It will be seen— 
(2) That there are 5 large positive errors and 8 large negative 
errors, for which no special explanation can be given. In Table I. 
there are 8 negative errors, no positive. 
(b) That in 6+5+104+5=26 cases, S has presumably been read 
in place of P. With other instruments there were only 5 such cases. 
(c) That in at least 17 cases a reading has been made which can 
be attributed to a reflected wave. There are, moreover, 9 readings 
intermediate between these and the normal readings, which are extreme 
cases of one or the other. The line of demarcation is not so sharp as 
before. Similarly there are 5 doubtful negative readings. 
(d) In the middle part of the table have been collected within the 
same limits as before what may be fairly regarded as normal readings. 
They number 25 in all, They do not of themselves suggest 
any corrections to the table for P, but we might use the same correc- 
tions as before. It is simpler, however, to resfrict attention to the 
second and largest group, the mean of the errors for which is + 0°4 m. 
If, however, we include in this the ‘ doubtful’ +1°8m., +1°4m., 
+12 m., and —1:0 m., —1°2 m., the mean of the errors rises to 
“+ 0°6 m. For other instruments this mean was + 0°31 m. 
The most significant fact is perhaps that of the whole 95 
readings-only 25-at a severe scrutiny; and at most (i.e., including 
EB 2 
